Generative Adversarial Imitation Learning

← Back to document page

Generative Adversarial Imitation LearningJonathan ErmonStanford Learning a policy from example expert behavior, without interaction withthe expert or access to a reinforcement signal. One approach is to recover theexpert s cost function with inverse reinforcement Learning , then extract a policyfrom that cost function with reinforcement Learning . This approach is indirectand can be slow. We propose a new general framework for directly extracting apolicy from data as if it were obtained by reinforcement Learning following inversereinforcement Learning . We show that a certain instantiation of our frameworkdraws an analogy between Imitation Learning and Generative Adversarial networks,from which we derive a model-free Imitation Learning algorithm that obtains signif-icant performance gains over existing model-free methods in imitating complexbehaviors in large, high-dimensional Introduct

networks [8], a technique from the deep learning community that has led to recent successes in modeling distributions of natural images: our algorithm harnesses generative adversarial training to fit distributions of states and actions defining expert behavior. We test our algorithm in Section 6, where

  Network, Learning, Adversarial, Generative, Imitation, Generative adversarial, Generative adversarial imitation learning