<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by HGR / Honda R&amp;D Center of Excellence on Medium]]></title>
        <description><![CDATA[Stories by HGR / Honda R&amp;D Center of Excellence on Medium]]></description>
        <link>https://medium.com/@HGRX?source=rss-fd02d7b9dda5------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*dFGaEFVIvsb3l4WVlamgJg.png</url>
            <title>Stories by HGR / Honda R&amp;amp;D Center of Excellence on Medium</title>
            <link>https://medium.com/@HGRX?source=rss-fd02d7b9dda5------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 09:21:24 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@HGRX/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[The Ultimate Material. The Real-World Challenge. ]]></title>
            <link>https://medium.com/@HGRX/the-ultimate-material-the-real-world-challenge-ba92797df61f?source=rss-fd02d7b9dda5------2</link>
            <guid isPermaLink="false">https://medium.com/p/ba92797df61f</guid>
            <category><![CDATA[honda]]></category>
            <category><![CDATA[engineering]]></category>
            <category><![CDATA[hgr]]></category>
            <category><![CDATA[semiconductors]]></category>
            <category><![CDATA[careers]]></category>
            <dc:creator><![CDATA[HGR / Honda R&D Center of Excellence]]></dc:creator>
            <pubDate>Tue, 10 Nov 2026 23:01:02 GMT</pubDate>
            <atom:updated>2026-11-10T23:01:02.273Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>The Ultimate Material. The Real-World Challenge. <em>— Honda’s Pursuit of Diamond Power Semiconductors</em></strong></h3><p>By Shuji Murai | HGR, Honda R&amp;D</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8VFshNN_YIprKW8o239Acg.jpeg" /></figure><p>Before joining Honda in October 2024, I’d spent my career in the semiconductor industry. To me, Honda was a company that made cars and motorcycles. To be honest, I didn’t even know Honda was conducting semiconductor research.</p><p>Now, I work in materials research at HGR, focusing on diamond power semiconductors. Diamond has such remarkable properties that it is often called the “ultimate semiconductor material.” <strong>But harnessing those properties in real products is anything but straightforward.</strong></p><p>Why would Honda, which isn’t a semiconductor manufacturer, take on such a difficult challenge? And what does it take to turn the ultimate material into technology people can actually use?</p><p>Together with Kentaro Ema, who also joined Honda from the semiconductor industry, I’d like to share where our research stands today — and where we hope to take it.</p><h3><strong>Why Honda Is Taking On the Ultimate Challenge</strong></h3><p><strong>Ema:</strong> I joined Honda in April 2025, also after working in the semiconductor industry. I’m now in the materials research domain, working on techniques for growing diamond crystals and on fabricating and evaluating semiconductor devices made from them.</p><p><strong>Murai:</strong> I mainly develop device fabrication processes and research modules built around those devices. Before joining, I had no sense of how far Honda’s semiconductor research had progressed. I was surprised to find that the company was already working with specialized materials like diamond.</p><p><strong>Ema:</strong> “Power semiconductor” may not be a familiar term. When most people think of semiconductors, they picture chips like CPUs that perform calculations. Power semiconductors do something different: they convert electrical power and control how it flows.</p><p>You’ll find them in automotive inverters and DC-DC converters, but also in trains and smartphone chargers. Ever noticed how warm a charger can get? That’s because some energy is lost as heat when electricity is converted.</p><p>The aim is to keep those losses as low as possible. Silicon and silicon carbide (SiC) are commonly used today. We’re looking beyond them, to diamond.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*s5RWHeyv484SeaAgmpOj3w.jpeg" /><figcaption><em>From left: Kentaro Ema and Shuji Murai</em></figcaption></figure><p><strong>Murai:</strong> Diamond can potentially withstand high voltages and operate at high speeds. It also has exceptionally high thermal conductivity, so it transfers heat very effectively. And it offers the prospect of stable operation at high temperatures and in environments with intense radiation.</p><p>That’s why it’s called the “ultimate semiconductor material.”</p><p>For semiconductor applications, we use synthetic rather than natural diamond. But outstanding properties on paper don’t automatically translate into performance in a device. Unlocking that potential is extremely difficult.</p><p><strong>Ema:</strong> First, diamond is incredibly hard. Making a semiconductor involves processes such as smoothing surfaces and creating grooves. The techniques for doing that reliably with diamond are still being developed.</p><p><strong>Murai:</strong> We sometimes use blades coated with diamond particles to cut conventional semiconductor materials. Now we’re trying to cut diamond itself. <em>[Laughs]</em></p><p><strong>Ema:</strong> Not exactly a material most people would choose to work with.</p><p><strong>Murai:</strong> I think Honda could make that choice precisely because we’re not a semiconductor manufacturer. If we simply tried to catch up with specialist manufacturers working on silicon or SiC, it would be hard to offer something distinctly Honda.</p><p>So why not take on a more advanced technology?</p><p>If we were going to research semiconductors ourselves, why not aim for the material considered the ultimate? That was our thinking from the start.</p><h3><strong>Don’t Wait for Perfection. Make It Work. — A World-First Demonstration</strong></h3><p><strong>Ema:</strong> One major obstacle to using diamond in power semiconductors is defects in the crystal.</p><p>This may not be the most flattering description, but today’s crystals are still fairly “dirty.” If you build a device where a defect is present, it may not behave as intended.</p><p><strong>Murai:</strong> There are different kinds of defects. Some affect performance; others don’t. You can’t tell which ones matter just by looking.</p><p>To make reliable products, we need a way to identify the defects that affect device performance before we fabricate the devices.</p><p>SiC and other semiconductor materials faced similar challenges before they became practical. Growing crystals with fewer defects is important, of course. <strong>But we can’t stop developing devices while we wait for perfect crystals.</strong></p><p><strong>Ema:</strong> Our idea was to make lots of small devices, identify the ones that performed well, and connect those together.</p><p>With one large device, there’s a greater chance that a defect somewhere will compromise its performance. Smaller devices let us separate those that work properly from those that don’t.</p><p>Research at universities and AIST has often used crystals just 3–4 millimeters square. We used a half-inch substrate, which meant finding a way to arrange and connect many more devices.</p><p>That created a challenge of its own: how to connect all those devices effectively.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7NJXEtQeZy6pQAniy1ZUwQ.jpeg" /></figure><p><strong>Murai:</strong> That’s the idea behind our integration approach. But selecting good devices and wiring them together isn’t enough to handle currents in the ampere range.</p><p>We had to work out how to arrange and connect them — and design the structure itself — to carry larger currents.</p><p><strong>Ema:</strong> Selecting individual devices isn’t particularly unusual in the semiconductor industry.</p><p>A transistor has to let current through when it’s supposed to and shut it off when it isn’t. Test individual devices, though, and you’ll find that some keep conducting when they shouldn’t, while others barely conduct at all.</p><p>So you have to measure them and select the ones that work.</p><p><strong>Murai:</strong> We automate those measurements now, but early on we did them by hand — around 400 devices, one at a time.</p><p>We’re still in the research phase. Unlike a semiconductor production line, we can’t yet make identical devices consistently.</p><p>We still handle materials with tweezers and place them in beakers. Even a tiny difference in processing can change a device’s characteristics.</p><p><strong>Ema:</strong> But we wanted to show that the idea could work in practice.</p><p><strong>Murai:</strong> At the time, getting a diamond transistor to carry a substantial current was a challenge in itself.</p><p>Someone once told us, “If it can’t handle ampere-level current, you can’t call it a power semiconductor.”</p><p>We’d initially aimed for a few hundred milliamperes. After hearing that, we decided to go for ampere-level current.</p><p>After a great deal of trial and error, we demonstrated high-speed switching at 2.5 amperes (A) in 2025.</p><p>We fabricated many MOSFETs (*1) on a half-inch diamond substrate, selected the best-performing devices, and connected them in parallel.</p><p><strong>Working with AIST, we achieved the world’s first demonstration of ampere-level high-speed switching using diamond MOSFETs.</strong></p><p><em>*1 MOSFET stands for Metal-Oxide-Semiconductor Field-Effect Transistor.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SIzsAEARhg_whFk1p0fl_Q.jpeg" /></figure><p><strong>Ema:</strong> Instead of improving just one small device, we combined several to switch ampere-level current on and off at high speed.</p><p>That gave us a tangible demonstration of diamond’s potential as a power semiconductor.</p><p><strong>Murai:</strong> Even when the prototype was ready, we didn’t know for sure that it would work. Some people on the team were asking, “Will it really work?”</p><p>We didn’t want to destroy it by applying too much current at once, so we started low and gradually increased it.</p><p>When we finally reached ampere-level current, the whole team cheered.</p><p>There had been some skepticism inside Honda, too: could diamond semiconductors really be useful?</p><p>Being able to show a working device made a real difference.</p><p><strong>Ema:</strong> Of course, we’re nowhere near finished. Traction inverters in cars handle currents of several hundred amperes — roughly 100 times our 2.5 A result.</p><p><strong>Murai:</strong> After that demonstration, we tried a device with even more elements integrated into it, but it didn’t perform as we’d hoped.</p><p>Adding more devices doesn’t necessarily produce a proportional increase in performance.</p><p><strong>Ema:</strong> One breakthrough brought the next challenges into focus. In terms of practical use, we’re only at the starting line.</p><h3><strong>Beyond Semiconductors: The Whole System Matters</strong></h3><p><strong>Murai:</strong> What makes research at Honda so interesting is that it doesn’t end with a material or an individual device.</p><p>We think ahead to fabricating the device, building it into a module, and making it work as part of an inverter.</p><p>Only when we build and test something do certain problems become apparent. We can then feed what we learn back into our semiconductor research.</p><p>Where does a problem in the material show up in the device or module? What’s missing when we try to use it in an inverter?</p><p>Being able to move between those levels as we set our research priorities is a real strength of Honda as a mobility manufacturer.</p><p><strong>Ema:</strong> It makes a huge difference to be able to see the whole process.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sOaRWxErbBtnD7efG7X3lw.jpeg" /></figure><p><strong>Murai:</strong> Coming to Honda changed how I think about making things.</p><p>Honda has built up a wealth of technology and experience through developing cars. Before I joined, I never imagined how much of that knowledge could inform semiconductor research.</p><p>People here see problems from angles I hadn’t encountered in my previous jobs.</p><p>That fresh perspective comes, in part, from Honda not being a semiconductor manufacturer.</p><p><strong>Ema:</strong> At the same time, those of us who specialize in semiconductors don’t necessarily know much about cars.</p><p>When colleagues with experience developing inverters or engines join the discussion, they can point to places where a particular semiconductor might make a difference.</p><p><strong>The goal isn’t just better semiconductor performance.</strong> We can think concretely about the value that performance could create.</p><p><strong>Murai:</strong> Our team includes materials and electrical engineering specialists, but also people who used to work on engines and connecting rods.</p><p>At first, I wondered whether people from such different backgrounds would find semiconductor research difficult.</p><p>But they’re remarkably keen to learn. They draw on what they already know while quickly picking up new skills.</p><p>I’ve also been struck by how openly people speak up, regardless of their age or experience.</p><p>Younger team members contribute ideas in meetings, and everyone discusses them seriously. When a question comes up, we investigate it and use what we learn in the next stage of research.</p><p>What matters is the idea, not who came up with it. That’s something I’ve really appreciated since joining Honda.</p><p><strong>Ema:</strong> A vital partner in this work is AIST (*2).</p><p>AIST has more than 30 years of experience in diamond research. We have a tremendous amount to learn from its researchers.</p><p><em>*2 AIST: National Institute of Advanced Industrial Science and Technology. One of Japan’s largest public research organizations, with 12 locations nationwide, including Tsukuba, Ibaraki Prefecture.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HNPclR3j2NqRF9kYFpMM_w.jpeg" /></figure><p>In February 2026, the <a href="https://www.linkedin.com/posts/honda_honda-rd-establishes-joint-research-laboratory-activity-7425448170287153152-sGEG">Honda R&amp;D–AIST Diamond × Electronics Joint Research Laboratory</a> was established at AIST’s Tsukuba Center, a major research hub for next-generation power semiconductors.</p><p><strong>Murai:</strong> Following the steps in a research paper doesn’t guarantee you’ll get the same results, even if you operate the equipment exactly as described.</p><p>There’s a great deal of know-how that comes only with experience.</p><p>It’s reassuring to be able to stand in front of the equipment with AIST researchers, work through problems together, and learn from their advice.</p><p><strong>Ema:</strong> Honda contributes a clear view of how we want to use the technology in mobility — and what performance those applications will demand.</p><p>By bringing AIST’s diamond research expertise together with Honda’s mobility expertise, we can move closer to technology that works in the real world.</p><p>That’s what makes the partnership valuable.</p><p><strong>Murai:</strong> Honda and AIST have been conducting joint research since 2023. In February 2026, we established the <a href="https://www.linkedin.com/posts/honda_honda-rd-establishes-joint-research-laboratory-activity-7425448170287153152-sGEG">Honda R&amp;D–AIST Diamond × Electronics Joint Research Laboratory</a>.</p><p>The aim is to advance research toward practical applications by combining AIST’s technologies with Honda’s mobility expertise.</p><p><strong>Ema:</strong> And Honda and AIST can’t do it all alone.</p><p>Bringing a semiconductor into a real product takes equipment, expertise, and people.</p><p>We want to keep moving forward alongside other engineers and partners who share our goal.</p><h3><strong>Making the Ultimate Part of Everyday Life</strong></h3><p><strong>Murai:</strong> What excites me most about the prospect of diamond power semiconductors becoming practical is what they could mean for heat management.</p><p>Today’s inverters need heat sinks and other components to dissipate heat. Those parts take up space and add weight.</p><p>If we can make the most of diamond’s properties, we may be able to shrink the cooling system and surrounding components — and make the entire inverter more compact.</p><p>That could change how cars are designed.</p><p>An inverter might fit somewhere it simply couldn’t fit before.</p><p>We could even imagine placing one near each wheel and controlling them individually.</p><p>So this isn’t just about making a better semiconductor.</p><p><strong>Ema:</strong> It’s exciting to think about the possibilities that could open up for mobility.</p><p>But we still have a lot to overcome before this becomes technology we can put into a real product.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*laD13d_h5ClOErvtmoTpaw.jpeg" /></figure><p><strong>Murai:</strong> That’s why we don’t just ask what we’ll be able to do once the research advances.</p><p><strong>We start with what we want to make possible in the future.</strong> Then we work backward to identify the technologies and people we’ll need.</p><p>We don’t want the ultimate material to remain a laboratory achievement. We want to carry the work through to real-world use.</p><p>At a semiconductor manufacturer, meeting specified standards and passing tests was one of the goals.</p><p>Honda, though, delivers mobility products to customers — and semiconductors are part of those products.</p><p>From the research stage onward, we need to ask what value we can ultimately bring to customers and define our challenges accordingly.</p><p><strong>Ema:</strong> Customers may never see the semiconductors inside a finished car.</p><p>But knowing that something we researched made it into a product and created new value for them would be incredibly rewarding.</p><p>One day, a material known as the “ultimate semiconductor material” might no longer seem extraordinary. It could simply become part of everyday technology.</p><p>Helping make that future possible is what makes this research at Honda so worthwhile.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ba92797df61f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Keep Blue in Motion.— Honda and Toray Take on Resource Circulation with Subcritical Water]]></title>
            <link>https://medium.com/@HGRX/keep-blue-in-motion-honda-and-toray-take-on-resource-circulation-with-subcritical-water-fd09f78c7d9a?source=rss-fd02d7b9dda5------2</link>
            <guid isPermaLink="false">https://medium.com/p/fd09f78c7d9a</guid>
            <dc:creator><![CDATA[HGR / Honda R&D Center of Excellence]]></dc:creator>
            <pubDate>Wed, 04 Nov 2026 02:46:01 GMT</pubDate>
            <atom:updated>2026-11-04T02:46:01.821Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nePalIp0SchXIkdqyi__fg.jpeg" /></figure><p>Hello everyone. I’m Satoshi Hirawaki from HGR.</p><p>At Honda, we are working with Toray Industries, Inc. on a chemical recycling technology that uses subcritical water* to break down a material called nylon 6 and turn it back into its raw material. Our goal is “Car to Car” resource circulation: recovering the raw material from nylon 6 in end-of-life automotive parts and using it once again to make materials for cars.</p><p>It all began in 2020, when we first took the idea to Toray.</p><p>For this conversation, I sat down with Hideki Matsumoto, Director of Toray’s Chemicals Research Laboratories, with whom I have been pursuing this joint research. We were also joined by Kozo Takahashi, Toray’s Corporate Vice President in charge of the Research &amp; Development Division, and Atsushi Ogawa, Chief Operating Officer of Honda R&amp;D Research Center of Excellence (HGR), to look back on our journey from our first meeting to where we are today.</p><p>What does it take to keep a blue-sky idea from remaining just an idea? We talked about the difficulties of keeping resources in circulation and what lies beyond them — not only in terms of technology, but also quality, cost, energy, and the systems that support society.</p><p>*<em>Subcritical water: Water heated to a high temperature under high pressure so that it remains liquid. Its properties differ from those of ordinary water, allowing it to be used to break down substances, among other applications.</em></p><h3><strong>It Started with a “Blue-Sky” Idea.</strong></h3><p><strong>Hirawaki:</strong> Our first contact with Toray was in 2020. We visited your head office and explained what we wanted to do. The document we brought was really only about three pages long. We were still at the stage of saying, “We’re about to get started.”</p><p>I was convinced that plastic recycling would become a major issue for society. But Honda is not a chemical manufacturer. If we were serious about this, we needed to work with a company that understood materials and chemistry. That was when we began looking for a partner.</p><p><strong>Ogawa:</strong> Looking back, some of what we were saying was pretty idealistic, wasn’t it?</p><p>We wondered whether recycling could be more local and decentralized. Instead of gathering huge volumes of used materials in one place and processing them again at a large facility, perhaps circulating them on a smaller scale could change the logistics costs, too.</p><p>We were saying ambitious things precisely because we knew so little.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*P3IU0dbh2emedM8gVmgJOw.jpeg" /><figcaption>From left: Kozo Takahashi, Hideki Matsumoto, Satoshi Hirawaki, and Atsushi Ogawa</figcaption></figure><p><strong>Matsumoto:</strong> Honestly, when I first heard the idea, I thought, “That’s pretty idealistic.” [Laughs] But Hirawaki had such a strong belief in this technology.</p><p>There are many ways to recycle, but in the end, it has to make economic sense. You can do it by using various chemicals and adding more processing steps. But every step costs money. Against that background, the idea of using water to recover the raw material was intriguing even from a chemist’s perspective.</p><p>I thought, “This might just turn into something big.”</p><p><strong>Ogawa:</strong> We were serious. Serious, but idealistic. [Laughs] Still, if you only talk about what is realistic from the outset, research like this never gets started. As you move forward, of course, the difficulties become clear. What about quality? How do you guarantee performance? How much energy will it use?</p><p>It is easy to say, “Recover it from cars and use it in cars again,” but actually doing that is far from simple.</p><p><strong>Hirawaki:</strong> I knew that the difficulties would emerge once we got a certain distance into the work. But I don’t think research for the future is something you begin only after you know exactly how to make it succeed.</p><p>First, you take a small step forward. Then, as you go, you gradually build greater certainty. Otherwise, you won’t get the equipment or the people you need.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vaFUae8V1XGMavZzV681tg.jpeg" /></figure><p><strong>Takahashi:</strong> I think that is a necessary instinct in research. That doesn’t mean you should keep every project going, whatever happens. But if you judge everything by today’s standards and stop simply because it won’t become a business right away, nothing that takes a long time will ever develop.</p><p>At Toray, we have nurtured businesses such as carbon fiber and water treatment with a long-term perspective. Even as we reassess along the way, we keep the things that need to be preserved for the future. I think that perspective matters.</p><p><strong>Matsumoto:</strong> And Honda doesn’t just think about things. You actually do them.</p><p><strong>Takahashi:</strong> Exactly. You take action. Working together, I’ve found that Honda has quite a playful spirit. Toray is… rather earnest. [Laughs] I think we could learn from that willingness to say, “Let’s try it,” and get moving.</p><h3><strong>Different Strengths, One Team.</strong></h3><p><strong>Matsumoto:</strong> Honda wasn’t planning to work exclusively with Toray from the beginning, was it? We heard you were talking to other companies, too, so we thought, “We’re not the only ones.” [Laughs]</p><p>But at a certain point, you told us, “We’re going to do this with Toray.” That meant a lot.</p><p><strong>Hirawaki:</strong> Once we began joint development, we became very open about sharing information. And when we started building the equipment together, I was struck again by Toray’s strengths.</p><p>Drawing up plans, creating process flows, and turning chemical reactions into working equipment. I thought, “This is what real expertise looks like.”</p><p>Honda is an automotive company, so we are good at bringing different parts together into a single product through assembly, welding, and other processes. But our work basically starts when something arrives already made into a material.</p><p><strong>Takahashi:</strong> That is quite different from Toray. We work with monomers and produce polymers. We handle chemicals and think about how to turn them into a viable process. Chemical engineering and processing, in particular, are fields we have worked in for many years.</p><p><strong>Ogawa:</strong> That really is an unfamiliar world to us. We can imagine carrying out a reaction in the lab, but what do you do when it becomes an actual facility? What kind of reactor do you need? What happens if impurities get in? Polymerization is not our area of expertise, either.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Db5ki4CT6ysKGBrs5j6nRA.jpeg" /></figure><p><strong>Matsumoto:</strong> And from our side, it’s more like, “How does a car even move?” [Laughs] Those great masses of steel drive around, and you even make rockets fly. I still don’t really understand it. [Laughs]</p><p><strong>Hirawaki:</strong> That’s exactly it. We are experts at combining all kinds of things into a finished product. Toray’s expertise lies further upstream, in making the materials themselves. Even when we use the same equipment, we are looking at different things.