Of Moments and Matching: A Game-Theoretic Framework for Closing the Imitation Gap

Abstract

We provide a unifying view of a large family of previous imitation learning algorithms through the lens of moment matching. At its core, our classification scheme is based on whether the learner attempts to match (1) reward or (2) action-value moments of the expert's behavior, with each option leading to differing algorithmic approaches. By considering adversarially chosen divergences between learner and expert behavior, we are able to derive bounds on policy performance that apply for all algorithms in each of these classes, the first to our knowledge. We also introduce the notion of moment recoverability, implicit in many previous analyses of imitation learning, which allows us to cleanly delineate how well each algorithmic family is able to mitigate compounding errors. We derive three novel algorithm templates (AdVIL, AdRIL, and DAeQuIL) with strong guarantees, simple implementation, and competitive empirical performance.

References in corpus (10)

• Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor
• Large Scale GAN Training for High Fidelity Natural Image Synthesis
• End to End Learning for Self-Driving Cars
• Formal Limitations on the Measurement of Mutual Information
• SQIL: Imitation Learning via Reinforcement Learning with Sparse Rewards
• Primal Wasserstein Imitation Learning
• Imitation Learning as $f$-Divergence Minimization
• Dual Policy Iteration
• Adversarial Soft Advantage Fitting: Imitation Learning without Policy Optimization
• Non-Adversarial Imitation Learning and its Connections to Adversarial Methods