Robust Safe Reinforcement Learning under Adversarial Disturbances

TL;DR

This paper introduces a robust safe reinforcement learning framework that ensures safety under worst-case external disturbances by combining Hamilton-Jacobi reachability analysis with a policy iteration scheme, improving safety guarantees in control tasks.

Contribution

It proposes a novel policy iteration algorithm for computing the maximal robust invariant set and integrates it into a constrained RL method for safety under adversarial disturbances.

Findings

Achieves zero safety constraint violations under worst-case disturbances.

Maintains high reward performance comparable to baseline methods.

Ensures safety even without adversarial disturbances.

Abstract

Safety is a primary concern when applying reinforcement learning to real-world control tasks, especially in the presence of external disturbances. However, existing safe reinforcement learning algorithms rarely account for external disturbances, limiting their applicability and robustness in practice. To address this challenge, this paper proposes a robust safe reinforcement learning framework that tackles worst-case disturbances. First, this paper presents a policy iteration scheme to solve for the robust invariant set, i.e., a subset of the safe set, where persistent safety is only possible for states within. The key idea is to establish a two-player zero-sum game by leveraging the safety value function in Hamilton-Jacobi reachability analysis, in which the protagonist (i.e., control inputs) aims to maintain safety and the adversary (i.e., external disturbances) tries to break down safety. This paper proves that the proposed policy iteration algorithm converges monotonically to the maximal robust invariant set. Second, this paper integrates the proposed policy iteration scheme into a constrained reinforcement learning algorithm that simultaneously synthesizes the robust invariant set and uses it for constrained policy optimization. This algorithm tackles both optimality and safety, i.e., learning a policy that attains high rewards while maintaining safety under worst-case disturbances. Experiments on classic control tasks show that the proposed method achieves zero constraint violation with learned worst-case adversarial disturbances, while other baseline algorithms violate the safety constraints substantially. Our proposed method also attains comparable performance as the baselines even in the absence of the adversary.