</p><p><strong>Takahashi:</strong> That is why working together has value. There was a time when Toray also had a strong preference for doing everything in-house — a belief that research was valuable only if we originated it ourselves.</p><p>But trying to do everything on our own isn’t the right approach. We need to stay clear about what we want to achieve. At the same time, if we lack something, we should join forces with people who have it. That attitude is becoming particularly important in research.</p><p><strong>Matsumoto:</strong> We were a little more reserved with each other at the start, though. [Laughs] As the research progressed, we kept encountering things that made us say, “If we’re really going to make this happen, we have to get past this barrier.” Being able to speak honestly about those things was a big step.</p><p><strong>Ogawa:</strong> You won’t get to practical application by just saying, “We did it,” and “That’s great.”</p><p><strong>Matsumoto:</strong> Exactly. It is precisely because we really want to make it happen that we can say, “This part still isn’t good enough.” I feel that was when we finally became one team.</p><h3><strong>Turning Nylon Back into Raw Material.</strong></h3><p><strong>Hirawaki:</strong> The resin we are working with is nylon 6. It is a familiar material used in automotive parts as well as clothing and other products.</p><p>Put very simply, we are trying to break it down with water and turn it back into its raw material.</p><p>Of course, this is not ordinary water. We heat and pressurize it to bring it into a subcritical state. We then use that water to break down nylon 6 into caprolactam, its raw material. Once we get it back to that stage, we can purify it and make nylon 6 again.</p><p><strong>Matsumoto:</strong> Nylon 6 recycling itself has been around for a long time. Relatively pure products, such as fishing nets, are fairly easy to recycle using conventional methods.</p><p>But automotive parts are not so simple. They contain not just nylon 6, but also glass fiber and various other substances. To truly achieve Car to Car resource circulation, we have to separate those components while recovering the raw material from nylon 6. We are exploring whether subcritical water could be one answer.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*walV4tPc8te44p1oSKqEUw.jpeg" /></figure><p><strong>Hirawaki:</strong> Research papers had already shown that subcritical water could break down nylon 6. But there is a huge difference between “It can be done” and “It can be used in the real world.”</p><p>What we were initially looking at was still inside a flask in the lab. Could we scale up the equipment and actually process a certain volume? Would we still recover the intended product at that larger scale? Unless we go that far, it will never be practical.</p><p><strong>Matsumoto:</strong> In a sense, demonstrating that something works in the lab is just the starting point. What happens when you scale it up? Is the quality consistent? How much energy does it need? What does it cost? Those are the questions we are now examining at the pilot facility.</p><p><strong>Ogawa:</strong> Being able to recover the raw material is a big advantage. If you have to say, “We don’t know what’s in it,” it is difficult to reuse it in a car.</p><p><strong>Matsumoto:</strong> That’s right. With mechanical recycling — reusing waste as feedstock to make materials again — you retain the material as it was used, so inevitably you face the question, “What does it contain?”</p><p>But if chemical recycling allows us to recover the raw material and purify it properly, it becomes easier to use as a material again. Resources that have already been used can move on to their next use instead of reaching the end of the line. There is real significance in that.</p><h3><strong>From Environmental Value to Value Customers Choose.</strong></h3><p><strong>Ogawa:</strong> In this field, people often ask, “Can recycling create new value?” Of course, that is the ideal.</p><p>But I think there is still work to do before we can talk about that. If recycled materials cannot deliver the same performance as current products — or if they can, but the cost is nowhere near viable — they won’t make it into real-world use.</p><p>At worst, we could end up simply saying, “We recycled it because regulations required us to.” We have to go further and make it something that delivers value to customers.</p><p><strong>Matsumoto:</strong> That is a very demanding test. If someone told you, “This PET bottle is made from recycled material, so this bottle of water costs 10,000 yen,” you wouldn’t buy it, would you? [Laughs]</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wJE6NHlQyFTAApQIRfAhLg.jpeg" /></figure><p><strong>Ogawa:</strong> No, you wouldn’t. And recycling is not automatically better for the environment, either. If you use huge amounts of energy collecting things, transporting them, heating them to high temperatures, cooling them again, and so on, you have to ask, “Is this really beneficial overall?”</p><p>That is why we have to look at everything: quality, performance, cost, and energy. Taking all of that into account, we have to create something that makes customers say, “Yes, I want that.”</p><p><strong>Takahashi:</strong> Economic viability is a prerequisite. At the same time, resources are finite. It may not be a question of oil running out any time soon. But if we think about our children and the generations beyond them, we have to ask whether we can keep using resources the way we do now.</p><p>So, while balancing economic viability, we need to develop the technologies to keep resources circulating, starting now. I think that is part of our responsibility.</p><p><strong>Ogawa:</strong> As an automotive company, it would be easy to focus solely on what makes economic sense. We could buy materials cheaply through existing supply chains and make products. But once we ask, “Can we keep doing that forever?” it becomes clear that this alone is not enough.</p><p>That doesn’t mean anything is acceptable just because it is for the future. Development always finds itself caught between “We have to do this for the future” and “But how do we put food on the table tomorrow?”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LfeCvneyXNB9ECRKBgfr-Q.jpeg" /></figure><p><strong>Takahashi:</strong> It is the same at Toray. And that is exactly why I think it is difficult for one company to do this alone. Even with the technology in place, resources won’t circulate without a collection system. Without sufficient volumes, it is hard to make the economics work. Technology alone won’t make it viable.</p><p><strong>Matsumoto:</strong> Ultimately, I think it comes down to putting the idea out there and saying, “Who’s with us?” Nothing will change unless we take the first step. How many people and companies will see what we are doing and say, “Let’s do this together”? That, too, is part of the challenge.</p><p><strong>Ogawa:</strong> A few decades from now, things that make people ask, “Can you really do that?” today may be commonplace. When that time comes, we can’t afford to have nothing ready. It is a bit of a chicken-and-egg question, but at the very least, we need to have an egg ready.</p><h3><strong>Laying the Groundwork Now for an Uncertain Future.</strong></h3><p><strong>Matsumoto:</strong> Research is ultimately about laying the groundwork for the future, isn’t it? If it could generate revenue tomorrow, that would be the easiest thing to explain. But that is not all there is to it.</p><p><strong>Ogawa:</strong> There is uncertainty involved.</p><p><strong>Matsumoto:</strong> Exactly. Someone has to take that risk. Looking to the future, making a certain level of commitment, and laying the groundwork — that is also part of research’s role.</p><p><strong>Ogawa:</strong> I think a bottom-up approach is powerful when you don’t know how things will turn out. It starts with someone’s desire to do something. That doesn’t mean every project should continue. But if you refuse to begin because you don’t have the answer from the outset, nothing new will emerge.</p><p><strong>Takahashi:</strong> Not every business at Toray has grown steadily all the time, either. There have been difficult periods, when other businesses supported us, followed by the growth of the next business.</p><p><strong>Matsumoto:</strong> Some things would make you want to stop if you looked only at today’s numbers.</p><p><strong>Takahashi:</strong> Yes. But sometimes, because we persevere, those efforts grow into something substantial later. That is why I think a long-term perspective is necessary.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YHT-BpusOF6XXlXWVaj6Hw.jpeg" /></figure><p><strong>Ogawa:</strong> That sounds a lot like Honda, doesn’t it?</p><p><strong>Matsumoto:</strong> I think it does. We both value technology, and neither of us gives up easily. [Laughs]</p><p><strong>Ogawa:</strong> We’re both pretty tenacious. [Laughs] I don’t think this research would have come this far if either company had been doing it alone. It began with Honda bringing in an idealistic idea, and Toray wondering, “Can that really be done?”</p><p>But instead of stopping at “That’s impossible,” you faced the reality with us and asked, “What makes it difficult?” and “How could we make it work?” Now we are even sharing the pain of it.</p><p><strong>Matsumoto:</strong> We’ve reached that point, haven’t we?</p><p><strong>Ogawa:</strong> Then we might as well enjoy it. We still don’t have the answer. But what began as blue-sky thinking has made it this far. So we can keep thinking through what comes next, together. That is how I see it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IhjJSf-IURN3-76LCezfbQ.jpeg" /></figure><p>■ Related Articles</p><ul><li><a href="https://medium.com/@HGRX/materials-research-supporting-hondas-future-from-minus-one-creating-the-invisible-foundations-fba55a0ad75e">Materials Research Supporting Honda’s Future from “Minus One” — Creating the Invisible Foundations…</a></li><li><a href="https://medium.com/@HGRX/beyond-impossible-delivering-desirable-c43b58a12b60">Beyond “Impossible,” Delivering “Desirable”</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fd09f78c7d9a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Gets Harder the Moment It Has a Body.—]]></title>
            <link>https://medium.com/@HGRX/ai-gets-harder-the-moment-it-has-a-body-1723d126e4cb?source=rss-fd02d7b9dda5------2</link>
            <guid isPermaLink="false">https://medium.com/p/1723d126e4cb</guid>
            <dc:creator><![CDATA[HGR / Honda R&D Center of Excellence]]></dc:creator>
            <pubDate>Wed, 28 Oct 2026 03:01:05 GMT</pubDate>
            <atom:updated>2026-10-28T03:01:05.238Z</atom:updated>
            <content:encoded><![CDATA[<h3>AI Gets Harder the Moment It Has a Body.— How Honda Is Turning Physical AI into Technology That Works in the Real World</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ta8TW1ku5KhHA2H2LyvIlQ.jpeg" /></figure><p>Hello everyone. I’m Akinobu Hayashi, a researcher in Honda’s Frontier Robotics domain. While coordinating Physical AI research within the Elemental Research Block, I’m also working on a project to improve the success rate of tasks performed with a multi-fingered hand. My specialty is software, particularly machine learning.</p><p>AI that handles language and images has already become part of our everyday lives. But the moment AI acquires a physical body and tries to touch and move objects in the real world, things suddenly become much more difficult.</p><p>At Honda, we’re taking on these challenges of the physical world through not only AI, but also hardware such as multi-fingered hands, sensors, and control systems.</p><p>In this article, I’ll be talking with Shuhei Kondo, who is working on the automated assembly of wire harnesses in automobile factories, and Naoki Morihira, who researches robot learning technologies. Together, we’ll explore where Physical AI stands today and the difficult problems we face.</p><h3>A Robot May Not Even Realize It Has Failed.</h3><p><strong>Hayashi: </strong>Let’s start by clarifying what we mean by Physical AI. AI that deals with language and images has become quite familiar to us. Now, that AI is acquiring a body and beginning to interact physically with other things in the real world. To me, that’s what the term “Physical AI” represents.</p><p>Our particular focus is on using robots equipped with hands to physically manipulate objects.</p><p><strong>Kondo: </strong>Until now, getting a robot to perform a task has required a great deal of preparation. We had to design the robot’s movements and plan how it would approach and pick up an object based on visual information.</p><p>With Physical AI, I hope we’ll eventually reach a point where anyone can easily use a robot for a wide variety of tasks simply by giving it verbal instructions.</p><p><strong>Morihira: </strong>Compared with conventional industrial robots, one of the biggest differences is autonomy. In the past, people defined rules such as “move this way in this situation” or “weld here.”</p><p>The shift is toward robots that can assess their surroundings, make autonomous decisions, and carry out tasks such as assembly on their own. I think that’s a major difference.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*xlhhEczK4jkc1nKB" /><figcaption>From left: Naoki Morihira, Akinobu Hayashi, and Shuhei Kondo</figcaption></figure><p><strong>Hayashi: </strong>But AI that actually moves a physical body faces challenges that are quite different from those involved in language or images, doesn’t it?</p><p><strong>Morihira: </strong>I think there are two major challenges. The first is that “the robot has to keep making decisions continuously, even as its own actions change what happens next.”</p><p>Suppose you move a robot once and it ends up two or three millimeters off target. That changes the relative positions of its end effector and the object, so the next camera image and contact conditions will be slightly different from what the training data anticipated. If the robot makes its next decision in that state, the error can grow even larger. The errors snowball.</p><p>The second challenge is that “the real world itself is simply too complex and varied.” Objects differ in type, color, and shape, while their orientation, friction, and other properties vary continuously.</p><p>The possible combinations are practically endless. A robot will almost never encounter exactly the same situation in actual operation that it saw during training.</p><p><strong>Hayashi: </strong>And sometimes a robot doesn’t even realize that it has failed.</p><p>In robotics and machine learning, there’s a concept called “partial observability” (*1). A robot judges its situation based on information from its sensors. It doesn’t have complete knowledge of everything in the world around it.</p><p>If it doesn’t know how hard an object is, for example, it can’t fully know the friction involved, either. It acts while making estimates from the information available — essentially thinking, “It’s probably like this.” Even when things gradually drift out of alignment, it may not recognize the deviation itself. If it then reaches a state it has never experienced before, it may have no idea how to recover.</p><p><em>*1 Partial observability: A condition in which a robot or AI cannot directly perceive the entire state of its environment and must make decisions by estimating that state from limited information, such as sensor data.</em></p><p><strong>Morihira: </strong>In a game or simulation, if something goes wrong, you can press the reset button and start over. But in the physical world, a robot might drop a part somewhere it can’t retrieve it, damage an object, or even break itself. This irreversibility — the fact that some mistakes can’t simply be undone — is another challenge unique to the real world.</p><h3>Not Everything Has to Be Left to AI — The Gap Between “Moving” and “Being Useful”</h3><p><strong>Hayashi: </strong>The wire-harness work you’re doing, Kondo, is exactly the kind of task that’s been difficult to automate until now, isn’t it?</p><p><strong>Kondo: </strong>Yes. But I don’t think the answer, at least for now, is to use AI for absolutely everything. There are still challenges in handling a wide variety of tasks and achieving high success rates. Training also requires enormous amounts of data and time. And the speed of execution is another issue.</p><p>That’s why we’re currently combining AI with the rule-based technologies we’ve developed over the years. Each has its advantages and disadvantages, so we’re at the stage of finding the right combination to accomplish each task.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*CRQFSKbiqGXIcEU6" /></figure><p><strong>Hayashi: </strong>I think that’s really important, too. We don’t leave everything to a large-scale model. In some parts of the system, we still use traditional recognition modules and motion controllers.</p><p>Deciding what today’s AI should handle and what should be left to conventional programs — and finding the right balance — is itself one of the difficult parts.</p><p><strong>Kondo: </strong>We rely heavily on simulation for training. Imagine hundreds of robots running inside a simulator, repeating a motion such as inserting a connector thousands or tens of thousands of times.</p><p>We test things like how to move a flexible object such as a wire harness so that its clip becomes easier to grasp. We learn those strategies in the simulator, deploy the learned policy on a real robot, and check how it performs. Then we repeat the cycle.</p><p><strong>Hayashi: </strong>Even when a person inserts a single connector, they don’t necessarily line it up perfectly from the start. They feel around a little, touching the edge of the opening as they guide it into place. Does a robot need to make that kind of movement, too?</p><p><strong>Kondo: </strong>Exactly. The robot estimates the connector’s position and orientation and moves it to the assembly location, but there will inevitably be an error of one or two millimeters.</p><p>If it tries to push the connector straight in, the edge can hit the opening and prevent insertion. So we teach it to probe slightly as it inserts the connector, or to produce a movement based on experience that says, “If I move in this direction, it’ll go in,” even when its perception is a little off. We use machine learning for that, too.</p><p><strong>Morihira: </strong>Even then, a learning-based method won’t succeed 100% of the time on its first attempt. For example, suppose the success rate per attempt is 90%. If we have a mechanism that allows the robot to retry correctly after one of the remaining 10% of attempts fails, permitting up to two attempts can bring the overall success rate to around 99%. The average task time would increase by only about 10%. The important thing is to design learning and the system as a whole around how to recover from failure, not just the first-attempt success rate.</p><p><strong>Hayashi: </strong>Even if the robot learns in simulation, transferring that learning to the real machine presents a different set of difficulties, doesn’t it?</p><p><strong>Kondo: </strong>It does. Collecting data and training repeatedly on an actual robot is difficult and can be dangerous, which is why we want to use simulators.</p><p>But it’s hard to make the physical models in a simulator match the physical properties of the actual environment perfectly. A movement that works in simulation may not play out the same way on the real robot.</p><p>You can’t use simulation effectively without getting Sim-to-Real and Real-to-Sim (*2) right. I think this is a challenge shared by researchers around the world.</p><p><em>*2 Sim-to-Real: Applying control methods or AI developed and trained in a simulated environment to real machines and real-world settings. A key challenge is bridging the gap between simulated and actual physical properties.</em></p><p><em>Real-to-Sim: Incorporating data and physical properties measured in the real world into a simulation, making the simulated environment more realistic and enabling training and verification under conditions closer to reality.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*HPhOKvi-kmHRRoir" /></figure><p><strong>Hayashi: </strong>I imagine the way you provide observation data can also cause problems.</p><p><strong>Morihira: </strong>Yes. For example, we sometimes first train a teacher model using information that’s only available inside the simulator. Then we use a technique called “distillation” to transfer its knowledge to a policy that operates using only information available on the real robot, such as camera images and tactile sensor readings.</p><p>But sometimes the observations we’re giving the model aren’t sufficient for it to make a confident decision or even tell whether it has achieved its goal. And yet we find ourselves asking, “Why can’t it do this?” That’s why I think it’s actually quite important to put ourselves in the model’s shoes and consider things from its point of view. What does the model know? Where does it get confused? Starting with those questions can sometimes reveal a new approach.</p><h3>Start with Intelligence Closest to the Body — What People Do Effortlessly Is Hard for AI</h3><p><strong>Hayashi: </strong>Within Physical AI, we sometimes divide the work into two broad roles.</p><p>At the higher level, there’s a layer that uses images and language to decide, in broad terms, “what to do.” Beneath that is a layer that determines “how to move,” according to the body of each particular robot.</p><p>If we call them System 2 and System 1 (*3), the higher-level System 2 is something we’re working on in collaboration with other companies and research institutions. System 1, on the other hand, is closer to the body. Because Honda develops the hardware itself, our current thinking is that we should take responsibility for that part ourselves.</p><p><em>*3 System 1 / System 2: In this article, these terms are used to explain the division of research roles. System 2 refers to explicit reasoning and higher-level decision-making using language and other information, while System 1 refers to intuitive, automatic movements closely linked to the body.</em></p><p><strong>Morihira: </strong>If System 2 is the kind of explicit thinking we can put into words, System 1 is closer to the physical movements we perform every day with almost no conscious thought.</p><p>No matter how intelligently the higher-level AI can plan, it can’t complete a real-world task unless the intelligence that actually moves the body is reliable.</p><p>That’s why, rather than immediately aiming for a general-purpose robot that can do anything when instructed in natural language, I think a more realistic path to practical use is first to develop embodied intelligence that can reliably complete individual tasks with a high success rate in an environment where the process and objects are defined. Then we can expand the range of tasks it can handle as we build up experience and data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*D0nHoAWieuT4j45L" /></figure><p><strong>Hayashi: </strong>Those physical movements people perform without even thinking are actually extremely difficult, aren’t they? If you compared the dexterity of today’s robot AI to that of a person, how old would the person be?</p><p><strong>Kondo: </strong>I have a two-year-old daughter, and she’s much more dexterous than the robot! (Laughs) She can fit blocks together and assemble things. But getting a robot to do the same thing with general-purpose AI is still difficult today.</p><p><strong>Morihira: </strong>Even children around eighteen months old can fit objects together or stack them without adults having to teach them every little detail.</p><p>This relates to what’s known as “Moravec’s paradox” (*4). For AI, logical tasks that even adults find difficult, such as playing chess, are relatively easy, while actions a one-year-old performs naturally, such as stacking blocks, are extraordinarily hard.</p><p>From early childhood, people learn the “rules of the world” — gravity, friction, how hard objects are — by touching and playing with things around them, even without a clear purpose.</p><p>Today’s task-specific AI, meanwhile, can learn patterns like “when this information comes in, move this way,” but that doesn’t mean it understands the underlying workings of the physical world — how the world will change if it applies force in a particular way.</p><p><em>*4 Moravec’s paradox: The observation that perception and physical actions performed naturally by infants can be more difficult for AI and robots than intellectual tasks, such as chess, that seem highly sophisticated to humans.</em></p><p><strong>Hayashi: </strong>If we want to put this technology into practical use, rather than trying to give robots the full versatility of a human right away, we need to start by enabling them to perform a specific task in, say, an assembly process, with a high success rate and without taking too much time. Then we can expand from there.</p><p>We’re also targeting processes that conventional industrial robots, including those equipped with today’s grippers, can’t handle.</p><p><strong>Morihira: </strong>One important point in accomplishing those tasks is that “embodied intelligence is closely tied to the structure of the body.” Language and mathematics are relatively recent concepts in human history, emerging over the past several thousand years. But feedback loops between vision, touch, and physical movement — such as seeing a predator and fleeing, or catching prey — have existed for hundreds of millions of years in the history of life.</p><p><strong>Hayashi: </strong>How does that idea of “hundreds of millions of years” connect to hardware?