Salience-Invariant Consistent Policy Learning for Generalization in Visual Reinforcement Learning

TL;DR

This paper introduces SCPL, a novel framework that enhances visual reinforcement learning generalization by focusing on task-relevant features and maintaining policy consistency across varied observations, leading to significant performance improvements.

Contribution

The paper proposes the Salience-Invariant Consistent Policy Learning (SCPL) algorithm, integrating value and dynamics modules with a policy consistency constraint to improve zero-shot generalization in visual RL.

Findings

SCPL outperforms state-of-the-art methods on multiple benchmarks.

Achieves up to 69% performance improvement in challenging environments.

Demonstrates the effectiveness of salience-guided and policy consistency modules.

Abstract

Generalizing policies to unseen scenarios remains a critical challenge in visual reinforcement learning, where agents often overfit to the specific visual observations of the training environment. In unseen environments, distracting pixels may lead agents to extract representations containing task-irrelevant information. As a result, agents may deviate from the optimal behaviors learned during training, thereby hindering visual generalization.To address this issue, we propose the Salience-Invariant Consistent Policy Learning (SCPL) algorithm, an efficient framework for zero-shot generalization. Our approach introduces a novel value consistency module alongside a dynamics module to effectively capture task-relevant representations. The value consistency module, guided by saliency, ensures the agent focuses on task-relevant pixels in both original and perturbed observations, while the dynamics module uses augmented data to help the encoder capture dynamic- and reward-relevant representations. Additionally, our theoretical analysis highlights the importance of policy consistency for generalization. To strengthen this, we introduce a policy consistency module with a KL divergence constraint to maintain consistent policies across original and perturbed observations.Extensive experiments on the DMC-GB, Robotic Manipulation, and CARLA benchmarks demonstrate that SCPL significantly outperforms state-of-the-art methods in terms of generalization. Notably, SCPL achieves average performance improvements of 14\%, 39\%, and 69\% in the challenging DMC video hard setting, the Robotic hard setting, and the CARLA benchmark, respectively.Project Page: https://sites.google.com/view/scpl-rl.