Deepak Vij·Oct 20From Imitation to Imagination: VLAs vs. World Action ModelsRobot foundation models are shifting from learning what to do to learning what will happen. For two years, the default recipe has been the…
Fancyfrank·Oct 15Exporting ACT with LEAPP: Take the Whole Policy With YouYou trained a robot policy. It works in simulation. Now comes the part that sounds easy: export the model and run it somewhere else.
TAS Design Group Inc.·Sep 30XR-1 Learned to Grasp Real Server Parts, but Placing Them Is Another StoryFour rounds of pick-and-place in simulation: 0/20 zero-shot, 73% grasp after fine-tuning, and placement that topped out at 62% on the…
Ambarish G K·Sep 29Why Is It So Hard to Make a Robot Move With Grace?Watch a good robot pick up a cup of coffee and it looks like nothing at all. The arm swings over, the fingers close, the cup lifts, no…
Arun Murugesan·Sep 28I Taught a ₹1000 Robot Arm to Pick and Place Objects Using SmolVLA 🤖How a hobby-servo arm, an Arduino Uno, and a fine-tuned vision-language-action model ended up doing real imitation learning on a budget…
Deepak Vij·Sep 22How Robot Brains Actually Move: The Nuances of VLA Models Like π0.5Lessons from my early days exploring physical AI: the questions that tripped me up, and the answers I wish I’d had from the start
BruceWayne·Sep 22Three Layers Below the Model: What Shipping FP8 Collectives Taught Us About Quantizers, Allocators…I had a simple-sounding problem. Multi-modal VLA training stack runs on 8× GPU— 96 GB of GDDR7 each, 503.8 bf16 TFLOPS each, and no NVLink…
Chanwoo Song·Sep 17현대차가 내놓은 자율주행 AI지난 11일 판교 42dot (현대차그룹의 자율주행 소프트웨어 회사) 본사에서 현대차그룹이 자율주행 미디어 데이를 열었습니다. Atria AI를 탑재한 차량이 서울 도심을 운전자 개입 없이 주행하는 영상을 처음 공개했고요. 유튜브에 올라온 지 며칠…
Chunpak·Sep 16How I chose an imitation-learning arm after writing about Vision-Language-ActionIn my last piece, I tried to explain Vision-Language-Action。 That article stayed in the clouds on purpose. This one comes down to the desk.
Mehrdad H. M. Farimani·Sep 13Are We Building Robotics Backwards?Why I’m starting to think we should build general foundations first, and specialize closer to the customer.