</p><p><strong>Morihira: </strong>Imagine leaping from tree to tree. If your senses or movements were even slightly off, you could fall and be eaten by a predator. Living things have been going through this kind of life-or-death optimization for hundreds of millions of years. In that process, not only “intelligence,” such as brains and nerves, developed in isolation. Natural selection also shaped the physical body itself, including the structure of the limbs. Intelligence and the body coevolved, adapting to each other.</p><p>That’s why I think Honda has tremendous value in evolving not just the software, but also the hardware and the embodied intelligence running on it as one integrated whole.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*gLI75XRBFvpcwko3" /></figure><p><strong>Kondo: </strong>From the perspective of someone developing software, having good hardware makes a huge difference.</p><p>When we’re training, we basically want the robot to keep moving all the time. Reliability and durability matter enormously: whether we’re training it or demonstrating it to visitors, we want it to keep running without stopping.</p><p>One reason we want to bring the hand closer to a human hand is that, if robots can be used for tasks currently performed by people in factories, we may be able to limit the need for major additional investment in surrounding equipment.</p><p>But we’re not at the level of a human hand yet. Sometimes the robot can’t fit into a narrow gap or has difficulty working inside a box. People also make extensive use of touch. They can feel exactly where something is making contact or slipping. To achieve that, the hardware and sensors need to evolve further.</p><p><strong>Hayashi: </strong>In fact, the hardware and software teams sometimes push conflicting requirements at each other.</p><p>The software side might say, “The fingers can’t reach with this many degrees of freedom,” or “If we want greater dexterity, we need you to change this part.” And the hardware side comes back with, “That’ll raise the cost,” or “That’ll make it more complicated.”</p><p>It’s the same with sensors. The hardware team might say, “We gave you this amazingly sensitive sensor, so make sure you use it!” And we’d respond, “But that part doesn’t even make contact in the first place!” (Laughs)</p><p>In a sense, we’re not just software developers; we’re also users of the hardware. Through exchanges like these, both sides change as we work out what needs to improve.</p><p><strong>Morihira: </strong>From the software developer’s perspective, the ideal would ultimately be a hand like a human’s, with many degrees of freedom and perfect tactile sensors.</p><p>In reality, though, there are all kinds of hardware constraints, such as space, wiring, and durability. Instead of trying to achieve an ideal 100% through hardware alone, we might ask, “Even if this hand only has four fingers, could we still accomplish 80% of the target tasks with this hardware if we get creative with software control and learning?”</p><p>I think it’s important to put every option — hardware and software — on the table and find the best solution for the system as a whole.</p><h3>If We’re Not Aiming to Be the Best in the World, There’s No Point.</h3><p><strong>Hayashi: </strong>How do you feel about where we stand compared with others around the world?</p><p><strong>Kondo: </strong>When it comes to AI specialized for a particular product or process, like wire harness assembly, I think we’ve reached a very high level even by global standards, partly by combining AI with rule-based methods.</p><p><strong>Morihira: </strong>I also think we’ve developed some very good hardware. The next step is to draw out its full potential through software and bring it to a level where it can operate reliably in an actual factory. We still need to tackle the remaining practical challenges one by one, including making the system easy for anyone on the factory floor to use and developing the error-recovery capabilities we talked about earlier.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*uupb0XDr-VDH_UfU" /></figure><p><strong>Hayashi: </strong>I also believe we’re very competitive in hardware and in specialized AI for applications such as wire harnesses. In some areas, I think our accuracy is higher than that of other companies.</p><p>On the other hand, in the field of general-purpose AI — the so-called generalist systems that are attracting attention — AI companies and research institutions in the United States and China are particularly strong. We’re working hard to keep up while looking for ways to make the most of our own strengths.</p><p>On the System 2 side, we use publicly available models, and to be honest, there’s still a gap between us and the cutting edge. That’s precisely why we’re focusing on System 1, where we can compete with hardware and data that no one else has.</p><p>For System 1, we have data directly tied to our hardware that is unique to us. We want to use that advantage to create technologies that are genuinely distinctive and valuable. But progress in the field is extremely fast, so we also need to keep a close eye on how quickly others are developing their technologies as we continue to advance our own.</p><p><strong>Kondo: </strong>Our manager, Mr. Yoshiike, regularly tells us to be so focused on becoming number one in the world that, if a research topic can’t reach that level, we should stop pursuing it.</p><p>So aiming to be the best in the world has naturally become part of how we approach our research and development.</p><p>Honda has hardware built on years of accumulated expertise in mechanics and electronics. That gives us a platform for conducting world-class R&amp;D. Now those of us on the software side need to push ourselves further, until we can say, “We were the first in the world to accomplish this difficult task.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*UD9H26rSFsC5kXNR" /></figure><p><strong>Hayashi: </strong>I think that standard also reflects everything Honda’s robotics research has built up over the years.</p><p>The generation before mine includes people who worked on ASIMO. They were genuine global frontrunners, and they evaluate our research against the standards they established.</p><p>Our generation joined because we wanted to make a contribution of our own. And the even younger researchers come in full of energy, thinking, “We can do it, too.”</p><p>The senior generation doesn’t dismiss that attitude, either. They say things like, “Go ahead and try it — that sounds interesting!” I think our culture has been shaped by both sides.</p><p><strong>Morihira: </strong>The research environment is also very different from the ASIMO era. Back then, development was largely closed, with each company pursuing its own technologies independently. Today, the field has become much more open. Research institutes and companies around the world publish new papers and technologies every day, and we’re flooded with information. The pace is so fast that even keeping up with it all is difficult. We have to identify what really matters, make choices about what to adopt, learn it for ourselves, and then roll up our sleeves and turn it into something real. I heard the phrase somewhere, “Learn or die.” That’s about the level of urgency I feel. I approach this work believing that if we’re not aiming to be the best in the world, there’s no point.</p><h3>Recreate Human Ability, or Go Beyond It?</h3><p><strong>Hayashi: </strong>To wrap up, let’s each talk about the barrier we most want to overcome next.</p><p>I’d like us to get better at handling multimodal information.</p><p>Language, images, touch, sound — we currently feed all sorts of information into machine-learning models and have them generate actions. But what we really need is the ability to choose based on the situation: “I need this information right now,” or “Here I need information at a much faster rate.”</p><p>I think tactile information, in particular, can improveboth task success rates and movement speed. I want to use it effectively enough to develop robots that can move faster and more accurately.</p><p><strong>Kondo: </strong>My first goal is to bring Honda’s robotic hand into production.</p><p>To do that, we need reliable, durable hardwareand electronics, of course. But on the software side, we also have to design around the reality that AI won’t succeed 100% of the time. We need to think about how the robot can recognize its own failure and recover.</p><p>I want to build a system that uses tactile and other sensor information to detect its own mistakes and figure out what it should do next.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*2UQnBnMXGpoaEWqM" /></figure><p><strong>Morihira: </strong>I want to work on “AI that can move on its own to gather the information it needs to make decisions,” instead of passively solving tasks with whatever sensor data it’s given or merely imitating data collected by humans.</p><p>For example, when a person is about to lift a box without knowing what’s inside, they don’t immediately try to lift it with all their strength. They might first touch it lightly or give it a little push to gauge its weight and how it feels. In the same way, when AI encounters something it doesn’t know, I want it to be able to ask itself, “How should I move to get the information I need?” and actively interact with its environment. This field is called Active Perception (*5). I believe taking it further could bring robot intelligence to a whole new level, and I find that an incredibly exciting research topic.</p><p><em>*5 Active Perception: An approach in which a system actively gathers information needed for decision-making by taking actions such as touching or pushing, rather than simply waiting to receive information passively.</em></p><p><strong>Hayashi: </strong>We’ve talked a lot about bringing robots closer to human capabilities, but in fact, there isn’t just one way of thinking about this within our organization.</p><p>Some people want to “recreate what humans can do,” while others say, “Shouldn’t we aim to go beyond humans?”</p><p>Personally, I’m more in the camp that wants to go beyond.</p><p>Taking on things that are difficult for people and expanding what’s possible beyond the reach of human effort alone — that, too, is part of the value robots can offer.</p><p>■Related Articles</p><ul><li><a href="https://medium.com/@HGRX/the-further-this-hand-reaches-the-further-human-potential-can-expand-hondas-multi-fingered-hand-bacb297baccb">The Further This Hand Reaches, the Further Human Potential Can Expand — Honda’s Multi-Fingered Hand…</a></li><li><a href="https://medium.com/@HGRX/that-step-was-leading-closer-to-people-from-p2-to-the-multi-fingered-hand-the-question-passed-f110270604b7">That Step Was Leading Closer to People — From P2 to the Multi-Fingered Hand: The Question Passed…</a></li><li><a href="https://medium.com/@HGRX/hgr-cadence-part-1-what-does-the-physical-in-physical-ai-really-mean-2bbfdef1de24">［HGR × Cadence Part 1] What Does the “Physical” in Physical AI Really Mean?</a></li><li><a href="https://medium.com/@HGRX/hgr-cadence-part-2-where-physical-ai-will-be-won-beyond-correlation-will-the-day-come-when-1c55e248c2a7">[HGR × Cadence Part 2] Where Physical AI Will Be Won — Beyond Correlation, Will the Day Come When…</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1723d126e4cb" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[So We Can Keep Flying — Honda’s SAF Research: Connecting How Fuel Is Made and Used]]></title>
            <link>https://medium.com/@HGRX/so-we-can-keep-flying-hondas-saf-research-connecting-how-fuel-is-made-and-used-d9e3509b0338?source=rss-fd02d7b9dda5------2</link>
            <guid isPermaLink="false">https://medium.com/p/d9e3509b0338</guid>
            <dc:creator><![CDATA[HGR / Honda R&D Center of Excellence]]></dc:creator>
            <pubDate>Wed, 21 Oct 2026 03:01:02 GMT</pubDate>
            <atom:updated>2026-10-21T03:13:50.605Z</atom:updated>
            <content:encoded><![CDATA[<h3>So We Can Keep Flying — Honda’s SAF Research: Connecting How Fuel Is Made and Used</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0-01scLjkOvfyVw6eOfKhg.jpeg" /></figure><p>Hello everyone. I’m Osami Yamamoto, and I work in research and development for SAF production technologies.</p><p>I joined Honda in 2004 and spent approximately 15 years researching and developing automotive exhaust gas purification catalysts, including three-way catalysts and NOx storage catalysts. Today, I’m applying that experience to research on producing Sustainable Aviation Fuel (SAF), a carbon-neutral aviation fuel made from CO₂ and biomass.</p><p>SAF is expected to play an important role in achieving carbon neutrality in aviation. While automobiles and motorcycles have options such as electrification and hybrid powertrains, high-energy-density hydrocarbon liquid fuels are expected to remain essential for aircraft, particularly those traveling long distances.</p><p>We are now applying the technologies Honda has developed through decades of mobility research and development to fuel production itself. Working alongside colleagues specializing in materials, catalysts, reactors, and plants, we continue to take on the challenge of bringing SAF to society and, ultimately, to our customers.</p><p>In this article, I’m joined by Kouta Okamura, who researches biomass gasification technology; Nozomu Hirata, who develops catalysts for SAF production; Takahiro Tsuchiyama, who works on scaling up those catalysts for use in reactors; and Hideyuki Utsunomiya, who focuses on turning our technologies into operational plants. Together, we’ll discuss where our research stands today and what we hope to achieve next.</p><h3>Pursuing Two SAF Technologies on Different Timelines</h3><p><strong>Okamura</strong>: First, I’d like to explain why Honda is pursuing two different approaches to SAF production technology.</p><p>At Honda, we’re working on two technologies: Bio-FT SAF (Biomass-to-Liquid via Fischer–Tropsch synthesis), which produces SAF from biomass, and PtL SAF (Power-to-Liquid), which produces SAF using CO₂ and other feedstocks. I joined Honda in 2020 and have been involved in the Bio-FT SAF project ever since.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qQwK0DY-8qTVpDPwzXyGrg.jpeg" /><figcaption>(From left) Osami Yamamoto, Kouta Okamura, Nozomu Hirata, Takahiro Tsuchiyama, and Hideyuki Utsunomiya</figcaption></figure><p><strong>Yamamoto</strong>: The two approaches have different target timelines. For Bio-FT SAF, we’re aiming for commercialization around 2030, while PtL SAF has a longer-term horizon of about2035 to 2040.</p><p><strong>Okamura</strong>: In the SAF industry, having a proven track record is extremely important. Even if you say, “The technology is ready,” potential customers may not be able to adopt it unless you can demonstrate that it has actually operated successfully as a plant.</p><p><strong>Yamamoto</strong>: That’s why we want to first establish a track record in production and earn customers’ trust through Bio-FT SAF. We can then carry that experience and credibility forward to PtL SAF and expand commercialization. That’s why we’re pursuing two technologies that use different feedstocks.</p><h3>Applying Engine Expertise to Fuel Production</h3><p><strong>Yamamoto</strong>: Fuel is ultimately something that gets consumed. So when customers choose a fuel, price — and therefore production cost — is naturally an important factor. One factor affecting that cost is conversion efficiency.</p><p>How much of the feedstock can we convert into SAF in a single reaction? If conversion efficiency is low, we have to circulate unreacted material back through the process for another reaction, which requires additional energy. That drives up production costs.</p><p><strong>Tsuchiyama</strong>: When it comes to PtL SAF, I believe the catalyst that drives the reaction is at the heart of our competitiveness. At the same time, we can’t fully realize the catalyst’s conversion performance without pairing it with the right reactor. We’re considering commercializing the catalyst developed by Yamamoto and his team, along with our reactor, as an integrated system.</p><p><strong>Okamura</strong>: With Bio-FT SAF, we first gasify biomass, then use the hydrogen and carbon monoxide in the resulting gas to produce liquid fuel. Honda is researching a technology that adds additional hydrogen during gasification to improve the efficiency of hydrogen and CO production.</p><p><strong>Yamamoto</strong>: For Bio-FT SAF, we drew on our experience analyzing the chemical reactions involved in combustion in automotive engines. We analyzed the biomass gasification process to identify which reactions were creating bottlenecks.</p><p>That led us to the idea that adding hydrogen during gasification might accelerate the reaction.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*HtdpH2m_ePQE350s" /></figure><p><strong>Okamura</strong>: Gasification and engine combustion are not the same, of course, but there are similarities in how we predict their reactions. Based on simulations, we predicted that introducing hydrogen at a particular point might shift the reaction toward more desirable products. When we tested the idea, that’s exactly what happened.</p><p><strong>Tsuchiyama</strong>: From there, the next challenge is reproducing the performance demonstrated in the lab at a scale where the technology can actually be used. That’s why the catalyst and reactor teams need to work closely together.</p><p><strong>Okamura</strong>: When I first joined Honda, we started by introducing hydrogen into a simulated gas representing the gas produced from biomass. From there, we progressed to using actual biomass. Today, we’ve reached the point where we can confirm the effects of hydrogen addition in bench-scale equipment.</p><p>But it isn’t simply a matter of adding hydrogen. The optimal amount depends on the biomass used and the operating conditions. We want to refine the technology until we can confidently say, “Honda’s hydrogen introduction technology delivers the best performance.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*PeZIGOvNe01yREuZ" /></figure><p><strong>Yamamoto</strong>: PtL SAF also builds on technologies Honda has developed over the years. In automotive research, I worked on exhaust gas purification catalysts, including three-way catalysts and NOx storage catalysts.</p><p><strong>Hirata</strong>: I joined Honda as a mid-career hire in 2025, and I’m now responsible for developing catalyst materials. I also studied catalysts as a student so I could apply that experience to my current work.</p><p><strong>Yamamoto</strong>: NOx and CO₂ exhibit common characteristics related to basicity that are relevant to catalyst material design. Drawing on our experience and knowledge from NOx storage material research, we’ve been exploring which promoter elements to add to the catalyst, and where to position them, to improve the performance of catalysts used for PtL SAF production.</p><p><strong>Hirata</strong>: We already have catalysts that we can confidently describe as good. But my job is to make them even better.</p><p>I want to develop catalysts that let us say with complete confidence, “If you use this catalyst, you can reliably produce high-quality fuel,” or “We can shorten your existing process by this much.” That’s the level I want to reach.</p><h3>Waiting Until It’s Finished Is Too Late</h3><p><strong>Tsuchiyama</strong>: It’s not as simple as completing the catalyst first and then scaling it up. When we move from the laboratory to bench scale, new challenges emerge that we couldn’t see before.</p><p>If we finish developing a catalyst, scale it up, and only then discover a fundamental problem, we could end up having to redo a significant amount of work. That’s why I believe catalyst material development and scale-up need to proceed in parallel.</p><p><strong>Hirata</strong>: Creating something from zero to one takes tremendous effort, and so does taking something from one to 100. But the stages in between aren’t separate. They’re seamlessly connected.</p><p><strong>Utsunomiya</strong>: I’m involved in both Bio-FT SAF and PtL SAF, focusing on scale-up and the considerations needed to turn these technologies into operational plants. That means thinking not only about the plants themselves, but also where to build them and what facilities and equipment they’ll need.</p><p><strong>Tsuchiyama</strong>: When we encounter problems at bench scale, we can feed those findings back to the catalyst team and accelerate development on both sides. That’s exactly the kind of work we should be doing in parallel.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*0OzzJ3EfJCI27AV6" /></figure><p><strong>Hirata</strong>: We actually had a catalyst material that we thought would only work well at a certain temperature. But the reactor team suggested, “Wouldn’t it be better to lower the temperature a little?”</p><p>Rather than sticking stubbornly to our original assumptions, we said, “Okay, let’s give it a try.” And when we did, the performance was surprisingly good.</p><p><strong>Utsunomiya</strong>: That’s one of the benefits of having material researchers and equipment researchers working closely together. We can exchange feedback very quickly.</p><p><strong>Hirata</strong>: Just recently, there was another approach that looked promising in theory, but I was hesitant to test because I thought it might be difficult to use in practice. The reactor team decided to experiment with it, and the results exceeded our expectations.</p><p>Seeing what actually happened when we tried it gave me a lot of encouragement. We haven’t overcome the challenge yet, but I feel like we’ve found a way forward.</p><h3>When Your Raw Material Is Wood, No Two Experiments Are the Same</h3><p><strong>Okamura</strong>: Bio-FT SAF presents a somewhat different set of challenges. Biomass gasification is generally treated as a chemical reaction, but quite a few phenomena are involved that can’t be fully explained by chemical reactions alone.</p><p>Many different reactions occur simultaneously and interact with one another, and even within the broader scientific community, not all of them are fully understood.</p><p><strong>Hirata</strong>: When I listen to the discussions happening next to me, I sometimes hear questions like, “What kind of wood is this, and where did it come from?” I remember being surprised by that at first.</p><p><strong>Okamura</strong>: Even when the wood comes from the same species of tree, its properties can vary depending on where it was grown. Different soil conditions mean that trees absorb different substances as they grow.</p><p>So even when someone tells us, “It’s the same kind of wood,” the experimental results can turn out completely differently.</p><p><strong>Utsunomiya</strong>: Those differences become even more important when we move toward actual plant operations. If you change regions, the feedstock supply chain structure and the climate will also change. We have to think about how to accommodate those variations.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*8yH9g6dR3F2lyM-H" /></figure><p><strong>Okamura</strong>: Sometimes, we can’t even get to the experiment we originally wanted to conduct.</p><p>Biomass particle size varies, and so does its moisture content. Depending on the biomass condition, it may clog the feeding system. Or too much tar may form during gasification and block pipes further downstream.</p><p><strong>Utsunomiya</strong>: Do we design equipment that can handle virtually any kind of feedstock? Or do we specify certain requirements and say, “The feedstock has to meet these conditions”?</p><p>How much flexibility should we build into the equipment? Those are important considerations when designing a plant.</p><p><strong>Okamura</strong>: That’s why, with Bio-FT SAF, we’re not necessarily trying to do everything ourselves at Honda. We’re also considering collaborating with manufacturers that have technologies to maintain stable gasification operations and combining their expertise with Honda’s hydrogen introduction technology.</p><h3>Talking Technology, Regardless of Position</h3><p><strong>Hirata</strong>: When I joined Honda as a mid-career hire, one of my first impressions was, “This is a young team.”</p><p>At my previous company, I was the youngest on the team, and even the next-youngest colleague was about ten years older than me.</p><p><strong>Utsunomiya</strong>: I also joined Honda mid-career. I imagined that because it was such a large company, meetings might be more formal and the hierarchy more rigid.</p><p><strong>Hirata</strong>: But here, it’s easy to talk to people. If someone sits nearby, I ask them directly. If they’re not around, I call or message them.</p><p>People from outside Honda sometimes tell us, “The younger employees at Honda really speak up, don’t they?”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*_kvL4FH5tUn1uI5M" /></figure><p><strong>Okamura</strong>: I joined Honda straight out of university, so this is the only environment I’ve ever known. But people who join mid-career or come from outside the company sometimes tell me, “There’s a unique atmosphere here.”</p><p><strong>Yamamoto</strong>: I think part of that comes from the fact that this department itself is still relatively new.</p><p>People who wanted to take on new technologies and explore new fields for Honda came together, and everyone started on a fairly equal footing. Maybe that’s what helped create today’s atmosphere.</p><p><strong>Utsunomiya</strong>: I feel that Honda really encourages people to take on challenges. When someone says, “I’d like to try this,” or “I want to do things this way,” people listen first rather than immediately dismissing the idea.</p><p><strong>Tsuchiyama</strong>: That’s true of Waigaya as well. Rather than rejecting someone’s opinion right away, everyone brings different ideas to the table and discusses them. I think that’s a great part of Honda’s culture.