Model-Based Imitation Learning Using Entropy Regularization of Model and Policy

Approaches based on generative adversarial networks for imitation learning are promising because they sample efficient in terms of expert demonstrations. However, training a generator requires many interactions with the actual environment model-free reinforcement is adopted to update policy. To improve efficiency using model-based learning, we propose Entropy-Regularized Imitation Learning (MB-ERIL) under entropy-regularized Markov decision process reduce number environment. MB-ERIL uses two discriminators. A policy discriminator distinguishes actions generated by robot from ones, and model counterfactual state transitions ones. We derive structured discriminators so that efficient. Computer simulations real experiments show achieves competitive performance significantly improves compared baseline methods.