TCRL: Temporal-Coupled Adversarial Training for Robust Constrained Reinforcement Learning in Worst-Case Scenarios

TL;DR

This paper introduces TCRL, a novel adversarial training framework that enhances the robustness of constrained reinforcement learning policies against temporally coupled perturbations in worst-case scenarios.

Contribution

TCRL is the first to explicitly model and defend against temporally coupled adversarial perturbations in constrained reinforcement learning.

Findings

TCRL outperforms existing methods in robustness against temporally coupled attacks.

TCRL maintains reward unpredictability while ensuring safety under worst-case perturbations.

Experimental results validate TCRL's effectiveness across various CRL tasks.

Abstract

Constrained Reinforcement Learning (CRL) aims to optimize decision-making policies under constraint conditions, making it highly applicable to safety-critical domains such as autonomous driving, robotics, and power grid management. However, existing robust CRL approaches predominantly focus on single-step perturbations and temporally independent adversarial models, lacking explicit modeling of robustness against temporally coupled perturbations. To tackle these challenges, we propose TCRL, a novel temporal-coupled adversarial training framework for robust constrained reinforcement learning (TCRL) in worst-case scenarios. First, TCRL introduces a worst-case-perceived cost constraint function that estimates safety costs under temporally coupled perturbations without the need to explicitly model adversarial attackers. Second, TCRL establishes a dual-constraint defense mechanism on the reward to counter temporally coupled adversaries while maintaining reward unpredictability. Experimental results demonstrate that TCRL consistently outperforms existing methods in terms of robustness against temporally coupled perturbation attacks across a variety of CRL tasks.