</p><p><strong>Yamamoto</strong>: When I moved into a management role, I made a conscious decision to listen to people first and avoid dismissing their ideas from the outset.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*iK3nqp_n0JOosjR5" /></figure><p><strong>Hirata</strong>: Even before joining Honda, I had the impression that it was a company that placed great importance on technology. That was actually one of the reasons I chose to work here.</p><p><strong>Yamamoto</strong>: In the moment, it might be faster for someone to make a decision and say, “Never mind, we’re going this way.”</p><p>But that approach can lead to having to redo work later. So even if it means slowing down a little, I try to make sure we discuss things together and find the best possible solution.</p><h3>Four Years Ago, This Fuel Didn’t Yet Exist</h3><p><strong>Yamamoto</strong>: Four years ago, our research was featured in Honda Stories, Honda’s owned media platform. Looking back, one of the biggest changes is that our scale-up research has progressed to the point where we can actually produce this clear liquid fuel.</p><p>Back then, we were still at a stage where we could only make catalysts in the laboratory.</p><p><strong>Okamura</strong>: I think Bio-FT SAF was also at the stage where we had only just begun experimenting with actual biomass in small-scale equipment.</p><p><strong>Yamamoto</strong>: Since then, we’ve established a reactor research project focused on scaling up our catalysts, and plant engineers have joined the team to help us think about the next stages.</p><p>I feel our research has matured.</p><p><strong>Okamura</strong>: With Bio-FT SAF, we’re now at the stage of actively talking with people from other companies.</p><p>We’ve continued presenting our research at academic conferences and other venues, so some people already know about our work. They’ll say things like, “Honda is working on biomass, right?” or “You’re using hydrogen in the process, aren’t you?”</p><p>Recently, we’ve started to see the benefits of communicating our research outside the company.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*zhAODf4K_Nanauje" /></figure><p><strong>Yamamoto</strong>: I believe we can’t take the next step unless people know what we’re doing.</p><p>Looking ahead to the next four years, I think we’re entering a phase where we need to take our technologies beyond Honda, find partners, and move toward real-world implementation.</p><p><strong>Okamura</strong>: First, we want to establish a track record for Honda through Bio-FT SAF. Then we can build on that experience and connect it to the next stage with PtL SAF.</p><p><strong>Utsunomiya</strong>: From my perspective, the challenge is figuring out how to reproduce the performance achieved through research in actual equipment.</p><p>We want to preserve our research results and deliver the same high level of performance at a practical scale. Ideally, we’d like to go even further and improve that performance.</p><p><strong>Yamamoto</strong>: At Honda, we’re not only working on “making” SAF. We’re also pursuing initiatives related to “using” SAF and establishing the rules and standards needed for its adoption.</p><p>Because we have gas turbine research and development facilities, we can evaluate SAF’s combustion characteristics and compare them with those of conventional fossil fuels. This lets us identify which characteristics are equivalent and where the differences lie.</p><p><strong>Okamura</strong>: With Bio-FT SAF, too, we don’t want to stop at demonstrating that the technology works. We want to take it all the way to commercialization and establish a proven track record for Honda.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*2gc-xXMRC1Otk8qm" /></figure><p><strong>Yamamoto</strong>: SAF alone won’t be enough to achieve carbon neutrality in aviation. We also need to advance and improve gas turbine technology and optimize flight routes.</p><p>We need to approach the challenge from both sides: those who make SAF and those who use it.</p><p>As I mentioned earlier, I developed three-way catalysts. When a catalyst I had helped develop was installed in a customer’s car, it made me genuinely happy.</p><p>The component was almost completely invisible from the outside. But even so, knowing that something I had researched and developed was being used in a customer’s vehicle — and that it was helping provide that customer with the joy of mobility — gave me a real sense of accomplishment.</p><p>I want to take SAF to that same point.</p><p>I want us to fly using the SAF we’ve made ourselves. And I want our work to give people the freedom to move wherever they want.</p><p>I think it’s only when we reach that point that we’ll truly feel the significance of everything we’ve put into this research.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d9e3509b0338" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Teaching AI a Captain’s “Feel.”—]]></title>
            <link>https://medium.com/@HGRX/teaching-ai-a-captains-feel-f43e049dc1c8?source=rss-fd02d7b9dda5------2</link>
            <guid isPermaLink="false">https://medium.com/p/f43e049dc1c8</guid>
            <dc:creator><![CDATA[HGR / Honda R&D Center of Excellence]]></dc:creator>
            <pubDate>Wed, 14 Oct 2026 03:01:02 GMT</pubDate>
            <atom:updated>2026-10-14T03:01:02.554Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>Teaching AI a Captain’s “Feel.”— How Honda Is Refining Human-Machine Collaborative Intelligence on Florida Waters [Part 2]</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2oLSLl5fLCf10NPg4I0rQA.jpeg" /></figure><p>Hello everyone. I’m Atsushi Ogawa, Chief Operating Officer of Honda R&amp;D Research Center of Excellence (HGR).</p><p>In Part 1, I went out onto the water in Florida and came to understand that thinking about intelligent marine systems requires us to understand not only the boat itself, but also the marine environment, how boats are used, and the people who spend time on the water.</p><p>And at the end of that discussion, we came to the “feel” that people who operate boats develop through experience. They sense the wind, the current, and the condition of the boat, then naturally choose the right action for the situation. So how can AI learn from that kind of human experience?</p><p>Continuing from Part 1, I spoke again with Chad Nere and software engineer Berker Logoglu to take a deeper look at the intelligent systems research being carried out in the field here in Florida.</p><h3><strong>From “Seeing” to “Understanding”</strong></h3><p><strong>Tajima</strong> The rules on the water are more ambiguous than on the road, aren’t they? It is not simply a matter of moving out of the way when another boat comes toward you.</p><p><strong>Chad</strong> On the road, lane widths are fixed, and everyone understands what white and yellow lines mean. On the water, things are not that clearly defined.</p><p><strong>Tajima</strong> Even when there is a channel, not everyone follows exactly the same path.</p><p><strong>Chad</strong> Right. Some people travel down the center, while others stay closer to one side. How much distance you leave from another boat also depends on the situation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*J2emFOihNLsoVTO7" /><figcaption>（From left）Atsushi Ogawa, Chad Nere, Tomoharu Tajima</figcaption></figure><p><strong>Ogawa</strong> We passed a boat that was fishing today as well.</p><p><strong>Chad</strong> Even if the rules technically allow you to pass at a higher speed, going by at high speed right next to them would not be good boating etiquette. You slow down a little and give them space. An experienced captain does that naturally.</p><p><strong>Ogawa</strong> So “as long as we do not collide” is not enough.</p><p><strong>Chad</strong> Yes. It is important not only to be safe, but also to behave in a way that feels natural to the people around you.</p><p><strong>Tajima</strong> Getting AI to do that sounds extremely difficult.</p><p><strong>Berker</strong> To do it, the system first has to understand its surroundings properly. Simply detecting that “there is an obstacle ahead” is not enough.</p><p><strong>Ogawa</strong> It has to understand what is there?</p><p><strong>Berker</strong> Whether it is a personal watercraft, a fishing boat, a sailboat, or a large vessel. Our own boat needs to behave differently depending on what it is.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*AAIAgpfhHAZcQlc4" /><figcaption>(From left) Chad Nere, Atsushi Ogawa, Berker Logoglu</figcaption></figure><p><strong>Chad</strong> Sailboats are a clear example. Even with the same boat, how you treat it differs depending on whether it is under engine power or sailing under canvas.</p><p><strong>Ogawa</strong> So it is not enough to say, “The shape is the same, so it is the same boat.”</p><p><strong>Chad</strong> Right. If it is sailing, you need to give it more room.</p><p><strong>Berker</strong> That is why cameras are important. Radar can tell you that an object is there and how far away it is, but in some cases it cannot tell you what kind of boat it is or what state it is in.</p><p><strong>Tajima</strong> So the system does not just “see.” It “understands the meaning” of what it sees.</p><p><strong>Berker</strong> What we need is not only to detect what is around us, but to understand “what it is.” How our boat should move changes depending on what kind of vessel the other one is and what state it is in.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*0556T1EOTi4Du_wA" /></figure><p><strong>Ogawa</strong> People do that unconsciously all the time. When we see a boat in the distance, even if we cannot identify the exact type, we still predict, “It will probably move like this.”</p><p><strong>Berker</strong> That is an important point. We cannot build a system where AI is unable to act unless it can perfectly classify every boat. It needs to make reasonable judgments even when it encounters a boat or situation it has never seen before.</p><p><strong>Chad</strong> That is another reason we need real-world data. Even the same kind of sailboat can vary in size and color. Some days the sky is bright, other days it is dark, and the color of the water changes too.</p><p><strong>Berker</strong> Data is extremely important for training AI. We are collecting data from different locations, times of day, boats, and environments so the system can also respond to situations it is seeing for the first time.</p><p><strong>Ogawa</strong> This is another place where what we discussed in Part 1 — “what is possible because we are in Florida” — comes into play.</p><p><strong>Chad</strong> Even on a weekday, you can see this many different boats. It is a very good environment for collecting data as well.</p><h3><strong>Turning Expert Skill into Algorithms</strong></h3><p><strong>Ogawa</strong> But collecting a lot of data alone does not mean the system can operate a boat the same way a person can, right?</p><p><strong>Berker</strong> Right. In marine applications especially, the experience of skilled operators is extremely important.</p><p><strong>Tajima</strong> The “feel” Chad mentioned at the end of Part 1.</p><p><strong>Chad</strong> Exactly. Captains are not calculating everything numerically. For example, if a boat that had four people on board last time has eleven people the next time, of course the boat will respond differently. But the captain is not thinking, “There are seven more people, so I should change the steering by exactly this many degrees.” They feel how much the boat responds when it turns, whether the bow is riding high or low, and how it feels under acceleration. They read all kinds of signals and adjust their operation naturally.</p><p><strong>Berker</strong> And even under the same wind, different boats respond differently. The pontoon we rode today has a large canopy, for example. That alone changes how the wind acts on the boat. Even the same model can behave differently if its configuration changes. So it is extremely difficult for humans to program every possible condition one by one as explicit rules.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*z6BbM18blZ41oTzQ" /><figcaption>The pontoon boat used that day. Supported by cylindrical flotation tubes beneath the deck, pontoons are highly stable and resistant to rocking, making them popular for activities such as parties on the water and relaxed cruising.</figcaption></figure><p><strong>Tajima </strong>So instead, you have the system learn directly from how people operate the boat.</p><p><strong>Berker</strong> Yes. With approaches such as reinforcement learning, imitation learning, and VLA (Vision-Language-Action) models — AI models that connect visual and language information to action — learning from the behavior of skilled operators becomes important.</p><p><strong>Ogawa</strong> So the way Chad and the team handle a boat can become training data.</p><p><strong>Berker</strong> That is how we see it. How does a skilled captain move the boat in strong winds or current? The AI learns from those actions.</p><p><strong>Chad</strong> Some of that experience has already been incorporated in very concrete ways.</p><p><strong>Ogawa</strong> Trailer loading?</p><p><strong>Chad</strong> In an early version of the system, as the boat approached the trailer, it would shift into neutral partway through and coast the rest of the way.</p><p><strong>Ogawa</strong> For a car, that almost sounds natural.</p><p><strong>Chad</strong> But on a boat, the moment you shift into neutral, a crosswind or current can push you off course very easily. So you keep some propulsion all the way to the end, stay in gear until just before the boat reaches the trailer, and keep controlling its heading.</p><p><strong>Berker</strong> We changed the algorithm accordingly — how fast the boat approaches, when to align the hull, and when to reduce the throttle. If the wind or current is strong, we preserve a little more momentum. The experience of the Florida team is already being incorporated into the control logic.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*KbDDfXj57hip0FT8" /><figcaption>With hands off the wheel, the boat automatically enters the trailer.</figcaption></figure><p><strong>Ogawa</strong> That is interesting. It is not that AI simply becomes smarter on its own. We first identify in the field “how skilled people actually do it,” and then have the AI learn from that.</p><p><strong>Chad</strong> The marine environment has many gray areas where clear rules alone cannot determine the right answer. That is why I think it is difficult to build driver-assistance or autonomous systems without incorporating human experience at all.</p><p><strong>Tajima</strong> The fact that Honda does so much testing in the field itself may become a differentiator.</p><p><strong>Berker</strong> I think so. The knowledge and experience held by skilled operators will become even more important going forward.</p><h3><strong>Expanding Reality in the Virtual World</strong></h3><p><strong>Ogawa</strong> On the other hand, even if field data is important, we cannot gather every boat and every environment in the world.</p><p><strong>Berker</strong> There are an enormous number of boat types, and even the same boat behaves differently when the environment changes. It is unrealistic to cover all of that through testing with real boats alone. That is why, in addition to real-world data, synthetic data — artificially generated training data — and simulation are also important. We are now creating environments close to the real world and reproducing conditions that are difficult to collect in reality.</p><p><strong>Ogawa</strong> So you use the virtual world instead of the real one?</p><p><strong>Berker</strong> It is not one or the other. We combine both. Real-world data is extremely important, but we cannot wait to encounter every rare edge case in reality. Simulation allows us to expand the range of cases.</p><p><strong>Tajima</strong> Establish the baseline in the real world, then increase the variations virtually.</p><p><strong>Berker</strong> Exactly. And then verify them again in the real world.</p><p><strong>Chad</strong> Even something as simple as the color of the water varies greatly by location.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*eJWwFqWrp4YKcM7o" /></figure><p><strong>Berker</strong> Even within Florida. The water near our base is brownish, but as you head toward the inlet, it shifts to green and blue. Because the sea and river water mix, there is a lot of variation within a single area. With synthetic data, we can create an even wider range of environments — bright blue water, dark blue water, brown rivers, and more.</p><p><strong>Ogawa</strong> Put that way, the real and virtual worlds are not opposites. It is not “the field matters, so we do not need simulation.”</p><p><strong>Berker</strong> In fact, because we understand the field, we know what we need to simulate.</p><p><strong>Ogawa</strong> We take what we learn in the real world into the virtual world, expand the number of cases there, and then bring it back into the real world. There is a learning loop between the field and simulation.</p><p><strong>Berker</strong> We want to strengthen that loop even further.</p><h3><strong>Refining Intelligence Without Simply Adding More Sensors</strong></h3><p><strong>Ogawa</strong> I can see how much more complex the marine environment is than the road. That makes you want to add lots of sensors. But in reality, you cannot just keep adding them indefinitely, can you?</p><p><strong>Berker</strong> That is one of the difficult parts. We are aiming for a system that is practical while keeping cost as low as possible. That is why we are focusing especially on camera-based perception. Of course, we are also studying combinations with other sensors to improve safety and reliability.</p><p><strong>Tajima</strong> The answer is not simply to install a large, expensive system.</p><p><strong>Berker</strong> That’s right. Not every boat has marine radar, and the amount and location of space available for sensors varies tremendously from one boat to another.</p><p><strong>Chad</strong> That is very different from cars. Boat shapes vary enormously. Some have hardtops, some have T-tops, some have bimini tops, some have rails at the bow, and some have flat decks.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*mIWcqy4SpaXvB3bL" /></figure><p><strong>Ogawa</strong> Even the place where you can install a sensor is different.</p><p><strong>Chad</strong> In the U.S. alone, there are around 2,000 boat manufacturers, and each of them makes many different models.</p><p><strong>Berker</strong> With cars, whether it is an SUV or a sedan, there are still some basic similarities in overall form. Boats are not like that.</p><p><strong>Ogawa</strong> And Honda does not manufacture every boat hull either.</p><p><strong>Chad</strong> That is another important point. If a boat manufacturer had to change the mold for the hull just to install a Honda system, it would be difficult for them to adopt it. Easy installation matters.</p><p><strong>Ogawa</strong> So it is not enough for the technology to work as a research project.</p><p><strong>Chad</strong> It has to be something real boat manufacturers can actually use.</p><p><strong>Berker</strong> That is why the system needs to be as compact as possible and easy to adapt to many different boats.</p><p><strong>Tajima</strong> The physical environment is more complex than for cars, yet the hardware we install needs to be smaller, fewer in number, and lower in cost. So instead of simply adding equipment, we need to make the “brain” smarter.</p><p><strong>Berker</strong> That is exactly the challenge.</p><p><strong>Ogawa</strong> It is not easy (laughs). But that is precisely why I think it is worth researching.</p><h3><strong>Taking Intelligence Trained at Sea to the Next Mobility Domain</strong></h3><p><strong>Ogawa</strong> And from HGR’s perspective, there is one more thing I want us to consider: not keeping this technology confined to marine applications.</p><p><strong>Berker</strong> The basic architecture of intelligent systems has many common elements. You perceive the surroundings, understand the situation, decide how to move, and then control the system. Whether it is a boat, a car, a drone, or a humanoid robot, the fundamental flow is similar.</p><p><strong>Ogawa</strong> Which means it can connect with other Honda domains as well.</p><p><strong>Berker</strong> We are already collaborating technically with Honda’s autonomous driving teams in areas such as perception. There is a great deal of overlap in the underlying algorithms.</p><p><strong>Ogawa</strong> At the same time, marine applications have requirements of their own.</p><p><strong>Berker</strong> They do. Perception distance is one example. In automotive applications, even on highways, the system may need to look about 250 meters ahead. In marine applications, we need to be aware of what is 500 meters away — and in some cases as far as one kilometer.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*r-M25FG7Pqj7Vbwx" /></figure><p><strong>Tajima</strong> Because boats respond more slowly, you have to detect things earlier.</p><p><strong>Berker</strong> That is why we are also working on technologies for longer-range perception. What we develop there may not be necessary in other domains today, but it could become necessary in the future.</p><p><strong>Ogawa</strong> After spending the day here, I’ve also come to see some similarities between marine mobility and mobility in the air. Both respond relatively slowly, and neither the sea nor the mountains have clearly defined lanes like roads do. Both are also strongly affected by the surrounding flow, while risks and relationships with what is around them are not always explicitly defined.</p><p><strong>Berker</strong> In that sense, I think there is potential to transfer what we learn in marine development to other forms of mobility.</p><p><strong>Ogawa</strong> Before coming to Florida, I thought boats might be technically a little easier because they move more slowly than cars. Once I came here, I found the exact opposite.</p><p>The environment is always changing. Every boat is different. There are many situations where there is no single clear answer. That is precisely why we need the wisdom people have built up through years of experience.</p><p>How can we teach that wisdom to AI? If we cannot test everything in the real world, how do we combine real-world testing with the virtual world? And how do we create technology that can handle a complex environment while still being installed on many different kinds of boats?</p><p>These are not easy challenges. But precisely because this intelligence is being trained in such a difficult environment, there may be things we can bring back to other mobility domains.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*9KZK81PFjCqgEr1t" /></figure><p><strong>Chad</strong> First, we need to build something that is genuinely useful on the water.</p><p><strong>Berker</strong> And keep learning from real environments.</p><p><strong>Ogawa</strong> Yes. We need to connect human experience with AI and turn it into technology that is truly useful in the field. The research taking place on Florida waters is not simply about building a “self-driving boat.” How can machines understand a reality that cannot be fully predicted? How do we transfer human wisdom into technology? And how should people and machines work together?</p><p>The questions we are confronting at sea may eventually connect to land, air, and robotics. Coming all the way to Florida, I felt I could see a little more clearly the possibilities ahead.</p><p>■Related Articles</p><p><a href="https://medium.com/@HGRX/helping-people-get-things-done-all-the-way-hondas-spirit-of-monozukuri-carried-forward-by-599b65c03348">Helping People Get Things Done, All the Way — Honda’s Spirit of Monozukuri, Carried Forward by…</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f43e049dc1c8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Out on the Water, the Assumptions Change.—]]></title>
            <link>https://medium.com/@HGRX/out-on-the-water-the-assumptions-change-8226b40faaa1?source=rss-fd02d7b9dda5------2</link>
            <guid isPermaLink="false">https://medium.com/p/8226b40faaa1</guid>
            <dc:creator><![CDATA[HGR / Honda R&D Center of Excellence]]></dc:creator>
            <pubDate>Wed, 07 Oct 2026 03:01:04 GMT</pubDate>
