|
6 | 6 |
|
7 | 7 | """ |
8 | 8 |
|
| 9 | +import numpy as np |
| 10 | +import math |
| 11 | +import matplotlib.pyplot as plt |
| 12 | + |
| 13 | +# Simulation parameter |
| 14 | +Qsim = np.diag([0.2])**2 |
| 15 | +Rsim = np.diag([1.0, math.radians(30.0)])**2 |
| 16 | + |
| 17 | +DT = 0.1 # time tick [s] |
| 18 | +SIM_TIME = 50.0 # simulation time [s] |
| 19 | +MAX_RANGE = 20.0 # maximum observation range |
| 20 | + |
| 21 | +# Particle filter parameter |
| 22 | +NP = 100 # Number of Particle |
| 23 | +NTh = NP / 2.0 # Number of particle for re-sampling |
| 24 | + |
| 25 | +show_animation = True |
| 26 | + |
| 27 | + |
| 28 | +def calc_input(): |
| 29 | + v = 1.0 # [m/s] |
| 30 | + yawrate = 0.1 # [rad/s] |
| 31 | + u = np.matrix([v, yawrate]).T |
| 32 | + return u |
| 33 | + |
| 34 | + |
| 35 | +def observation(xTrue, xd, u, RFID): |
| 36 | + |
| 37 | + xTrue = motion_model(xTrue, u) |
| 38 | + |
| 39 | + # add noise to gps x-y |
| 40 | + z = np.matrix(np.zeros((0, 3))) |
| 41 | + |
| 42 | + for i in range(len(RFID[:, 0])): |
| 43 | + |
| 44 | + dx = xTrue[0, 0] - RFID[i, 0] |
| 45 | + dy = xTrue[1, 0] - RFID[i, 1] |
| 46 | + d = math.sqrt(dx**2 + dy**2) |
| 47 | + if d <= MAX_RANGE: |
| 48 | + dn = d + np.random.randn() * Qsim[0, 0] # add noise |
| 49 | + zi = np.matrix([dn, RFID[i, 0], RFID[i, 1]]) |
| 50 | + z = np.vstack((z, zi)) |
| 51 | + |
| 52 | + # add noise to input |
| 53 | + ud1 = u[0, 0] + np.random.randn() * Rsim[0, 0] |
| 54 | + ud2 = u[1, 0] + np.random.randn() * Rsim[1, 1] |
| 55 | + ud = np.matrix([ud1, ud2]).T |
| 56 | + |
| 57 | + xd = motion_model(xd, ud) |
| 58 | + |
| 59 | + return xTrue, z, xd, ud |
| 60 | + |
| 61 | + |
| 62 | +def motion_model(x, u): |
| 63 | + |
| 64 | + F = np.matrix([[1.0, 0, 0, 0], |
| 65 | + [0, 1.0, 0, 0], |
| 66 | + [0, 0, 1.0, 0], |
| 67 | + [0, 0, 0, 0]]) |
| 68 | + |
| 69 | + B = np.matrix([[DT * math.cos(x[2, 0]), 0], |
| 70 | + [DT * math.sin(x[2, 0]), 0], |
| 71 | + [0.0, DT], |
| 72 | + [1.0, 0.0]]) |
| 73 | + |
| 74 | + x = F * x + B * u |
| 75 | + |
| 76 | + return x |
| 77 | + |
| 78 | + |
| 79 | +def ekf_slam(xEst, PEst, u, z): |
| 80 | + # Predict |
| 81 | + xEst = motion_model(xEst, u) |
| 82 | + # [G,Fx]=jacobF(xEst, u); |
| 83 | + # PEst= G'*PEst*G + Fx'*R*Fx; |
| 84 | + |
| 85 | + return xEst, PEst |
| 86 | + |
| 87 | + # % Update |
| 88 | + # for iz=1:length(z(:,1))%それぞれの観測値に対して |
| 89 | + # %観測値をランドマークとして追加 |
| 90 | + # zl=CalcLMPosiFromZ(xEst,z(iz,:));%観測値そのものからLMの位置を計算 |
| 91 | + # %状態ベクトルと共分散行列の追加 |
| 92 | + # xAug=[xEst;zl]; |
| 93 | + # PAug=[PEst zeros(length(xEst),LMSize); |
| 94 | + # zeros(LMSize,length(xEst)) initP]; |
| 95 | + |
| 96 | + # mdist=[];%マハラノビス距離のリスト |
| 97 | + # for il=1:GetnLM(xAug) %それぞれのランドマークについて |
| 98 | + # if il==GetnLM(xAug) |
