Imitation Learning from Observations by Minimizing Inverse Dynamics Disagreement
This work addresses imitation learning for robotics or AI systems where only observational data (e.g., videos) is available, making it more practical but incremental as it builds on existing LfO methods.
The paper tackles the problem of imitation learning from state-only demonstrations (LfO) by identifying and minimizing the inverse dynamics disagreement between the imitator and expert, which bridges the gap to learning from demonstrations (LfD). It achieves consistent improvements over other LfO methods on challenging benchmarks.
This paper studies Learning from Observations (LfO) for imitation learning with access to state-only demonstrations. In contrast to Learning from Demonstration (LfD) that involves both action and state supervision, LfO is more practical in leveraging previously inapplicable resources (e.g. videos), yet more challenging due to the incomplete expert guidance. In this paper, we investigate LfO and its difference with LfD in both theoretical and practical perspectives. We first prove that the gap between LfD and LfO actually lies in the disagreement of inverse dynamics models between the imitator and the expert, if following the modeling approach of GAIL. More importantly, the upper bound of this gap is revealed by a negative causal entropy which can be minimized in a model-free way. We term our method as Inverse-Dynamics-Disagreement-Minimization (IDDM) which enhances the conventional LfO method through further bridging the gap to LfD. Considerable empirical results on challenging benchmarks indicate that our method attains consistent improvements over other LfO counterparts.