Adversarial Inverse Optimal Control for General Imitation Learning Losses and Embodiment Transfer

Abstract

We develop a general framework for inverse optimal control that distinguishes between rationalizing demonstrated behavior and imitating inductively inferred behavior. This enables learning for more general imitative evaluation measures and differences between the capabilities of the demonstrator and those of the learner (i.e., differences in embodiment). Our formulation takes the form of a zero-sum game between a predictor attempting to minimize an imitative loss measure, and an adversary attempting to maximize the loss by approximating the demonstrated examples in limited ways. We establish the consistency and generalization guarantees of this approach and il- lustrate its benefits on real and synthetic imitation learning tasks.

Cite

Text

Chen et al. "Adversarial Inverse Optimal Control for General Imitation Learning Losses and Embodiment Transfer." Conference on Uncertainty in Artificial Intelligence, 2016.

Markdown

[Chen et al. "Adversarial Inverse Optimal Control for General Imitation Learning Losses and Embodiment Transfer." Conference on Uncertainty in Artificial Intelligence, 2016.](https://mlanthology.org/uai/2016/chen2016uai-adversarial/)

BibTeX

@inproceedings{chen2016uai-adversarial,
  title     = {{Adversarial Inverse Optimal Control for General Imitation Learning Losses and Embodiment Transfer}},
  author    = {Chen, Xiangli and Monfort, Mathew and Ziebart, Brian D. and Carr, Peter},
  booktitle = {Conference on Uncertainty in Artificial Intelligence},
  year      = {2016},
  url       = {https://mlanthology.org/uai/2016/chen2016uai-adversarial/}
}