Learning Plan-Satisficing Motion Policies from Demonstrations

Yanwei Wang

Keywords: Task and Motion Imitation, Certifiable Imitation Learning

Abstract: Learning from demonstration (LfD) methods have shown promise for solving multi-step tasks; however, these approaches do not guarantee successful reproduction of the task given perturbations. In this work, we identify the roots of such a challenge as the failure of the learned continuous policy to satisfy the discrete plan implicit in the demonstration. By utilizing modes (rather than subgoals) as the discrete abstraction and motion policies with both mode invariance and goal reachability properties, we show our learned continuous policy can simulate any given discrete plan. Consequently, the imitator is robust to both task- and motion-level perturbations and guaranteed to achieve task success.

Submission Number: 2