| 99 | + # mdist=[mdist alpha];%新しく追加した点の距離はパラメータ値を使う |
| 100 | + # else |
| 101 | + # lm=xAug(4+2*(il-1):5+2*(il-1)); |
| 102 | + # [y,S,H]=CalcInnovation(lm,xAug,PAug,z(iz,1:2),il); |
| 103 | + # mdist=[mdist y'*inv(S)*y];%マハラノビス距離の計算 |
| 104 | + # end |
| 105 | + # end |
| 106 | + |
| 107 | + # %マハラノビス距離が最も近いものに対応付け |
| 108 | + # [C,I]=min(mdist); |
| 109 | + |
| 110 | + # %一番距離が小さいものが追加したものならば、その観測値をランドマークとして採用 |
| 111 | + # if I==GetnLM(xAug) |
| 112 | + # %disp('New LM') |
| 113 | + # xEst=xAug; |
| 114 | + # PEst=PAug; |
| 115 | + # end |
| 116 | + |
| 117 | + # lm=xEst(4+2*(I-1):5+2*(I-1));%対応付けられたランドマークデータの取得 |
| 118 | + # %イノベーションの計算 |
| 119 | + # [y,S,H]=CalcInnovation(lm,xEst,PEst,z(iz,1:2),I); |
| 120 | + # K = PEst*H'*inv(S); |
| 121 | + # xEst = xEst + K*y; |
| 122 | + # PEst = (eye(size(xEst,1)) - K*H)*PEst; |
| 123 | + # end |
| 124 | + |
| 125 | + # xEst(3)=PI2PI(xEst(3));%角度補正 |
| 126 | + |
9 | 127 |
|
10 | 128 | def main(): |
11 | 129 | print(__file__ + " start!!") |
12 | 130 |
|
| 131 | + time = 0.0 |
| 132 | + |
| 133 | + # RFID positions [x, y] |
| 134 | + RFID = np.array([[10.0, 0.0], |
| 135 | + [10.0, 10.0], |
| 136 | + [0.0, 15.0], |
| 137 | + [-5.0, 20.0]]) |
| 138 | + |
| 139 | + # State Vector [x y yaw v]' |
| 140 | + xEst = np.matrix(np.zeros((4, 1))) |
| 141 | + xTrue = np.matrix(np.zeros((4, 1))) |
| 142 | + PEst = np.eye(4) |
| 143 | + |
| 144 | + xDR = np.matrix(np.zeros((4, 1))) # Dead reckoning |
| 145 | + |
| 146 | + # history |
| 147 | + hxEst = xEst |
| 148 | + hxTrue = xTrue |
| 149 | + hxDR = xTrue |
| 150 | + |
| 151 | + while SIM_TIME >= time: |
| 152 | + time += DT |
| 153 | + u = calc_input() |
| 154 | + |
| 155 | + xTrue, z, xDR, ud = observation(xTrue, xDR, u, RFID) |
| 156 | + |
| 157 | + xEst, PEst = ekf_slam(xEst, PEst, ud, z) |
| 158 | + |
| 159 | + # store data history |
| 160 | + hxEst = np.hstack((hxEst, xEst)) |
| 161 | + hxDR = np.hstack((hxDR, xDR)) |
| 162 | + hxTrue = np.hstack((hxTrue, xTrue)) |
| 163 | + |
| 164 | + if show_animation: |
| 165 | + plt.cla() |
| 166 | + |
| 167 | + for i in range(len(z[:, 0])): |
| 168 | + plt.plot([xTrue[0, 0], z[i, 1]], [xTrue[1, 0], z[i, 2]], "-k") |
| 169 | + plt.plot(RFID[:, 0], RFID[:, 1], "*k") |
| 170 | + plt.plot(xEst[0], xEst[1], ".r") |
| 171 | + plt.plot(np.array(hxTrue[0, :]).flatten(), |
| 172 | + np.array(hxTrue[1, :]).flatten(), "-b") |
| 173 | + plt.plot(np.array(hxDR[0, :]).flatten(), |
| 174 | + np.array(hxDR[1, :]).flatten(), "-k") |
| 175 | + plt.plot(np.array(hxEst[0, :]).flatten(), |
| 176 | + np.array(hxEst[1, :]).flatten(), "-r") |
| 177 | + plt.axis("equal") |
| 178 | + plt.grid(True) |
| 179 | + plt.pause(0.001) |
| 180 | + |
13 | 181 |
|
14 | 182 | if __name__ == '__main__': |
15 | 183 | main() |
0 commit comments