Vulnerability Analysis for Safe Reinforcement Learning in Cyber-Physical Systems
Open Access
- 21 April 2026
- journal article
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Cyber-Physical Systems
- Vol. 10 (2) , 1-23
Abstract
Safe Reinforcement Learning (RL) has been applied to synthesize control policies that maximize task rewards while adhering to safety constraints within simulated secure cyber-physical systems. However, the vulnerability of safe RL to adversarial attacks remains largely unexplored. We argue that understanding the safety vulnerabilities of learned control policies is crucial for ensuring true safety in real-world scenarios. To address this gap, we first formally define the safe RL problem with formal language (signal temporal logic) and demonstrate that even optimal policies are susceptible to observation perturbations. We then introduce novel safety violation attacks that exploit adversarial models trained with reversed safety constraints to induce unsafe behaviors. Lastly, through both theoretical analysis and experimental results, we demonstrate that our approach is more effective at violating safety constraints than existing adversarial RL methods, which primarily focus on reducing task rewards rather than compromising safety.Keywords
Funding Information
- National Sleep Foundation (CNS-2442914 and CNS-2333980)
This publication has 31 references indexed in Scilit:
- Joint Differentiable Optimization and Verification for Certified Reinforcement Learning2023
- An STL-based Approach to Resilient Control for Cyber-Physical Systems2023
- Learning From Demonstrations Using Signal Temporal Logic in Stochastic and Continuous DomainsIEEE Robotics and Automation Letters, 2021
- Malicious Attacks against Deep Reinforcement Learning Interpretations2020
- TrojDRL: Evaluation of Backdoor Attacks on Deep Reinforcement Learning2020
- Adversarial Sensor Attack on LiDAR-based Perception in Autonomous Driving2019
- LTL and Beyond: Formal Languages for Reward Function Specification in Reinforcement Learning2019
- One Pixel Attack for Fooling Deep Neural NetworksIEEE Transactions on Evolutionary Computation, 2019
- Multiple shooting, CEGAR-based falsification for hybrid systems2014
- Probabilistic Temporal Logic Falsification of Cyber-Physical SystemsACM Transactions on Embedded Computing Systems, 2013