具身智能观察

Counterfactual Behavior Cloning: Offline Imitation Learning from Imperfect Human Demonstrations

技术动态

来源:arXiv cs.RO发布时间待核实

arXiv:2505.10760v2 Announce Type: replace Abstract: Learning from humans is challenging because people are imperfect teachers. When everyday humans show the robot a new task they want it to perform, humans inevitably make errors (e.g., inputting noisy actions) and provide suboptimal examples (e.g., overshooting the goal).

Counterfactual Behavior Cloning: Offline Imitation Learning from Imperfect Human Demonstrations | 具身智能观察