            <atom:updated>2026-10-07T03:01:04.978Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>Out on the Water, the Assumptions Change.— The “Reality” of Marine Mobility Honda Is Exploring in Florida [Part 1]</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aykTTdqZEO1GpRHWSImhzw.jpeg" /></figure><p>Hello everyone. I’m Atsushi Ogawa, Chief Operating Officer of Honda R&amp;D Research Center of Excellence (HGR).</p><p><a href="https://medium.com/@HGRX/helping-people-get-things-done-all-the-way-hondas-spirit-of-monozukuri-carried-forward-by-599b65c03348">Together with Tomoharu Tajima, Head of the Solution Development Office,</a> I visited Honda’s marine R&amp;D facility in Florida, USA. Florida is one of the world’s leading marine markets, where boats of all shapes and sizes travel across the ocean, rivers, and waterways, and countless people enjoy life on the water.</p><p>My purpose was to see Honda’s intelligent systems research there with my own eyes. I got on a boat, headed out onto the water, and experienced the entire sequence all the way through to loading the boat back onto its trailer. Along the way, I spoke with Chad Nere, who oversees the Florida base, and software engineer Berker Logoglu.</p><p>Once we were there, we began to notice many things that simply could not be understood by talking about the technology alone. Why does Honda conduct this research here in Florida? To begin, I’d like to share what I learned from spending a full day out on the water.</p><h3><strong>Turning the Idea That “Boats Are Slow” on Its Head</strong></h3><p><strong>Ogawa</strong> Thank you for giving us the opportunity to take the boat out and experience boating-assistance technologies such as autonomous navigation and trailer loading. After spending a full day on the water, what changed my perspective the most was the impact of the environment. With cars, rain and wind are simply “conditions.” But with boats, the wind and currents themselves become “variables” that actually move the boat.</p><p><strong>Chad</strong> Yes. It feels almost as if the road itself were constantly moving.</p><p><strong>Ogawa</strong> Even the inlet we passed through today (Note 1) had a stronger current than it looked like from the surface.</p><p><strong>Chad</strong> Sebastian Inlet has a very strong tidal current. And the way it flows changes depending on the time of day. Then you add wind on top of that.</p><p>Note 1: An inlet is a narrow body of water where the sea or a lake extends inland. Sebastian Inlet is an artificial inlet in Florida and is known around the world as a major destination for surfing and fishing.</p><p><strong>Tajima</strong> I heard today was relatively calm, but watching the boat being loaded onto the trailer, it clearly wasn’t easy. Even a little wind was enough to move the boat.</p><p><strong>Berker</strong> That is why real-world testing is so important in marine development. On the water in particular, it is difficult to reproduce everything through simulation alone.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*6hcRAHSzdEtFb7kV" /><figcaption>(From left) Berker Logoglu, Atsushi Ogawa, Chad Nere</figcaption></figure><p><strong>Ogawa</strong> Is that another difference from autonomous driving in cars?</p><p><strong>Berker</strong> Yes. Cars are affected by weather too, of course. But the road itself does not move because of wind or current.</p><p><strong>Chad</strong> On the water, the environment is constantly exerting forces on you.</p><p><strong>Ogawa</strong> To be honest, before coming here, I thought, “Boats are slower than cars, so maybe automation is a little easier.” But it is actually the opposite.</p><p><strong>Berker</strong> Boats also take time to respond. While they are responding, they are still being affected by wind and current, which makes control difficult.</p><p><strong>Tajima</strong> So being slow does not mean you have more margin. It means the boat keeps being pushed around from the outside during that slower response.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*xPWaCYnEiO84dPfX" /><figcaption>Chad Nere,Tomoharu Tajima</figcaption></figure><p><strong>Berker</strong> Exactly. So once we develop an algorithm in the office, we come here and test it in the real world. We develop, test, understand what was different, and revise it. Then we repeat the cycle. The result changes when the weather or current changes, so we develop while continuously learning from the real world.</p><p><strong>Ogawa</strong> After experiencing it today, I really understand the weight behind the words “real world.” In the office, we tend to separate things into the boat, the trailer, the sensors, and the control system. But in the field, all of those things are moving at the same time.</p><p><strong>Berker</strong> That is both the difficulty and the value of testing in the field.</p><h3><strong>Bringing the “Reality” of the Market into Development</strong></h3><p><strong>Ogawa</strong> Even just being out here today, we saw an incredible variety of boats — large boats, pontoons, and smaller boats.</p><p><strong>Chad</strong> Florida has a huge variety of boats. That is one reason this location is so well suited to testing.</p><p><strong>Ogawa</strong> How did Honda come to establish a base here in the first place?</p><p><strong>Chad</strong> It started with a search for a marine product testing site. At the time, I was based in North Carolina, where Honda R&amp;D America had a facility, and I traveled around with one Japanese colleague to evaluate different locations. We looked not only along the East Coast, but the West Coast as well. We literally drove around to find the best place.</p><p><strong>Ogawa</strong> What criteria did you use to choose the location?</p><p><strong>Chad</strong> Obviously, it had to be on the water. But we also needed good airport access and the right surrounding environment. Most important of all was the testing environment. Here, we have the Intracoastal Waterway (Note 2), and we can reach the Atlantic Ocean through the inlet. Even when the ocean is rough, we can still test on the protected inland side.</p><p>Note 2: The Intracoastal Waterway is a vast network of inland waterways extending along the Atlantic and Gulf coasts of the United States for about 3,000 miles (approximately 4,800 km).</p><p><strong>Ogawa</strong> So one location gives you access to a wide range of different conditions.</p><p><strong>Chad</strong> Another major advantage is that we can test on the kinds of boats people actually use in the United States.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*mmPigQDjvwYPJdft" /></figure><p><strong>Ogawa</strong> Was the facility always like it is today?</p><p><strong>Chad</strong> Not at all (laughs). At first, we just rented a small office and garage a little south of Sebastian. It was so cramped that we did most of the work outside. We did not even have an air-conditioned work area.</p><p><strong>Ogawa</strong> In Florida, that sounds rough (laughs).</p><p><strong>Chad</strong> It was hot (laughs). But around that time, we were developing a lot of outboard motors.</p><p><strong>Ogawa</strong> And you were actually installing them on American boats here.</p><p><strong>Chad</strong> Of course we also test in Japan. But there are things you only discover when you use the products on boats that are especially common in the U.S., such as pontoons and aluminum fishing boats.</p><p><strong>Ogawa</strong> Did you actually find problems?</p><p><strong>Chad</strong> We did — things related to timing, cooling, and so on. There were cases where testing here allowed us to make improvements before mass production.</p><p><strong>Ogawa</strong> So instead of discovering issues after a product reaches the market, you can find them in the field beforehand.</p><p><strong>Chad</strong> Testing here has ultimately helped us make better products.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*w9d2gvbzOu0kbmgU" /></figure><p><strong>Ogawa</strong> Seen that way, Florida has long been a place where you “feed reality back into development.”</p><p><strong>Chad</strong> And Florida itself is also a huge marine market. Around one million boats are registered here. It is not just large offshore fishing boats. There are many smaller boats, and the pontoon market is growing as well. There are also many boats using high-output outboards, or multiple engines — two, three, even four.</p><p><strong>Tajima</strong> So many different sizes and use cases are concentrated in one region.</p><p><strong>Chad</strong> That’s right. If you are developing products for the marine industry, it is an extremely important market.</p><p><strong>Ogawa</strong> In the era of engine development, this was a place to confirm whether a product would truly work on American boats. But with intelligent systems, it seems like you have to go one step further.</p><p><strong>Berker</strong> I agree. With intelligent systems, we need to understand not only the performance of the boat, but also the surrounding environment, how other boats move, and — at an even deeper level — the culture of boating itself.</p><h3><strong>Looking Beyond the Boat to the People Who Use It</strong></h3><p><strong>Ogawa</strong> I felt that very strongly today. It is not just about what happens on the water.</p><p><strong>Chad</strong> Everything is part of boating life — from launching the boat into the water to coming back and loading it onto the trailer.</p><p><strong>Ogawa</strong> If you are only looking at documents in Japan, “loading a boat onto a trailer” can sound like a single operation. But when you see it in person, it is completely different.</p><p><strong>Chad</strong> Coming to the actual site matters. We can explain things in online meetings and share them in documents. But there are things those methods alone cannot convey.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*2T_33HlsTqIG2m_T" /><figcaption>In the United States, it is common for owners to transport boats by trailer and launch and retrieve them themselves.</figcaption></figure><p><strong>Ogawa</strong> Is that true even when engineers come from Japan?</p><p><strong>Chad</strong> Sometimes they realize, “This is different from what I imagined.” The scenario they had in mind may differ slightly from how the boat is actually used.</p><p><strong>Tajima</strong> And if they discover that difference early, they can feed it back into development.</p><p><strong>Chad</strong> Exactly. We can revise the design or the algorithm if needed. And when engineers who have actually experienced the site return to Japan, they can pass that understanding on to other team members.</p><p><strong>Ogawa</strong> There is a difference between reading a document and saying, “This is what I actually felt when I was there.”</p><p><strong>Chad</strong> I think there is. Of course, we explain things as thoroughly as we can. But that can never replace actually coming here and experiencing the environment and culture for yourself.</p><p><strong>Ogawa</strong> I experienced that myself, especially at the boat ramp. Even when all you are doing is loading the boat onto a trailer, there is already another person waiting behind you.</p><p><strong>Chad</strong> The pressure is even greater on weekends. There can be a line the whole time. And by evening, everyone is hot, tired, and hungry (laughs). It is especially difficult for people who are not used to it. Not everyone goes boating every week.</p><p><strong>Tajima</strong> If you cannot get the boat onto the trailer smoothly, you start worrying about the line behind you.</p><p><strong>Chad</strong> That is why technology could make a big difference. Make the operation easier. Make it safer. Reduce the pressure on the captain.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*H2o3XQWIY_dGdxqy" /><figcaption>Honda is developing technology that can automatically guide a boat from the water back onto its trailer.</figcaption></figure><p><strong>Tajima</strong> If you look only at the technical challenge, you can summarize it in one phrase: “automated trailer loading.” But when you are actually here, you also see the people waiting behind the user.</p><p><strong>Chad</strong> You understand why the function has value.</p><p><strong>Tajima</strong> It is not “we automate it because we can.” The starting point is a situation where someone is struggling.</p><p><strong>Berker</strong> That is important from the software side as well. If we do not understand the situations in which people struggle, we cannot decide what kind of behavior the system needs.</p><p><strong>Ogawa</strong> If you looked only at the technology, you might arrive at the wrong optimal answer.</p><p><strong>Berker</strong> Absolutely. When it comes to how the boat should behave, we receive a lot of feedback from Chad and the Florida team.</p><p><strong>Ogawa</strong> So this is not simply a place where finished technology is brought for final confirmation.</p><p><strong>Berker</strong> Right. Development itself progresses together with the field.</p><h3><strong>Automation in Service of Enjoyment</strong></h3><p><strong>Ogawa</strong> After seeing the field this closely, I also find myself asking a more fundamental question: “What are intelligent systems for?” Is Honda aiming for a boat that people never have to operate at all?</p><p><strong>Chad</strong> At least when it comes to recreational boats in the U.S., I do not think that is necessarily the goal. Most people are not buying a boat simply to get from one place to another.</p><p><strong>Ogawa</strong> They buy one to enjoy it.</p><p><strong>Chad</strong> Yes. And operating the boat yourself is part of the enjoyment. Personally, I do not want a system that does absolutely everything for me. But I do want a system that supports me.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*SYu1S3NSu5JPJRKL" /></figure><p><strong>Ogawa</strong> In what situations?</p><p><strong>Chad</strong> I have four children, so when my family is on the boat, there are all kinds of things happening besides operating the boat. If I look away for a moment, it could alert me to another boat I have not noticed, or identify a hazard close to the surface. Something like that would be extremely valuable.</p><p><strong>Tajima</strong> So rather than “driving for you,” it “covers what you might miss.”</p><p><strong>Chad</strong> A system that supports navigation and safety.</p><p><strong>Ogawa</strong> The boat-ramp example is the same. Instead of handing everything over to the machine, it helps with the most difficult moments.</p><p><strong>Chad</strong> If that allows customers to enjoy boating more confidently and safely, then I think it has real value.</p><p><strong>Ogawa</strong> I see. When we hear the term “intelligent systems,” we tend to think in terms of “how far can we automate?” But from the field, the question looks a little different.</p><p><strong>Berker</strong> I agree. To decide what the technology should be for, it is important to understand the actual users and the environment.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*8lLDfD6AqtJ5i8Fv" /></figure><p><strong>Tajima</strong> You launch the boat, spend time on the water, and then come back. Where in that journey do people feel anxious? Where could technology provide support and make the experience more enjoyable?</p><p><strong>Chad</strong> That is what matters.</p><p><strong>Ogawa</strong> I feel like I understand much more clearly why it mattered to come to Florida. How does the water move? How are boats actually used? What do people struggle with, and what do they enjoy? Only after seeing those things can we ask, “Then what should the technology do?”</p><p><strong>Berker</strong> On the water, many factors act at the same time — wind, current, waves, the condition of the boat, and more. Even so, an experienced captain is not operating the boat by calculating everything as a mathematical equation.</p><p><strong>Tajima</strong> Chad, you are not calculating every time that “the wind is like this, so I should use exactly this much throttle,” are you?</p><p><strong>Chad</strong> No. A lot of it is a “feel” that comes from experience.</p><p><strong>Ogawa</strong> That is where things get really interesting. How does AI learn that “feel”? In Part 2, I would like to dig into that question. How do we put the intuition people acquire through experience into algorithms? How do we cover conditions that cannot be tested on the real ocean alone? And where can the intelligence trained on Florida waters lead next? That is what I want to explore.</p><p>■Related Article</p><p><a href="https://medium.com/@HGRX/helping-people-get-things-done-all-the-way-hondas-spirit-of-monozukuri-carried-forward-by-599b65c03348">Helping People Get Things Done, All the Way — Honda’s Spirit of Monozukuri, Carried Forward by…</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8226b40faaa1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Helping People Get Things Done, All the Way — Honda’s Spirit of Monozukuri, Carried Forward by…]]></title>
            <link>https://medium.com/@HGRX/helping-people-get-things-done-all-the-way-hondas-spirit-of-monozukuri-carried-forward-by-599b65c03348?source=rss-fd02d7b9dda5------2</link>
            <guid isPermaLink="false">https://medium.com/p/599b65c03348</guid>
            <dc:creator><![CDATA[HGR / Honda R&D Center of Excellence]]></dc:creator>
            <pubDate>Wed, 30 Sep 2026 04:01:02 GMT</pubDate>
            <atom:updated>2026-09-30T04:01:02.206Z</atom:updated>
            <content:encoded><![CDATA[<h3>Helping People Get Things Done, All the Way — Honda’s Spirit of Monozukuri, Carried Forward by Solution Development</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AlNe-lrTgzNvhPZRlPPtXQ.jpeg" /></figure><p>Hello everyone. I’m Atsushi Ogawa, Chief Operating Officer of Honda R&amp;D Research Center of Excellence (HGR).</p><p>Within HGR, we have the Solution Development Division, or SD, whose roots include areas such as power products and marine products. What makes this field distinctive is the sheer breadth of its lineup — from general-purpose engines to generators, lawn mowers, snow blowers, and outboard motors for large boats — all products that are closely connected to people’s everyday lives.</p><p>Naturally, the research is just as diverse: analyzing soil, cutting grass, pumping water, propelling boats. The things they work with are quite different from automobiles, but whenever I heard about what they were doing, I always thought, “That sounds interesting.”</p><p>This time, I sat down with Tomoharu Tajima, Head of Solution Development, to talk about what this field has carried with it over the years — and where it might go next.</p><h3>Taking Technology All the Way to the People Who Use It</h3><p><strong>Ogawa:</strong> We’ve had some interaction with your area over the years, and I’ve always found the work interesting. You were using particle methods to analyze soil, for example, or simulating how snow gets thrown by a snow blower.</p><p>Then there was the pump impeller — the rotating component inside the pump. You analyzed it yourselves, designed it yourselves, and then said, “In an internal comparison with the conventional design under specified evaluation conditions, we improved efficiency by 30%.” In automotive development, we used to celebrate an improvement of just a few percent, so I remember thinking, “What do you mean, 30%?”</p><p><strong>Tajima:</strong> Our organization was formed by several different streams coming together. One of our major roots is power products, and we’ve since been joined by people who have worked in intelligent technologies and software. On one side, you have people who have spent their careers working very close to the physical world — with things like soil and grass. On the other, you have people with expertise in intelligence and software. In some ways, they’re almost complete opposites.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*x2X5Ll27aa8JyN9h" /><figcaption><em>From left: Atsushi Ogawa and Tomoharu Tajima</em></figcaption></figure><p><strong>Ogawa:</strong> Even the way you approached analysis was distinctive.</p><p><strong>Tajima:</strong> We worked closely with the automobile and motorcycle teams, so we would take technologies being used there and ask ourselves, “Could we use this to analyze soil?” or “Could we use it to analyze grass?”</p><p>But the person doing the analysis doesn’t just stop at the analysis. In the end, we all operate the actual machine together and make sure that we really achieved what we intended. There’s a culture here of working together all the way from the calculations through to final verification.</p><p><strong>Ogawa:</strong> Another thing is how close you are to the customer. With B2B products like GX engines (*1), you have application engineers who work directly with OEM customers. At first, I assumed that because it was B2B, maybe you didn’t think that much about the people who would ultimately use the products. But I was completely wrong.</p><p><em>1 GX engines: A series of Honda general-purpose engines used to power industrial and construction equipment.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/0*Lo-0PHGiAIgd_qd5.png" /><figcaption><em>Honda iGX390: A high-performance, air-cooled, four-stroke, single-cylinder OHV gasoline engine equipped with an electronic governor.</em></figcaption></figure><p><strong>Tajima:</strong> General-purpose engines are used in all kinds of environments, in all kinds of countries. Even with the same pump, what you need changes depending on whether it’s pumping muddy water or supplying clean water to a rice field. Sometimes we’ll actually buy the complete machine, replace its engine with ours, and test it ourselves to see whether it really works the way it should.</p><p><strong>Ogawa:</strong> You understand the OEM, but you’re also looking beyond them to the end user. You follow the whole process — from simulation and design to calibration and where the product will ultimately be sold.</p><p>Automotive organizations are large and highly specialized, so I remember asking myself, “Are we really seeing this much of the whole picture?”</p><p><strong>Tajima:</strong> There’s a phrase that has always remained very much alive within our organization: <strong>“Helping People Get Things Done.”</strong></p><p>Our business may not be particularly large in scale, but <strong>we have brought Honda to people in countries where motorcycles and automobiles may not yet be within reach.</strong> Our products can make their daily lives easier, and the value they create can help people take the next step forward. I think the people in power products genuinely believe that profit alone is not the ultimate goal.</p><p><strong>Ogawa:</strong> Every time I hear that, I think it’s wonderful. Before asking whether something makes sense as a business, there is first this desire to help people get things done.</p><p><strong>Tajima:</strong> I think part of that comes from the fact that we’re not a very large organization. There aren’t many people who say, “This is my specialty, and this is all I do.” In fact, we couldn’t work that way even if we wanted to. You have to look into the areas next to your own and simply focus on making the best possible product.</p><p>People sometimes tell us, “Your organization has preserved something of the old Honda spirit.”</p><p><strong>Ogawa:</strong> Before SD joined the Advanced Technology Research Institute, I also wondered how an area rooted in long-established, hardware-centered businesses would adapt to the age of software.</p><p>But surprisingly, you were among the fastest to adapt. You develop algorithms, write the software, and even implement it in the ECU yourselves. You don’t simply hand the work off to someone else. That makes you faster and reduces rework. I remember thinking, “Ah, this is part of their culture.”</p><p><strong>Tajima:</strong> And we also have to think about the business beyond the technology.</p><p><strong>Ogawa:</strong> Exactly. Of course you have to create value for customers, but you also think about everything beyond that — how to actually get that value into their hands. I think there’s a lot we can learn from that when we consider new businesses.</p><h3>Turning Differences into Connections</h3><p><strong>Tajima:</strong> One of the biggest changes since joining HGR has been regaining the perspective of looking at technology from a clean slate.</p><p>We’ve always worked with limited resources, so it was easy to fall into the mindset of, “We have to find a way to make this work with what we have.” We would simply keep pushing until we reached the goal. I think that tendency was quite strong.</p><p><strong>Ogawa:</strong> The people in the Intelligence Domain who research things like automated driving approach things a little differently.</p><p><strong>Tajima:</strong> They go all the way back to the question, “What is actually the best solution?” They look at the fundamental principles and ask whether there might be a better way. I realized that was something we needed to add to our own approach as well.</p><p><strong>Ogawa:</strong> Has that actually changed the technology?</p><p><strong>Tajima:</strong> It has. We had already been developing our own environmental perception technology for marine navigation assistance before the organizations came together. But when we simply kept building on what we already understood and could already do, there were areas where accuracy didn’t improve as much as we expected.</p><p>We began learning from the people working in intelligence technologies, and some of them joined the work directly. That helped us make significant progress.</p><p>At sea, for example, you have to distinguish the obstacles you truly need to avoid while filtering out waves and foam. Being able to discuss challenges like that together has made a big difference.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*tp8fr1TJDesAClJD" /><figcaption><em>Operating a research boat in Tokyo Bay</em></figcaption></figure><p><strong>Ogawa:</strong> To be honest, at first I didn’t think there would necessarily be that much synergy. Even within Honda, when someone asks, “Can motorcycles and automobiles create synergies?” I don’t think it’s as simple as putting people from the same company together and expecting something new to happen. (laughs)</p><p><strong>Tajima:</strong> If synergy itself becomes the objective, that doesn’t feel right either.</p><p><strong>Ogawa:</strong> Exactly. Even if we use the same words — automated driving or autonomous control — the customers are different, the value we provide is different, the environmental conditions and boundary conditions are different, and the fields where the technology is used are different.</p><p>So I don’t think we need to rush to call everything “synergy.” It may actually be better to start from the assumption that there isn’t any.</p><p>Then, when someone runs into a problem, they can talk to someone else. And sometimes you realize, “Oh, this is something we can work on together.” I think that’s about the right level.</p><p><strong>Tajima:</strong> Marine technology seems to attract a lot of interest from people around us. You included, quite a few of the domain leaders have even gone and gotten their boat licenses.