基于简化概率选择框架的双足机器人模仿学习

薛雯,赵硕,李永强

Imitation learning for bipedal robots based on simplified probabilistic framework for options

Wen XUE,Shuo ZHAO,Yongqiang LI

表 2 在典型连续动作控制任务中不同模仿学习方法的平均回报与标准差

Tab.2 Mean return and standard deviation of different imitation learning methods in typical continuous action control tasks

方法 $ \overline{R}\pm \sigma $ 类人机器 人行走 机械臂推物 二维行走 机器人 游动机器人 双足行走 机器人 类人机器 人起身 四足蚂蚁 机器人 单腿跳跃 机器人 双足猎豹 模型 GAIL-Option −1701.46± 1109.85 −636.24 ± 239.12 509.71 ± 132.52 12.38 ± 41.99 257.79 ± 10.60 360.16 ± 198.64 1958.55 ± 226.74 489.35 ± 74.15 1924.50 ± 217.21 SPFFO −158.63 ± 48.26 −576.64 ± 33.26 1217.23 ± 756.50 −3.97 ± 7.60 −1.74 ± 7.84 275.29 ± 40.56 1027.70 ± 256.64 553.32 ± 279.83 2390.93 ± 358.77 PFFO 498.22 ± 28.01 −391.40 ± 14.19 333.87 ± 132.80 41.52 ± 6.75 203.25 ± 86.42 885.02 ± 25.37 2251.33 ± 214.70 329.40 ± 194.13 1898.57 ± 574.98 本研究 517.44 ± 46.11 −384.08 ± 24.51 2296.74 ± 250.19 45.94 ± 1.95 264.40 ± 16.82 915.27 ± 48.67 2407.19 ± 106.10 704.11 ± 61.90 2871.36 ± 155.93 DVL 153.56 ± 63.20 −44.11 ± 0.12 192.49 ± 175.22 31.01 ± 3.05 −95.51 ± 18.47 812.97 ± 298.35 2010.40 ± 684.32 76.29 ± 43.06 340.72 ± 326.62 ISWBC −536.76 ± 594.76 −387.42 ± 13.04 1593.60 ± 446.14 42.98 ± 5.65 256.99 ± 31.92 −173.89 ± 146.13 2412.23 ± 72.90 488.13 ± 79.01 2738.28 ± 103.02 HIPS 531.89 ± 34.90 −384.13 ± 17.24 1292.01 ± 189.64 25.08 ± 16.21 244.24 ± 29.73 865.51 ± 18.63 2338.80 ± 88.94 431.03 ± 69.39 1699.79 ± 187.79