Temporal Logic Imitation: Learning Plan-Satisficing Motion Policies from Demonstrations
This addresses the problem of unreliable task reproduction in LfD for robotics under perturbations, offering a guaranteed solution, though it appears incremental as it builds on existing LfD and LTL methods.
The paper tackles the challenge of learning from demonstration (LfD) in tasks with human-in-the-loop perturbations by ensuring learned continuous policies satisfy discrete plans, proving that their policy can simulate any linear temporal logic (LTL) formula and guaranteeing task success with robustness to perturbations.
Learning from demonstration (LfD) has succeeded in tasks featuring a long time horizon. However, when the problem complexity also includes human-in-the-loop perturbations, state-of-the-art approaches do not guarantee the successful reproduction of a task. In this work, we identify the roots of this challenge as the failure of a 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 prove our learned continuous policy can simulate any discrete plan specified by a linear temporal logic (LTL) formula. Consequently, an imitator is robust to both task- and motion-level perturbations and guaranteed to achieve task success. Project page: https://yanweiw.github.io/tli/