</p><p><strong>Ogawa:</strong> And it’s not as though we’re making a deliberate effort to do that. People are simply curious. Boats look interesting, so they want to try one. And once they do, the work naturally starts to feel closer.</p><p>We don’t even need to invoke the idea of Honda’s “Three Realities” principle. The result may look similar, but I think the starting point is slightly different.</p><p>It begins with whether you’re genuinely interested.</p><h3>From One Machine to the Next “General-Purpose” Platform</h3><p><strong>Tajima:</strong> Intelligence technologies now account for a much larger share of what we do. Autonomous lawn mowers are one example, but outdoor automation presents a different kind of difficulty from automobiles.</p><p>You’re dealing with nature, so there are many uncertainties. And simply moving from one place to another isn’t enough. If it’s a lawn mower, it has to actually finish cutting the grass. At the same time, it has to be safe.</p><p>So the challenge is how to improve the quality of the work while ensuring safety. We’re still exploring how best to balance those two things.</p><p><strong>Ogawa:</strong> It reminds me a little of how I felt when I moved from HondaJet to F1.</p><p>Aviation has a long history. The rules are well established, and there are textbooks for almost everything. Competition takes place within that framework.</p><p>When I moved to F1, almost none of that existed. The attitude was basically, “As long as you stay within the regulations, you can do whatever you want.”</p><p>Outdoor automation feels somewhat similar to me. There still isn’t a fully established framework. But the requirement to ensure safety doesn’t change. There’s a certain amount of freedom in how you get there, and that’s what makes it both difficult and fascinating.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*8u34AT40UGHHo8H6" /><figcaption><em>The ProZision series of electric riding lawn mowers, unveiled in October 2025</em></figcaption></figure><p><strong>Tajima:</strong> When you’re dealing with nature, you can’t operate solely according to predetermined rules.</p><p><strong>Ogawa:</strong> Unlike cars, there aren’t even lanes. At sea, you have waves. You have sunlight. You even have foam created by your own boat. Unless you’re actually working in that environment, it’s hard to understand what makes it so difficult.</p><p><strong>Tajima:</strong> In marine applications, I think the fact that we already had outboard motors has also been important. No matter how well you can sense the environment or plan an action, if the boat doesn’t respond the way you expect in the end, you can’t automate its operation.</p><p>So we build new perception technologies on top of the control technologies we already have. I think that approach is very characteristic of us.</p><p><strong>Ogawa:</strong> You don’t have to throw away everything you already have just because a new technology comes along.</p><p><strong>Tajima:</strong> Looking a little further ahead, I think <strong>what has happened inside factories is going to spread outdoors next.</strong></p><p>Automation has evolved from fixed systems such as conveyor belts to machines like AMRs — autonomous mobile robots — that can move freely. In controlled indoor environments, automation has already advanced considerably. I think the next step is for that same shift to reach outdoor work.</p><p><strong>Ogawa:</strong> And lawn mowers are the first step.</p><p><strong>Tajima:</strong> With lawn mowers, we’re learning how to perceive off-road environments and how to operate within them. But those challenges aren’t unique to lawn mowers. Many different kinds of outdoor work machines should face similar issues.</p><p>That’s why we don’t want the automated driving and autonomous control — or AD — technology we develop here to remain limited to lawn mowers.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*sQ5XKkUDse-3whVl" /></figure><p><strong>Ogawa:</strong> It’s the same kind of thinking as a general-purpose engine.</p><p><strong>Tajima:</strong> Exactly. With power products, we took the engine itself and made it available for use in all kinds of customers’ products.</p><p>In the same way, I wonder whether we could create something like a <strong>general-purpose AD system</strong> as a new Honda core technology that could take the place of the general-purpose engine. If it could be used in many different environments, it might even become one of our next pillars. I’d like to take it that far.</p><p><strong>Ogawa:</strong> And technically, outdoor automated driving is tackling some extremely difficult challenges.</p><p>Even with automated driving for cars, much of what is possible today is still centered on limited environments such as highways. Extending that into urban areas has taken time. But SD is trying from the outset to enter environments where people and autonomous machines coexist.</p><h3>Turning Failure into the Next Starting Point</h3><p><strong>Ogawa:</strong> I see SD as something of a model for how to approach new businesses.</p><p>When you think about it, you’ve launched all kinds of new products over the years. You found your own customers, built supply chains, and expanded around the world. As we discussed earlier, you don’t just think about the upstream part of the process. You think all the way downstream, until the product actually reaches the customer.</p><p><strong>Tajima:</strong> Of course, not everything has worked out. One of the airports that is now collaborating with us on lawn mower R&amp;D is actually a place where a different research project once didn’t go well.</p><p>Even so, the customer we worked with at the time came back to us and said, “Let’s work together again.” We’re very grateful for that.</p><p><strong>Ogawa:</strong> I think that’s what new business is really like.</p><p>It isn’t a matter of assembling everything you need and then moving neatly through the steps in order. <strong>You try something, and it doesn’t work. So you go back and try again. Then you go back again. You find your way forward through that process.</strong></p><p><strong>Tajima:</strong> In reality, it’s rarely a clean process.</p><p><strong>Ogawa:</strong> Everyone tries too hard to make it look clean. In reality, it’s much messier than that.</p><p>But you only understand that once you’ve been through it. People who have experienced failure gradually develop a sense for things: “This part is dangerous,” or “This is where we can push ahead.” You can’t develop that instinct without actually doing the work.</p><p><strong>Tajima:</strong> When you take a new business into the market, there are also many challenges that have nothing to do with the technology itself.</p><p>First, finding customers is difficult. And even if you find them, can you mass-produce the product? Will suppliers be willing and able to support you? How do you build the supply chain?</p><p>We’ve struggled with many of those things ourselves. I think other domains will face the same challenges when they start new businesses, so I’d like us to get more deeply involved and share what we’ve learned.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*pVH_htOddVMvRI-0" /></figure><p><strong>Ogawa:</strong> Exactly. You haven’t just developed technology. You’ve taken it out into the world as a business and followed it all the way through until it reaches the people it is meant to serve.</p><p>The fact that you have that experience is itself one of this field’s strengths. And now, intelligence and robotics technologies are being added to it.</p><p><strong>Tajima:</strong> We want to keep taking on new challenges while making the most of what we’ve built over the years.</p><p><strong>Ogawa:</strong> It may not always be obvious from the outside, but <strong>I think SD has actually served as a starting point for many of Honda’s businesses and technologies.</strong></p><p>Let’s build on everything you’ve accumulated so far — and use it to create Honda’s next businesses.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=599b65c03348" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What Took Flight Was Confidence.]]></title>
            <link>https://medium.com/@HGRX/what-took-flight-was-confidence-7c1d931a81d3?source=rss-fd02d7b9dda5------2</link>
            <guid isPermaLink="false">https://medium.com/p/7c1d931a81d3</guid>
            <dc:creator><![CDATA[HGR / Honda R&D Center of Excellence]]></dc:creator>
            <pubDate>Wed, 23 Sep 2026 10:01:03 GMT</pubDate>
            <atom:updated>2026-09-23T10:01:03.414Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>What Took Flight Was Confidence. - The Technology and Decisions Behind the First Flight of Honda’s Full-Scale eVTOL Model</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ganih-hfcbiD-Pdw2WydXA.jpeg" /></figure><p>Hello everyone. I’m Koya Kuwamura, and I work on eVTOL research and development.</p><p>This April, the first full-scale model we have been developing, known as the F1 aircraft, successfully completed its maiden flight. I served as the test conductor, monitoring data transmitted from the aircraft in real time and confirming the test was progressing as planned. Sitting next to me was Test Director Drew Paolicelli, who monitored the aircraft’s overall condition and managed the risks associated with the flight.</p><p>At the moment of first flight, we were not looking up at the sky. We watched the data displayed across the telemetry room, following the aircraft until it flew safely and returned to the ground.</p><p>How did we take an aircraft that “should fly in theory” and bring it to the point where we could confidently say, “It is ready to fly”?</p><p>Drew and I look back on the journey to that day.</p><h3><strong>The 30 Seconds After Landing</strong></h3><p><strong>Kuwamura:</strong> During the first flight, the pilot and flight test engineer were at the ground station, while we were inside the telemetry room in the hangar. We were sitting right next to each other.</p><p><strong>Drew:</strong> There were about 20 engineers in the telemetry room, each monitoring different parameters during the flight.</p><p>As test conductor, Koya checked incoming data for problems and made sure we met the test objectives. As test director, I was responsible for understanding the aircraft’s overall condition and managing the flight’s overall risk.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*NRPjj8aID1mBdxo6" /><figcaption>(From left) Drew Paolicelli and Koya Kuwamura</figcaption></figure><p><strong>Kuwamura:</strong> Even when the aircraft lifted off, the room stayed remarkably quiet.</p><p><strong>Drew:</strong> I think everyone was extremely tense. We had planned everything very carefully, with safety as our highest priority. But even after all that preparation, this was still the first time the aircraft had ever actually flown.</p><p><strong>Kuwamura:</strong> For me, the longest moment actually came after it landed.</p><p><strong>Drew:</strong> We deliberately chose not to watch video of the aircraft during the flight. We wanted to stay focused on the data rather than have our attention drawn to the visuals.</p><p>That meant the data alone could not immediately tell us whether the aircraft had completely settled on the ground. It was probably only about 30 seconds, but it felt as though everyone in the room was holding their breath.</p><p><strong>Kuwamura:</strong> And the moment we confirmed everything was safe, the music from <em>Top Gun</em> suddenly started playing.</p><p><strong>Drew:</strong> That was me. That was when the applause and cheering finally broke out.</p><p><strong>Kuwamura:</strong> We had all been concentrating so intensely that when the music suddenly came on, I couldn’t help laughing.</p><p><strong>Drew:</strong> Everyone started high-fiving, and we gathered around one screen. During the flight, we had all been focused on monitoring the data, so none of us had actually watched the aircraft fly. We replayed the video over and over. I think we must have watched it about 20 times.</p><iframe src="https://cdn.embedly.com/widgets/media.html?type=text%2Fhtml&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;schema=twitter&amp;url=https%3A//x.com/HGRX_jp/status/2072455208873971770&amp;image=https%3A//i.embed.ly/1/image%3Furl%3Dhttps%253A%252F%252Fpbs.twimg.com%252Famplify_video_thumb%252F2072312499870076928%252Fimg%252FMp2mYCeZ36Pjpxu6%253Fformat%253Dwebp%2526name%253Dlarge%26key%3Da19fcc184b9711e1b4764040d3dc5c07" width="500" height="281" frameborder="0" scrolling="no"><a href="https://medium.com/media/e87a80732b27ccfb7dd06770327ba709/href">https://medium.com/media/e87a80732b27ccfb7dd06770327ba709/href</a></iframe><p><strong>Kuwamura:</strong> But pretty soon after that, we were already opening up the data again.</p><p><strong>Drew:</strong> Exactly. Once we had celebrated, our attention immediately shifted to, “What do we do next?” We already had another flight scheduled for later that same week.</p><h3><strong>Turning Predictions into Confidence</strong></h3><p><strong>Kuwamura:</strong> One of the biggest achievements of this first flight was that, for the first time, we completed the entire development cycle with a full-scale aircraft: building it, integrating the systems, testing it, and ultimately flying it.</p><p>We had conducted extensive validation using subscale models. But this time, we gained confidence not only in the aircraft itself, but also in our development process and the tools we used. We came away feeling, “Yes, we can move forward this way.”</p><p><strong>Drew:</strong> For me, what mattered most was proving that we could execute as a team.Understanding what needs to happen technically is not enough to make an aircraft fly. Many people have to work together for months, then bring everything together at the end to take the aircraft from liftoff through landing.</p><p>The first flight proved more than the fact that the aircraft could fly. It also proved that this team could take one aircraft through a complete flight together.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*DeBFkQdxp-74QHnl" /></figure><p><strong>Kuwamura:</strong> Comparing the subscale and full-scale programs, what stood out to me most was the difference in the number of people involved.</p><p>With a subscale model, a small team can move quickly through relatively simple communication. But once you move to full scale, the systems become more complex and specialists from many different disciplines become involved.</p><p>No one person can look only at their own system. Everyone has to look at the same aircraft and understand the same risks.</p><p><strong>Drew:</strong> My experience was almost the opposite. I came from more highly structured flight-test programs for large fixed-wing aircraft, so I initially had to adjust to how subscale testing worked.</p><p>Subscale testing is about developing technology quickly and testing it quickly. Once you move to full scale, the expectations around test planning and risk management change dramatically.</p><p>It is not simply that the aircraft gets bigger. The responsibility that comes with flying it changes, too.</p><p><strong>Kuwamura:</strong> From a design standpoint, we were confident that it would fly.</p><p><strong>Drew:</strong> On paper, we believed it should fly. But until it actually takes off and lands again, you never truly know. You can only gain a level of confidence when the systems you designed physically exist and are tested in the real environment.</p><p>We had done everything we could to prepare. At that point, the only thing left was to have confidence in our work, fly the aircraft, and confirm it for ourselves.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*hDLb_zhGkzwPJB5D" /></figure><p><strong>Kuwamura:</strong> One of the things we especially wanted to observe during the first flight was how the aircraft behaved close to the ground.</p><p>We can predict that behavior through simulation, but simulation can never reproduce reality perfectly. In particular, the dynamic effects of ground effect* can only be fully understood when the aircraft is actually moving freely through the air.</p><p>At the same time, we had a fairly clear understanding of what we did not know.</p><p>Earlier in my career, when I was developing eVTOL aircraft at a startup, we often flew first and learned from what happened.</p><p>This time, we had accumulated much more research and development knowledge. We understood the uncertainties and could fly specifically to validate them.</p><p>That was a major difference.</p><p><em>*Ground effect: A phenomenon that occurs when an aircraft flies close to the ground, where interaction between the ground and the airflow generated by the rotors or other lifting surfaces affects the aircraft’s lift and behavior.</em></p><p><strong>Drew:</strong> You could feel the tension in the telemetry room rise as the aircraft came closer to the ground.</p><p>But the actual behavior was very close to what our models had predicted. Confirming on the very first flight that our models accurately predicted the real aircraft’s behavior was a huge reassurance.</p><p>It also gave us greater confidence to proceed to the next flight.</p><h3><strong>We Don’t Decide “Go” Just Once</strong></h3><p><strong>Kuwamura:</strong> Our testing for the first flight did not begin with someone handing us a completed aircraft.</p><p>As assembly progressed, each engineer integrated their respective systems and verified that they were operating correctly. The test team was already working alongside the design engineers from that stage.</p><p><strong>Drew:</strong> In a more conventional aircraft development program, the test team and design team can be more clearly separated.</p><p>The test team may receive the completed aircraft and begin pre-flight ground testing.</p><p>This time, however, design and test worked together much earlier. By getting involved in system-level integration, the test team also developed a much deeper understanding of the aircraft.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*QWNnS6osr0qelnD0" /><figcaption><em>F1 aircraft before flight testing</em></figcaption></figure><p><strong>Kuwamura:</strong> At the beginning, there were really only a few of us.</p><p><strong>Drew:</strong> It was Koya, me, and just a handful of design engineers.</p><p>We had meetings every morning and evening. As assembly progressed, more and more people gradually joined.</p><p>By the time we approached ground testing, more test personnel had come on board, and before we knew it, the entire team responsible for the first flight had come together.</p><p><strong>Kuwamura:</strong> Nobody ever declared, “This is where we build the team.”</p><p>It felt as though the team came together naturally as the aircraft itself moved closer to completion.</p><p>The schedule was demanding, but whenever a problem came up, someone stepped in to help, even if it was outside their formal area of responsibility. That was the atmosphere we had.</p><p><strong>Drew:</strong> And that directly supported our preparation for first flight.</p><p>We did not make a single decision that said, “Yes, the aircraft is ready to fly.”</p><p>We had a Flight Readiness Review, or FRR. We had safety reviews. We had a Go/No-Go decision the day before. We had another briefing on the morning of the flight.</p><p>Then, immediately before takeoff, we went around the telemetry room and received final confirmation from everyone there.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*JkfFDufOLHJjTDZQ" /><figcaption><em>Telemetry room at the development site in San Luis Obispo, California, USA</em></figcaption></figure><p><strong>Kuwamura:</strong> Anyone could say, “We should not fly yet.”</p><p><strong>Drew:</strong> Exactly.</p><p>Each time we passed through another gate, we reconfirmed that we were ready.</p><p><strong>Kuwamura:</strong> We also chose not to conduct a tethered hover test, where the aircraft is restrained to the ground while lifting off.</p><p><strong>Drew:</strong> We had built enough confidence through model predictions, integration testing, and reviews that we were able to proceed directly to free flight.</p><p><strong>Kuwamura:</strong> By the time we said “Go,” almost all of the major preparation we could do had already been completed.</p><p>Of course, successfully completing the flight itself was a major achievement.</p><p>But to me, reaching the point where we could genuinely say “Go” was just as important.</p><p><strong>Drew:</strong> Right before takeoff, I actually felt a little sad.</p><p>We had done everything we could. The aircraft was operating correctly.</p><p>At that point, all we could do was hand it over to the pilot and support the flight.</p><h3><strong>Taking Turns on the Gas and the Brake</strong></h3><p><strong>Kuwamura:</strong> Looking back at the road to first flight, Drew and I really were together every day.</p><p>This project is being developed by teams in both Japan and the United States, so it was extremely important to make sure that people from different locations and different technical disciplines were all looking at the same aircraft in the same way.</p><p><strong>Drew:</strong> The members who came over from Japan also played an important role in connecting the engineers in Japan with the U.S. team.</p><p>The team was larger than the one for the subscale aircraft, but it was still relatively small.</p><p>That made it even more important for each person to look beyond their own area of responsibility and understand the aircraft as a whole.</p><p><strong>Kuwamura:</strong> Drew is extremely calm.</p><p>When I’m pushing hard to move forward, he might say, “Let’s check this one more time.”</p><p>But you were not always the one putting on the brakes.</p><p><strong>Drew:</strong> No, sometimes it was the other way around.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*_QESGDp74rxR8cPl" /></figure><p><strong>Kuwamura:</strong> We needed both safety and speed.</p><p>So it was never a case of one person always pressing the accelerator and the other always pressing the brake.</p><p>Depending on the situation, we would switch roles — one of us pushing to move forward, the other stopping to make sure we had checked what needed to be checked.</p><p><strong>Drew:</strong> Koya puts a tremendous amount of importance on data.</p><p>He wants to make decisions only after truly understanding what we are doing.</p><p>At the same time, he avoids getting lost in the details and keeps sight of the project as a whole.</p><p>New issues kept coming up on the way to first flight. There were always more things we could have analyzed.</p><p>But Koya was very good at identifying what mattered most in the moment and keeping the team focused on it.</p><p><strong>Kuwamura:</strong> Drew also never confines himself to his own area of responsibility.</p><p>He voluntarily became involved in integration testing from a very early stage.</p><p>Watching him made me realize I couldn’t afford to look only at my own scope either.</p><p>I think we influenced each other in that way as we worked our way toward first flight.</p><h3><strong>Toward the Next “Go” — Until We Can Picture the Future</strong></h3><p><strong>Kuwamura:</strong> First flight was a major milestone, but in terms of the overall development program, we are still only at the beginning.</p><p>From here, what matters is how quickly we can learn from the data we gather through flight testing and connect those lessons to the next step.</p><p><strong>Drew:</strong> I also want us to maintain the energy we had leading up to first flight through the rest of the development program.</p><p>The aircraft is currently remotely piloted, and no one is onboard yet.</p><p>To turn it into a vehicle that carries people and that anyone can use with confidence, we still have a great deal of design and testing ahead of us.</p><p><strong>Kuwamura:</strong> Moving quickly is important.</p><p>But ultimately, we want to create a product people will actually ride in.</p><p><strong>Drew:</strong> The true goal is to create something safe and reliable enough that people do not feel any special anxiety about using it.</p><p>They should be able to board it much like they board an airplane today, travel to their destination, and go see the people they want to see.</p><p><strong>Kuwamura:</strong> Before joining Honda, I worked at a startup developing eVTOL aircraft.</p><p>In a small team, I gained experience in everything from system design to flight testing. At first, I thought, “Maybe we can do all of this ourselves.”</p><p>But if you want to create something your family and friends can actually use, simply making it fly isn’t enough.</p><p>You need to build up safety and reliability, achieve certification, and have the development foundation required to deliver it to society as a real product.</p><p>One reason I joined this project was that Honda is thinking not only about building prototypes, but about taking the technology all the way to commercialization.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*sK3ymGWiRZlw8GMw" /></figure><p><strong>Drew:</strong> Before Honda, I worked in flight testing for fixed-wing business jets.</p><p>In that world, many of the fundamental technologies are already well established. A lot of the work is about optimization — making the aircraft a little faster, or a little more comfortable.</p><p>eVTOL is different. It is still new.</p><p>As an aerospace engineer, you do not get many opportunities to create something from zero that has never existed before.</p><p>That is what brought me to Honda.</p><p><strong>Kuwamura:</strong> Honda has experience delivering a wide variety of products to society, from motorcycles and automobiles to Power Products and aircraft.</p><p>At the same time, the eVTOL team itself is relatively small and can move with a strong sense of speed.</p><p>I think what makes this project so exciting is that we can use both strengths to create something that can genuinely reach society.</p><p><strong>Drew:</strong> The idea of eVTOL has been talked about for many years now.</p><p>But for many people, it may still look like an experimental “flying taxi” that hasn’t moved beyond the prototype stage.</p><p>I see this aircraft every day, so I can picture a future where eVTOLs are flying over our cities.</p><p>Now that Honda has flown a full-scale model, I hope more people can imagine that future and think, “Maybe this really can happen.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*JStPuvg3vXe9WBH4" /></figure><p><strong>Kuwamura:</strong> If this kind of mobility becomes part of everyday life, people will be able to reach places that are difficult to get to today without worrying about traffic congestion.</p><p>Personally, I love hot springs, so I would be very happy if I could decide on a whim and get to one much more easily. <em>(laughs)</em></p><p>But it goes beyond that.</p><p>It could make it easier to see the people you want to see, and give you greater freedom to visit the places you have always wanted to go.</p><p>The first flight made that vision feel just a little more real.</p><p>Now we want to work seriously to make that future part of everyday life.</p><p><strong>Drew:</strong> That is the future we imagine every day while we work.</p><p>I hope the people who saw this flight can imagine just a little of that same future, too.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7c1d931a81d3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Can the Speed We Showed at 271.4 Meters Carry Us to 100 Kilometers? — Where Honda’s Sustainable Rock]]></title>
