Provably Efficient Generative Adversarial Imitation Learning for Online and Offline Setting with Linear Function Approximation

Abstract

In generative adversarial imitation learning (GAIL), the agent aims to learn a policy from an expert demonstration so that its performance cannot be discriminated from the expert policy on a certain predefined reward set. In this paper, we study GAIL in both online and offline settings with linear function approximation, where both the transition and reward function are linear in the feature maps. Besides the expert demonstration, in the online setting the agent can interact with the environment, while in the offline setting the agent only accesses an additional dataset collected by a prior. For online GAIL, we propose an optimistic generative adversarial policy optimization algorithm (OGAP) and prove that OGAP achieves $\tilde{\mathcal{O}}(H^2{d}^{3/2}{K}^{1/2}+K{H}^{3/2}d{N}_1^{-1/2})$ regret. Here $N_1$ represents the number of trajectories of the expert demonstration, $d$ is the feature dimension, and $K$ is the number of episodes. For offline GAIL, we propose a pessimistic generative adversarial policy optimization algorithm (PGAP). For an arbitrary additional dataset, we obtain the optimality gap of PGAP, achieving the minimax lower bound in the utilization of the additional dataset. Assuming sufficient coverage on the additional dataset, we show that PGAP achieves $\tilde{\mathcal{O}}(H^{2}d{K}^{-1/2}+H^2{d}^{3/2}{N}_2^{-1/2}+H^{3/2}d{N}_1^{-1/2})$ optimality gap. Here $N_2$ represents the number of trajectories of the additional dataset with sufficient coverage.

54 pages, in submission

References in corpus (12)

• Conservative Q-Learning for Offline Reinforcement Learning
• Benchmarking Batch Deep Reinforcement Learning Algorithms
• Apprenticeship Learning using Inverse Reinforcement Learning and Gradient Methods
• Model-Based Reinforcement Learning with Value-Targeted Regression
• Information-Theoretic Considerations in Batch Reinforcement Learning
• Provably Good Batch Reinforcement Learning Without Great Exploration
• Information Theoretic Regret Bounds for Online Nonlinear Control
• Off-Policy Evaluation via the Regularized Lagrangian
• Toward the Fundamental Limits of Imitation Learning
• On the Global Convergence of Imitation Learning: A Case for Linear Quadratic Regulator
• Offline Learning from Demonstrations and Unlabeled Experience
• Provably Breaking the Quadratic Error Compounding Barrier in Imitation Learning, Optimally