Adversarial Soft Advantage Fitting: Imitation Learning without Policy Optimization

Abstract

Adversarial Imitation Learning alternates between learning a discriminator -- which tells apart expert's demonstrations from generated ones -- and a generator's policy to produce trajectories that can fool this discriminator. This alternated optimization is known to be delicate in practice since it compounds unstable adversarial training with brittle and sample-inefficient reinforcement learning. We propose to remove the burden of the policy optimization steps by leveraging a novel discriminator formulation. Specifically, our discriminator is explicitly conditioned on two policies: the one from the previous generator's iteration and a learnable policy. When optimized, this discriminator directly learns the optimal generator's policy. Consequently, our discriminator's update solves the generator's optimization problem for free: learning a policy that imitates the expert does not require an additional optimization loop. This formulation effectively cuts by half the implementation and computational burden of Adversarial Imitation Learning algorithms by removing the Reinforcement Learning phase altogether. We show on a variety of tasks that our simpler approach is competitive to prevalent Imitation Learning methods.

References in corpus (13)

• Trust Region Policy Optimization
• Soft Actor-Critic Algorithms and Applications
• Variational Inference with Normalizing Flows
• Guided Cost Learning: Deep Inverse Optimal Control via Policy Optimization
• Reinforcement Learning with Deep Energy-Based Policies
• A Connection between Generative Adversarial Networks, Inverse Reinforcement Learning, and Energy-Based Models
• DualDICE: Behavior-Agnostic Estimation of Discounted Stationary Distribution Corrections
• Provably Efficient Maximum Entropy Exploration
• Discriminator-Actor-Critic: Addressing Sample Inefficiency and Reward Bias in Adversarial Imitation Learning
• Off-Policy Actor-Critic
• SQIL: Imitation Learning via Reinforcement Learning with Sparse Rewards
• Goal-conditioned Imitation Learning
• Imitation Learning via Off-Policy Distribution Matching

Cited by in corpus (4)

• Feedback in Imitation Learning: The Three Regimes of Covariate Shift
• All by Myself: Learning Individualized Competitive Behaviour with a Contrastive Reinforcement Learning optimization
• RAIL: A modular framework for Reinforcement-learning-based Adversarial Imitation Learning
• Of Moments and Matching: A Game-Theoretic Framework for Closing the Imitation Gap