            <link>https://medium.com/@HGRX/can-the-speed-we-showed-at-271-4-meters-carry-us-to-100-kilometers-where-hondas-sustainable-rock-8fbd139304ff?source=rss-fd02d7b9dda5------2</link>
            <guid isPermaLink="false">https://medium.com/p/8fbd139304ff</guid>
            <dc:creator><![CDATA[HGR / Honda R&D Center of Excellence]]></dc:creator>
            <pubDate>Wed, 16 Sep 2026 07:01:02 GMT</pubDate>
            <atom:updated>2026-09-16T07:01:02.030Z</atom:updated>
            <content:encoded><![CDATA[<h3>Can the Speed We Showed at 271.4 Meters Carry Us to 100 Kilometers? — Where Honda’s Sustainable Rocket Project Stands in Phase 2</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SIJO5a051c35J1mV2GXf-Q.jpeg" /></figure><p>Hello everyone. I’m Kazuki Kawano, and I work on attitude control for Honda’s sustainable rocket in the New Mobility Research Domain at HGR.</p><p>I joined Honda straight out of university in 2023 and was assigned to the rocket team. Today, I work on what we call GNC — Guidance, Navigation, and Control. In simple terms, it is the system that enables a rocket to understand where it is, where it needs to go, and how it should move to get there.</p><p>In June 2025, Honda’s sustainable rocket experimental vehicle successfully completed a launch-and-landing test in Taiki, Hokkaido. It reached an altitude of 271.4 meters, remained in flight for 56.6 seconds, and landed just 37 centimeters from its target point. It was the first successful launch-and-landing demonstration by a private company in Japan.</p><p>But for us, the goal is still far ahead.</p><p>What we are looking toward next is 2029: reaching an altitude of 100 kilometers, the boundary between Earth and space, in a suborbital flight. Numerically, that is roughly 370 times higher than 271.4 meters. But height is far from the only challenge we have to overcome. The vehicle, the testing, the operations, and the scale of the team will all change.</p><p>Once a team has successfully flown a rocket, what needs to change next? How do we carry the speed that was possible because we were a small team into a much larger development effort? And how do we evolve without losing the freedom to get our hands dirty and build things ourselves — something that has always been part of Honda?</p><p>This time, I sat down with Takahiro Nakamura, Omi Sakai, and Makoto Horikawa to talk about what was happening behind the scenes during that flight, and where the rocket team stands today as we enter Phase 2.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*PZ2diaaclWD21ySL" /><figcaption><em>From left: Makoto Horikawa, Kazuki Kawano, Takahiro Nakamura, Omi Sakai</em></figcaption></figure><blockquote><strong>Takahiro Nakamura</strong><br>Works on aerodynamic studies and analysis of the environments the vehicle encounters from launch through its journey into space, while also contributing to system-level studies. Joined Honda in 2020. When he joined, the team consisted of only about seven people.</blockquote><blockquote><strong>Omi Sakai</strong><br>Responsible for development of the turbopump that delivers propellant from the tanks to the combustion chamber, building reliability in the engine area, and creating test environments. Joined Honda straight out of university in 2015, worked on automotive engine development, and transferred to the rocket team through an internal job posting in 2021.</blockquote><blockquote><strong>Makoto Horikawa</strong><br>Responsible for defining the rocket’s flight trajectory and for flight safety. Joined Honda in 2024. At his previous company, he worked on rocket trajectory analysis and booster separation analysis.</blockquote><blockquote><strong>Kazuki Kawano</strong><br>Responsible for GNC — Guidance, Navigation, and Control — design. Joined Honda straight out of university in 2023 and has since worked on test operations and implementation of control functions for the rocket team.</blockquote><h3>Flying Wasn’t Enough to Call It a Success — What Four Engineers Were Watching from One Kilometer Away</h3><p><strong>Kawano:</strong> As we work toward reaching suborbital altitude in 2029, we’re now facing a much bigger wall than before. But I think that in order to understand how we should overcome it, we first need to look back at what that previous “success” really meant.</p><p>What were each of you watching that day? I was inside the control room, so I only saw the rocket through the screens.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*s81CJ97u7IxnIUp5" /></figure><p><strong>Sakai:</strong> Nakamura-san and I were next to the vehicle right up until shortly before launch, making all kinds of preparations. Once all the checks were complete and launch was approaching, we jumped into a car and headed back to the control room.</p><p>Once I got there, I watched the monitor in front of me with the emergency stop button right at my hand. If anything abnormal happened during the flight, I had to press it and stop the test. That was how I watched the entire flight.</p><p><strong>Nakamura:</strong> We were also the people who evacuated last and would be the first to run back to the vehicle afterward.</p><p><strong>Horikawa:</strong> I was sitting next to the person coordinating the operation and supporting them. I was monitoring things like how much propellant remained in the tanks and advising whether the conditions required for flight were still being met.</p><p><strong>Kawano:</strong> The window of conditions in which we can actually fly is extremely narrow, isn’t it? A lot of things have to line up.</p><p><strong>Horikawa:</strong> Exactly. I was monitoring the other conditions as well, drawing interpolated lines in Excel and saying things like, “In about three minutes, this is going to become difficult,” or, “Actually, I think we can go now.”</p><p>Basically, I was working behind the scenes.</p><p><strong>Nakamura:</strong> Although you were the one giving some of the most important advice.</p><p><strong>Kawano:</strong> I really was just reading numbers off the monitor. Everyone else was carrying out their own responsibilities with such focus that I told myself, as long as nothing abnormal happened, my job was simply to read those numbers correctly.</p><p>It had been cloudy all morning that day. Everyone was hoping the weather would clear — and somehow, it did.</p><p>After that, the only thing going through my mind was: just come back down safely.</p><p><strong>Nakamura:</strong> Kawano-san was operating from a building about one kilometer away from the launch site. One of his important responsibilities was pressing the final button that meant, “You are cleared to fly.”</p><p>He was also doing the countdown, so the voice you hear in the video that was released publicly is his.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FlTCMHnNY0cs%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DlTCMHnNY0cs&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FlTCMHnNY0cs%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/8cd1e72e65c9f6c247896ec189a3e5d6/href">https://medium.com/media/8cd1e72e65c9f6c247896ec189a3e5d6/href</a></iframe><p><strong>Kawano:</strong> When the vehicle came down safely, I thought, “Yes!”</p><p>But that still wasn’t the end of the test.</p><p><strong>Sakai:</strong> Right. It comes back, it lands, and you think, “We did it.” But the experiment still isn’t finished.</p><p>The moment it lands, Nakamura-san and the others head straight back out to the rocket.</p><p>From previous experience, we knew there were cases where things around the vehicle could be burning. So we took fire extinguishers and rushed out to make sure there really were no problems, either with the vehicle itself or the surrounding area.</p><p>We all checked with each other: “Everything okay?” “Everything okay?”</p><p>Then I said, “All right, I’ll make the call.”</p><p>Over the radio, I announced, “Site safety confirmed. The test is now complete.”</p><p>I really savored those words when I said them. That was the moment when I finally felt: it’s over. We succeeded.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*YGtTiR4W-Moc8fOA" /></figure><p><strong>Nakamura:</strong> What happened that day wasn’t something we had prepared for only that one day. It was the result of everything we had built up beforehand.</p><p>Even before the vehicle was flying, we kept practicing those procedures. The operations team had internalized the idea that nobody relaxes until safety has been confirmed and the person responsible for site safety officially declares that the area is safe.</p><p><strong>Kawano:</strong> So just because it flew doesn’t mean it was a success.</p><p><strong>Nakamura:</strong> Exactly. Only after the final safety confirmation is it a success.</p><p><strong>Kawano:</strong> Nakamura-san and Horikawa-san, you had both experienced rocket launches at your previous companies. How was this launch-and-landing test different?</p><p><strong>Horikawa:</strong> At my previous company, I experienced about three launches of orbital rockets at the launch site.</p><p>But they weren’t reusable vehicles.</p><p>Once restrictions were lifted after launch and you went to look at the launcher, it would be covered in soot — and there was nothing there. You never saw the vehicle itself sitting outside in front of you again.</p><p>It almost made you wonder whether there had really been a rocket there in the first place.</p><p>This time, though, the rocket actually came back down. And it landed where we expected it to.</p><p>Seeing it there gave me this very fresh feeling: it really took off from here and came back here.</p><p><strong>Nakamura:</strong> I had always worked with rockets that were simply gone once they launched, so actually seeing one come back felt completely new.</p><p>A rocket that takes off and lands.</p><p>There still aren’t many examples of that in Japan, either. So my most straightforward feeling was simply: we finally made it this far.</p><h3>What Scared Us More Than 300 Meters Was 50 Centimeters — One Do-or-Die Moment After Another</h3><p><strong>Kawano:</strong> To be honest, the 300-meter test made me nervous too, but the test before that — when we lifted the vehicle just 50 centimeters off the ground — is burned into my memory even more strongly.</p><p>Until then, that vehicle had never left the ground.</p><p>Then it rose slightly and its legs separated from the surface.</p><p>My heart was pounding at that moment. I remember thinking, we have to get this thing flying this year.</p><p><strong>Nakamura:</strong> We would take on one challenge after another, things wouldn’t work, and the schedule would slip. That happened repeatedly.</p><p>So from very early on, we had discussions again and again: if we can’t get the performance we need, do we force the project to end by the original deadline? Or do we extend it and keep trying?</p><p>We went through situations like that many times before we finally reached that point.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*rAgbnxAkthvBJFkY" /></figure><p><strong>Horikawa:</strong> It feels like we had our backs against the wall more than once.</p><p><strong>Kawano:</strong> Nakamura-san, you felt that way on the day of the flight too, right?</p><p><strong>Nakamura:</strong> As the scale of the tests grew, I became increasingly aware that if a malfunction or failure caused some kind of trouble for the people around us, it would not have been surprising for the entire project to be cancelled.</p><p>That was the level of responsibility involved in what we were doing.</p><p>Another source of anxiety was timing.</p><p>We had already postponed the test several times, and the remaining launch window was getting very short. We were reaching a point where, if we missed one of those few remaining opportunities, we might not be able to conduct the test at all.</p><p>On top of that, some of the reasons for the previous delays came from issues I had caused myself.</p><p>That added another kind of pressure.</p><p><strong>Kawano:</strong> Which made the reaction after the successful flight all the more incredible. Sakai-san, you spent quite a while reading comments online afterward, didn’t you?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*DCvTejk4fNnnwqzN" /></figure><p><strong>Sakai:</strong> The response was much more positive than I had expected.</p><p>These days, even when something is good news, you often see at least some negative comments. So I assumed there would probably be a bit of that too.</p><p>But on social media and elsewhere, there were just so many positive reactions.</p><p>I realized there were this many people who were happy about what we had done and excited about what might come next.</p><p>It was so far beyond what I expected that I felt incredibly rewarded.</p><p><strong>Kawano:</strong> It made me realize that rockets have this power to excite people.</p><p>But for us, it wasn’t as though the spotlight suddenly appeared out of nowhere. I think what people saw was the view that finally opened up after we had made it through one difficult situation after another.</p><h3>Speed Was Our Weapon. But What Comes Next? — Keeping What Makes Honda, Honda</h3><p><strong>Sakai:</strong> During the tests in Taiki, I think the speed we could achieve precisely because we were a small team became a major strength.</p><p>But whether we can produce that same speed in the next phase is not a simple question.</p><p>In Phase 2, the vehicle will be larger, the tests and operations will become more complex, and the scope of what we have to consider will expand.</p><p>That’s exactly why I think we first need to look carefully at how we actually worked during Phase 1.</p><p><strong>Kawano:</strong> This is the only team I’ve known since becoming a working engineer, so I’d actually like to ask you two instead.</p><p>Nakamura-san and Horikawa-san, what did you think when you first came to Honda?</p><p><strong>Horikawa:</strong> First of all, we couldn’t understand each other’s terminology.</p><p>I didn’t know Honda’s internal terminology, and when I used specialized terminology from the space industry, nobody understood me either.</p><p>I would pull up some data and say, “You can just look at this,” and people who weren’t familiar with it would ask, “Wait, what was this again?”</p><p>So you start from explaining that.</p><p>But it also forced me to go back and ask myself, “Why do we need to do this in the first place?”</p><p>Although if you think about it too much, it does become a little tiring and you start thinking, “Come on, just understand what I mean.” <em>(laughs)</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*uWMBhUC-ZJ-14-U3" /></figure><p><strong>Nakamura:</strong> It isn’t just that the terminology is different. Our backgrounds are different, so our ways of thinking are fundamentally different too.</p><p>That means we have to align not only on what the words mean, but on questions like, “What exactly is different about the way we’re thinking?”</p><p>Sometimes it feels almost like people from different countries trying to reconcile two different cultures.</p><p><strong>Kawano:</strong> And those differences appear in the way we run tests too.</p><p><strong>Nakamura:</strong> In my previous work, the usual approach was to carefully define the requirements for a system and then develop that system very deliberately.</p><p>At Honda, though, you can say, “It doesn’t have to be perfect. Let’s build something and test it once.”</p><p>You have suppliers nearby who can make the parts, people inside the company who can assemble them, and facilities where you can run the test. Then you can immediately feed the results back into the next design.</p><p>The ability to complete that whole cycle quickly, with everything physically close together, really impressed me.</p><p><strong>Kawano:</strong> That freedom and speed probably showed themselves most clearly when something went wrong at the test site.</p><p><strong>Sakai:</strong> At one point, part of the test vehicle was damaged.</p><p>We didn’t have a spare part.</p><p>But if we didn’t repair it, there was absolutely no way we could fly.</p><p>So we asked ourselves what we could do, searched around the test site together, and found a sheet of metal that looked usable.</p><p><strong>Kawano:</strong> Then everyone started cutting it, grinding it, bending it.</p><p>And eventually, people even started sewing.</p><p><strong>Nakamura:</strong> We were sewing, hammering metal — you name it. We made this emergency repair and got the vehicle back into a condition where we could fly it again the next day.</p><p>And we did all of that within a single day.</p><p>Even when something goes wrong, the next challenge immediately begins.</p><p>I remember thinking, this is unbelievably fast.</p><p>And it wasn’t only the hardware.</p><p>We found an issue with a control parameter for landing and realized, “This isn’t good.”</p><p>That had to be fixed overnight too.</p><p>With hardware, it’s relatively easy to judge visually whether you’ve restored something properly. Software is much harder to verify.</p><p>And this was one of the most critical parts of the system.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ISUE7DmJ9RugRYfR" /></figure><p><strong>Kawano:</strong> In my case, when I reviewed the data afterward, I realized that the radio communications hadn’t been working properly.</p><p>Something like that absolutely cannot happen while the vehicle is flying.</p><p>I struggled with it, thought through the problem, and came up with an answer within a single day.</p><p><strong>Nakamura:</strong> Whether it’s hardware or software, if something happens, you somehow fix it by the next day, implement the change, and get everything ready for another test.</p><p>The people who can determine whether it is safe to proceed and say, “You can run the test,” are right there on site. And the people who can actually make the repair are gathered there too.</p><p>That’s the kind of environment we have.</p><p><strong>Kawano:</strong> From a perspective outside Honda, how unusual is that?</p><p><strong>Nakamura:</strong> Running multiple engine firing tests in a single day is almost unthinkably fast by the standards of the traditional space industry.</p><p>And Honda wasn’t publicizing any of that at all.</p><p>I found it a little amusing that we didn’t even seem to realize that this speed itself had value.</p><p><strong>Horikawa:</strong> In some ways, maybe it’s closer to what happens in racing.</p><p>There are people on the team with Formula 1 experience, and I hear stories from them often.</p><p>Once a race weekend begins, you have no choice but to solve whatever happens right there. Time is limited.</p><p>I think the feeling may be similar.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*RByXOnDDJ-_rCqhQ" /></figure><p><strong>Kawano:</strong> And that speed is something we’re still arguing about today.</p><p><strong>Sakai:</strong> We definitely are.</p><p>Even in component testing for the Phase 2 rocket, there are times when people say, “You’ve only done this level of analysis. You shouldn’t run the test yet,” or, “You need to understand it in greater detail before testing.”</p><p>I understand what they mean.</p><p>But throughout this rocket development, there have been so many things we only discovered because we actually tried them.</p><p>There are so many things you simply cannot know until you do them.</p><p>So once we are prepared to a reasonable level, I think it is better to actually try.</p><p><strong>Horikawa:</strong> For people who come from the space industry, the mindset of trying to solve everything through analysis becomes deeply ingrained.</p><p>Testing costs money, so except for the tests that are genuinely necessary, there is this sense of responsibility that you should first do everything possible through analysis.</p><p>And because there is often a lot of reference data available, you can even validate the accuracy of the analysis itself.</p><p>That makes you think, “Why wouldn’t you analyze this first? Isn’t it normal to use analysis to make sure the answer is right?”</p><p>But for the people doing something for the first time, the feeling is more like, “It’s our first time. Just let us try it.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*rqsBmQLtf3MBMsek" /></figure><p><strong>Nakamura:</strong> That said, if you go into testing without developing the specifications deeply enough, you can end up with a lot of problems.</p><p>There is another side to the “build it, test it, and discover it didn’t work” approach.</p><p>That’s where I think the experience and knowledge people like us bring from the space industry can be useful.</p><p><strong>Horikawa:</strong> And even in the traditional space industry, when an actual launch is approaching, companies other than Honda work with that same kind of intensity.</p><p>It’s just that the pace tends to slow down in other situations.</p><p>Still, Honda is certainly faster overall, and that is something I want us to preserve.</p><p>The difficult part for me is that if I simply apply everything I used to consider standard practice in the space industry, I feel like we could erase what is good about Honda.</p><p>There are things where I think, “We really have to do this.”</p><p>But at the same time, I worry that doing them will make that freedom disappear.</p><p>That’s something I’m still struggling with.</p><h3>Taking a Small Team’s Speed to the Next Scale — What Has to Change in Phase 2</h3><p><strong>Sakai:</strong> What became clear in Phase 1 was how much speed came from being a small team.</p><p>The people responsible for all the different areas were gathered in the same place. They could make decisions immediately and act on them.</p><p>I think that was a major strength.</p><p><strong>Kawano:</strong> That is another way this differs from automotive development, isn’t it?</p><p><strong>Sakai:</strong> In automotive development, work is divided among many more people, and often you have to build agreement across a number of groups before moving to the next step.</p><p>Being small gave us an advantage there.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*zNXr-OTLK5e5bKbV" /></figure><p><strong>Kawano:</strong> But what we’re targeting next is suborbital flight in 2029.</p><p>The vehicle will be larger. The tests will be larger. Operations will be larger. The scope of what we have to consider will be much broader.</p><p>Can we keep working exactly the same way?</p><p>That’s one of the difficult questions of Phase 2.</p><p><strong>Nakamura:</strong> That’s exactly what we’re struggling with right now.</p><p>For the activities ahead, I don’t think we can simply continue working exactly as we have until now.</p><p>The scale of what we need to do is becoming much larger.</p><p>We are reaching a point where it will be difficult to continue with a model in which a small number of people see everything, everyone gathers together, and decisions are made on the spot.</p><p>I think we are entering a stage where we have to move toward a certain degree of specialization and division of responsibilities.</p><p><strong>Sakai:</strong> The propulsion and testing areas will change significantly as well.</p><p>During Phase 1, both the vehicle and the experiments were relatively small in scale, so we could respond to the issue directly in front of us and make decisions on the spot.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*piTOxIgXJA1a7b0Q" /></figure><p><strong>Kawano:</strong> Things are going to change a lot in your area too, Sakai-san, aren’t they?</p><p><strong>Sakai:</strong> We’ll have to move forward simultaneously on turbopump development, building reliability, creating test environments, and many other things.</p><p>So it can’t simply be about “moving fast.”</p><p>We need to prepare in ways that allow us to make decisions quickly, and the team needs to share the assumptions and criteria behind those decisions in advance.</p><p>We need to turn the speed that was possible because we were a small team into something we can reproduce as a system.</p><p><strong>Kawano:</strong> I think the nature of the challenge will change for GNC as well in Phase 2.</p><p>It isn’t enough simply to make the vehicle fly.</p><p>The system has to continuously determine where the vehicle is and how it should return.</p><p>As altitude increases, the scope we have to consider changes. The influence of external disturbances changes. The errors we can tolerate change too.</p><p><strong>Nakamura:</strong> Because it has to work not only on the way up, but all the way through bringing the vehicle back.</p><p><strong>Kawano:</strong> During Phase 1, much of our attention was focused on how to make the experiment directly in front of us succeed.</p><p>Next, we have to bring the vehicle back from a much higher altitude, under more demanding conditions, and still return it to where we intend.</p><p>That cannot be solved by control technology alone.</p><p>The vehicle, propulsion, trajectory, and approach to safety all have to be connected.</p><p>It isn’t enough for GNC alone to become smarter.</p><p>We are entering a stage where the whole team has to align on how we want the vehicle to fly — and how we want it to return.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*MNpk_ZhcHbgb4weX" /></figure><p><strong>Horikawa:</strong> From the perspective of flight safety too, the further we progress, the broader the area we have to consider becomes.</p><p>A rocket cannot simply be allowed to fly.</p><p>As altitude increases, the variation in possible trajectories becomes wider, and the potential impact area that has to be considered in the event of an emergency becomes larger.</p><p>More and more things simply won’t work if we approach them with the same mindset we used in Phase 1.</p><p>At the same time, if we lock everything down too tightly in the name of safety, we lose Honda’s speed and freedom.</p><p>But if we allow too much freedom, flight safety itself becomes impossible to maintain.</p><p>Designing the right balance between those two is going to be extremely important in Phase 2.</p><p><strong>Nakamura:</strong> Scaling up without losing speed.</p><p>That’s the difficult part.</p><p><strong>Horikawa:</strong> Flight safety is sometimes seen simply as something that puts on the brakes.</p><p>But that isn’t really what it is.</p><p>It is also the job of defining how far we can safely go so that we can keep taking on challenges.</p><p>I want us to create an approach to safety that can handle the scale of what comes next without losing Honda’s speed.</p><p><strong>Kawano:</strong> Phase 2 isn’t simply about building a bigger rocket.</p><p>The mechanisms for making decisions quickly have to change. The controls for bringing the vehicle back have to change. Our approach to safety has to change.</p><p>Everything has to evolve to match the next scale.</p><p>And in the middle of all that, the question is how we preserve our speed.</p><p><strong>Nakamura:</strong> Exactly.</p><p>Reaching suborbital altitude in 2029 is already an extremely high hurdle.</p><p>And if you look beyond that, to put it in one word: it’s chaos.</p><p>Unlike last time, we don’t have results that we can immediately show publicly either.</p><p>For a while, I think we’ll be going back underground, so to speak. Much of what we’re doing will be difficult to see from the outside.</p><p><strong>Kawano:</strong> It may be hard to see from outside, but internally there are a huge number of things that have to change.</p><p><strong>Nakamura:</strong> Exactly.</p><p>And even so, we still have to maintain our sense of speed.</p><p>We have to move fast.</p><p>Figuring out how to do that is the challenge we’re facing next.</p><h3>Beyond 100 Kilometers, One More Time — Turning Joy into Our Next Thrust</h3><p><strong>Sakai:</strong> As engineers, I think we experienced a kind of joy that not everyone gets to experience.</p><p>Knowing what that feels like is an enormous strength.</p><p><strong>Kawano:</strong> You mean the joy of that success in Taiki.</p><p><strong>Sakai:</strong> I was incredibly happy at that moment.</p><p>But the years leading up to it had been filled with difficult things.</p><p>We went through all of those challenges, and then at the end of it all, we discovered just how happy it could make us feel.</p><p>Having experienced that entire cycle once makes a difference.</p><p>We’re going through another difficult period now.</p><p>But part of what keeps me going is knowing that if we make it through this too, we may get to feel “that” one more time.</p><p><strong>Nakamura:</strong> That success filled my internal “satisfaction meter” until it almost overflowed.</p><p>But now I also feel as though I’m gradually spending that stored-up satisfaction every day.</p><p>I’m anxiousabout when the meter might finally hit zero.</p><p>But until we reach the next “that,” I suppose we’ll survive by finding smaller moments of satisfaction along the way.</p><p><strong>Kawano:</strong> I’ve still only experienced it that one time.</p><p>So I want to see the next “that” as soon as possible.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*F608A3bWMKovSrJF" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8fbd139304ff" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Untangling Flow, Building Trust [Part 2] — Beyond Aerodynamics: How a Technology Refined Together…]]></title>
            <link>https://medium.com/@HGRX/untangling-flow-building-trust-part-2-beyond-aerodynamics-how-a-technology-refined-together-43fc858b5494?source=rss-fd02d7b9dda5------2</link>
            <guid isPermaLink="false">https://medium.com/p/43fc858b5494</guid>
            <dc:creator><![CDATA[HGR / Honda R&D Center of Excellence]]></dc:creator>
            <pubDate>Wed, 02 Sep 2026 03:01:04 GMT</pubDate>
            <atom:updated>2026-09-02T03:01:04.075Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>Untangling Flow, Building Trust [Part 2] — Beyond Aerodynamics: How a Technology Refined Together Is Accelerating Honda</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*R8VM-HYqtvjRcdi0LyM0OA.jpeg" /></figure><p>Hello everyone. I’m Atsushi Ogawa, Chief Operating Officer of Honda R&amp;D Research Center of Excellence (HGR).</p><p><a href="https://medium.com/@HGRX/untangling-flow-building-trust-part-1-15-years-of-turning-stanford-born-cfd-into-a-technology-9bcfb1ad2c92">In Part 1, Honda aerodynamic engineer Kei Ambo, Guillaume Brès and Frank Ham of Cascade Technologies, and I looked back on the evolution of CFD — Computational Fluid Dynamics — at Honda and Cascade.</a></p><p><a href="https://medium.com/@HGRX/untangling-flow-building-trust-part-1-15-years-of-turning-stanford-born-cfd-into-a-technology-9bcfb1ad2c92">Untangling Flow, Building Trust [Part 1] — 15 Years of Turning Stanford-Born CFD into a Technology…</a></p><p>We talked about how the two sides first met, the pursuit of accuracy at the 0.1% error level, and how episodes such as the development of physically rotating wheels came to symbolize the trust we had built in each other’s expertise.</p><p>But producing an accurate answer alone does not yet make a technology a true tool for development. The answer also needs to come back when it is needed, at the speed it is needed. The technology has to become part of a development process that anyone can use. And it needs to expand into new areas.</p><p>That is where the Honda–Cascade collaboration entered its next stage.</p><p>In Part 2, we explore how a collaboration that began with the pursuit of accuracy moved through GPU acceleration to change Honda’s development process itself — and how the technology spread into entirely new fields. And after 15 years of working together, we look at the new possibilities now coming into view.</p><h3><strong>Three A’s Tell the Story of How the Collaboration Evolved</strong></h3><p><strong>Frank:</strong> I think we could describe everything we’ve talked about so far with three A’s. The first is <strong>Accuracy</strong>. If the answer isn’t correct, there is no point in using the technology in the first place.</p><p>The second is <strong>Automation</strong>. Honda played a major role in automating the simulations so they could be integrated into its internal development processes.</p><p>And the third is <strong>Acceleration</strong>. We had been working for years to make the calculations faster, but GPUs accelerated that effort dramatically.</p><p><strong>Ambo:</strong> Accuracy alone isn’t enough for actual vehicle development. We receive a geometry from the designers, run the calculation, and need to get the answer back to them as quickly as possible. For example, <strong>a simulation including rotating wheels took around 30 hours on CPUs but with 16 GPUs, it was reduced to approximately three hours.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Mx20ocHoMWVt2V-29cdcgA.jpeg" /><figcaption><em>From left: Kei Ambo, Guillaume Brès, Atsushi Ogawa, and Frank Ham</em></figcaption></figure><p><strong>Frank:</strong> That’s an enormous difference.</p><p><strong>Ambo:</strong> Real-world time matters enormously in vehicle development. If the result doesn’t come back while the designer is still considering the next geometry, the simulation drops out of the development cycle.</p><p><strong>Frank:</strong> No matter how accurate it is, if it’s too slow, people eventually stop using it.</p><p><strong>Ogawa:</strong> In the past, you would start a calculation and have to wait overnight. You wouldn’t know the result until the next morning. If you can get the answer back within a few hours instead, it changes the development process itself.</p><p><strong>Guillaume:</strong> Combining the moving solver with GPU acceleration also allowed us to move into fields beyond external aerodynamics. The collaboration expanded dramatically — from rotating blades on eVTOL aircraft (*1) to vehicle wind noise, and even turbofans.</p><p><em>*1 eVTOL (electric Vertical Take-Off and Landing): An electrically powered aircraft capable of vertical takeoff and landing.</em></p><h3><strong>Solving Faster Means Making the Next Move Sooner</strong></h3><p><strong>Ambo:</strong> The GPU solver wasn’t simply a port of the existing CPU solver, was it?</p><p><strong>Frank:</strong> No. To get high performance from GPUs, we rebuilt it from the architectural level.</p><p><strong>Ambo:</strong> Many solvers take software designed for CPUs and modify it to run on GPUs. But simply doing that doesn’t always produce enough of a speed increase.</p><p><strong>Guillaume:</strong> We rewrote almost everything. We used mixed-precision computing and designed it to run on GPUs from both NVIDIA and AMD, aiming to extract as much performance as possible from the hardware.</p><p><strong>Frank:</strong> Cascade was still a small company at the time, and that gave us the freedom to take a fairly bold risk. We wanted to see how close the software could get to the hardware’s theoretical performance. In fact, in an evaluation known as <strong>roofline analysis (*2)</strong>, we have reached roughly 80% of the hardware’s performance limit.</p><p><em>*2 Roofline analysis: A method used to evaluate how closely actual computational performance approaches a system’s theoretical hardware performance limit.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*XMV67UVs54Kite58" /></figure><p><strong>Ambo:</strong> We wouldn’t have reached that level simply by porting the existing solver.</p><p><strong>Frank:</strong> I don’t think so. But when you rewrite a solver, you also have to confirm that the new solver still gives the same correct answers as before. So we reran all of the many validation cases we had accumulated through our collaboration with Honda. That was an extremely important process.</p><p><strong>Guillaume:</strong> It was a long effort that took several years. Honda continued working with us throughout the process.</p><h3><strong>From “Colorful Fluid Dynamics” to a Technology That Supports Development</strong></h3><p><strong>Ogawa:</strong> Years ago, people used to call CFD “Colorful Fluid Dynamics jokingly<strong>.</strong> It could produce beautiful, colorful pictures. But could you really trust the numbers? For a long time, the answer wasn’t clear.</p><p><strong>Ambo:</strong> If all you want to do is visualize a flow using colors, a result can look convincing enough. But to use it for actual development decisions, you have to predict forces and pressures with very high accuracy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*3c71HgwyH8xLsfxy" /><figcaption><em>An illustration of airflow visualized using traditional CFD</em></figcaption></figure><p><strong>Ogawa:</strong> In aircraft development in the past, even after performing a CFD simulation, you ultimately had to confirm the numbers in a wind tunnel.</p><p>Today, however, in the development of eVTOL aircraft, there are cases where we can skip a wind tunnel test, trust the CFD results, and move directly to the next step. The role of CFD has changed enormously over the past 30 years.</p><p>Young engineers today may not even know that CFD was once jokingly called “Colorful Fluid Dynamics.” They begin from the assumption that the numbers produced by high-fidelity CFD can be trusted.</p><p><strong>Frank:</strong> I saw that change myself when I was at Stanford.</p><p>In the past, experimental researchers didn’t always take CFD results very seriously. But as high-fidelity methods improved, experimental researchers began coming to our offices and asking us, “What is actually happening in this flow?”</p><p><strong>Ogawa:</strong> First, CFD allowed us to understand the flow. Today, we can also trust the numbers themselves. That is a profound difference.</p><p><strong>Frank:</strong> Experiments can struggle to measure certain locations, whereas CFD gives you pressure and velocity information throughout the entire flow field. But that is only useful if you can trust it.</p><p><strong>Ambo:</strong> At the same time, evaluating CFD accuracy also requires a very high level of expertise on the experimental side. How do you measure static pressure? How do you handle the tubes and sensors? What kind of pressure gradients exist within the wind tunnel itself? Many subtle experimental conditions are involved.</p><p><strong>Ogawa:</strong> Especially with unsteady phenomena, the act of measuring them is itself difficult.</p><h3><strong>Beyond Aerodynamics, the Questions Keep Expanding</strong></h3><p><strong>Ambo:</strong> Our collaboration began with external vehicle aerodynamics. But I think aerodynamics was only the first entry point. Now we are moving into areas where multiple phenomena interact, such as noise and heat.</p><p><strong>Guillaume:</strong> Once we had both the moving solver and GPU acceleration, the range of problems we could address expanded dramatically. In an eVTOL aircraft, multiple rotors are moving, and the flow from each one interacts with the airframe and with other rotors.</p><p>And we don’t only need to predict the flow. We also need to predict the sound it generates. Our work with Honda has expanded into areas including HVAC (*3) noise and aerodynamic noise in automobiles, as well as turbofans and combustors in aviation.</p><p>*3 HVAC (Heating, Ventilation, and Air Conditioning): Systems used for heating, ventilation, and air conditioning.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*9QgAPRa9coHyEx4w" /><figcaption><em>Honda’s experimental eVTOL aircraft currently under development</em></figcaption></figure><p><strong>Frank:</strong> We had long been interested not only in predicting flow, but also in <em>hearing</em> the sound produced by that flow.</p><p>Predicting aerodynamic noise with high accuracy is extremely difficult. But even in the early stages of our work with Honda, we knew we would eventually move in that direction.</p><p><strong>Guillaume:</strong> Honda develops not only automobiles, but also aircraft, eVTOLs, and a wide variety of power units.</p><p>As a result, many members of our team have been involved in a Honda project in one form or another. The collaboration has become incredibly broad — not just external aerodynamics, but combustion, noise, eVTOL, HVAC, and turbofans.</p><p><strong>Ogawa:</strong> Another interesting aspect is that our work with Cascade has connected different teams inside Honda.</p><p>Honda has a diverse range of products, but that diversity doesn’t automatically generate technical exchange between teams. Even the racing-car and production-car aerodynamics teams used to operate separately.</p><p>But when different groups use the same simulation technology and discuss their challenges with the same partner, connections begin to form across organizational boundaries. Cascade has also helped connect Honda internally.</p><p><strong>Guillaume:</strong> Whenever Ogawa says, “Could we do this next?” or “Could we use this in another field?” I’ve always tried to answer, “Yes.”</p><p>Which is probably why I’ve been busy all these years. (Laughs)</p><h3><strong>The Devil Is in the Details</strong></h3><p><strong>Ambo:</strong> One thing that has been very important to our continued collaboration is being able to talk directly with Cascade’s R&amp;D team. We are aiming for extremely high levels of accuracy. If you want to discuss state-of-the-art technology, you need to speak directly with the people who develop the solver and can change it.</p><p><strong>Guillaume:</strong> Honda specifically asked to talk directly with Cascade’s R&amp;D team rather than going through general application engineers.</p><p>Even after we became part of Cadence, we have been able to maintain that direct relationship. That isn’t necessarily the standard way of working at a large company.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ffJ9EQiLTzWtOlJL" /></figure><p><strong>Frank:</strong> In my experience, Cadence as a large company is unique in this way: for example, the same kind of direct relationship exists within Cadence’s EDA (*4) business for particularly important customers.</p><p>If the issue is usability, a customer-facing team may be able to solve it. But if the issue concerns predictive accuracy or the fundamentals of the technology itself, it is important to speak with the people who can actually change the technology.</p><p><em>*4 EDA (Electronic Design Automation): A category of tools used to automate electronic circuit design for semiconductors.</em></p><p><strong>Ogawa:</strong> With a large company, you sometimes have to go through the local subsidiary or a sales contact. But for technically important issues, it has enormous value to have the developers talk directly to one another.</p><p><strong>Ambo:</strong> Guillaume-san in particular has served as our main point of contact for many years while maintaining a deep understanding of the situation on the Honda side. He knows how many users Honda has and what kinds of problems each team is facing. He has always stood between Cascade and Honda and helped maintain the balance across the overall relationship.</p><p><strong>Guillaume:</strong> Thank you. Perhaps I’ve played a role in guiding the partnership over the years. But in reality, everything we’ve achieved has come from both companies working together as one team.</p><p>There are so many team members on both sides that played critical roles in this success. In particular on the Cadence/Cascade side, these achievements would not have been possible without key developments from Sanjeeb Bose and Chris Ivey on core technology and numerical methods, Dave Philips on external aerodynamics and Lee Shunn on combustion applications.</p><p><strong>Ambo:</strong> From Honda’s perspective, we’re genuinely grateful that we’ve been able to build this kind of relationship.</p><p><strong>Guillaume:</strong> I’m glad to hear that. It was only possible because so many people on the Honda side became involved, including Minoru Teramura, Hisato Tanaka and Hiroaki Nagaoka, just to name a few.</p><p>That is how a relationship that began with external aerodynamics expanded into so many different areas. We have built trust at every level.</p><p>Incidentally, inside Cascade we have a nickname for Ogawa-san: <strong>“Ageless Wonder.”</strong></p><p><strong>Ogawa:</strong> Ageless Wonder?</p><p><strong>Guillaume:</strong> Fifteen years have passed, and your passion for technology and your energy haven’t changed at all. That’s why we call you that. If someone at Cascade says, “We have a meeting with the Ageless Wonder,” everyone immediately knows who they mean.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*eXRFep6ED2u9WBK3" /></figure><p><strong>Ogawa:</strong> I’m honored. (Laughs)</p><p>As I said in Part 1, I felt I could trust Cascade from the beginning because the team clearly distinguished between what they could do and what they couldn’t.</p><p>Even for problems they couldn’t yet solve, they understood exactly what was causing the difficulty. That meant we could discuss what would need to change to move forward.</p><p><strong>Frank:</strong> That may be part of the culture that came from Stanford’s Center for Turbulence Research. We placed great importance on getting the right answer for the right reasons. And to do that, you have to acknowledge what you can and cannot do honestly. If you blur the things you cannot do, you cannot move the technology forward.</p><p><strong>Ogawa:</strong> Professor Parviz Moin at Stanford is an internationally renowned researcher, but even today, he listens to what everyone has to say. He is always trying to learn something. He is willing to learn even from someone much younger than himself, or from someone with a completely different area of expertise.</p><p><strong>Frank:</strong> Studying turbulence may naturally make you humble. Turbulence is still not a fully solved problem. No one has all the answers. So you need both the willingness to challenge the problem and the humility to recognize what you still don’t know.</p><p><strong>Ambo:</strong> Cascade’s engineers always come up with a wide variety of possible solutions. And each idea is remarkably original. I think that comes from an environment where people can discuss mathematics and numerical methods at great depth, combined with a very strong foundation in the fundamentals.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*H-0AJ5HpjBVOBVrk" /></figure><p><strong>Frank:</strong> At the Center for Turbulence Research, we paid enormous attention to detail. To solve flows around something as complex as a vehicle, you need not only state-of-the-art methods, but also mathematical rigor.</p><p><strong>Ogawa:</strong> State-of-the-art methodology is important, of course. But going back to the fundamentals is equally important. Even with a principle that everyone supposedly knows, it takes deep understanding to stand up and say, “This is what really matters.”</p><p><strong>Frank:</strong> Going back to <strong>first principles</strong>.</p><p><strong>Ogawa:</strong> Exactly. You understand the difficulties hidden in the details, while remaining committed to a simple underlying concept. You could say, “God is in the details.”</p><p><strong>Guillaume:</strong> In English, we tend to say, <strong>“The devil is in the details.”</strong></p><p>Don’t hide the inconvenient details. Be transparent about what you know, what you don’t know, and where improvement is possible. I think that commitment to detail — and that honesty — has been the foundation of the trust we have built.</p><h3><strong>Different Fields, Connected by “Flow”</strong></h3><p><strong>Guillaume:</strong> We should also talk a little about what comes next.</p><p>Within CFD, there are themes such as machine learning and AI, optimization, and digital certification. We talked earlier about replacing wind tunnel testing. With Honda’s turbofan team, we are also pursuing work with future digital certification in mind.</p><p>And when you look at the broader relationship between Cadence and Honda, there is an even larger collaboration emerging in the field of semiconductors. The Cascade team will not necessarily be at the center of that effort, but it represents an important next step in the Honda–Cadence partnership.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*iLqpQNhDJ9m2CjEC" /></figure><p><strong>Ogawa:</strong> Honda has another team designing semiconductor logic chips.</p><p>EDA tools have automated much of chip design, but thermal analysis of chips and packages has not yet been fully integrated with physical design. Traditionally, chip physical design is completed first, and thermal analysis follows. If a problem is found at that stage, you have to go back to the design. And sometimes it isn’t easy to go back, which can result in a major loss of time. That is why the physical design of the chip and thermal analysis, including the package, need to be carried out in parallel.</p><p>Honda has completed the conceptual design of a neuromorphic chip (*5) and is now moving into logic design. When we move from there into physical design, I believe Cadence’s capabilities in CFD and thermal analysis could become extremely important.</p><p><em>*5 Neuromorphic chip: A semiconductor chip designed to emulate the structure of neural circuits in the brain.</em></p><p><strong>Ambo:</strong> Aerodynamics was only the beginning. From here, I think we can expand into heat, noise, and other even more complex phenomena.</p><p><strong>Guillaume:</strong> A relationship that began with fluid simulation is now expanding all the way into semiconductors. I’m excited to see where this relationship goes next.</p><p><strong>Ogawa:</strong> Fluid simulation and chip design used to belong to completely separate teams — both within Honda and within Cadence. But now those two worlds are beginning to connect. I certainly didn’t imagine this kind of development when we first started working together. It’s an interesting turn of events, isn’t it?</p><p>A relationship that began by untangling flow is now connecting entirely different fields. That, too, is a next step none of us could see 15 years ago.</p><p>■Related Article</p><p><a href="https://medium.com/@HGRX/untangling-flow-building-trust-part-1-15-years-of-turning-stanford-born-cfd-into-a-technology-9bcfb1ad2c92">Untangling Flow, Building Trust [Part 1] — 15 Years of Turning Stanford-Born CFD into a Technology…</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=43fc858b5494" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>