Shields for Safe Reinforcement Learning
Open Access
- 29 October 2025
- journal article
- Published by Association for Computing Machinery (ACM) in Communications of the ACM
- Vol. 68 (11) , 80-90
Abstract
Reinforcement learning (RL) is a prominent machine learning technique used to optimize an agent’s performance in potentially unknown environments. Despite its popularity and success, RL lacks safety guarantees, both during the learning phase and deployment. This paper reviews a runtime enforcement method called shielding that ensures provable safety for RL. We describe the underlying models, the types of guarantees that can be delivered, and the process of computing shields. Furthermore, we describe several techniques for integrating shields into RL, discuss the advantages and potential drawbacks of this integration, and highlight the current challenges in shielded learning.Keywords
This publication has 23 references indexed in Scilit:
- Provable Correct and Adaptive Simplex Architecture for Bounded-Liveness Properties2023
- TEMPEST - Synthesis Tool for Reactive Systems and Shields in Probabilistic Environments2021
- Neural Simplex Architecture2020
- Run-Time Optimization for Learned Controllers Through Quantitative Games2019
- Handbook of Model Checking2018
- Shield synthesisFormal Methods in System Design, 2017
- Deep Reinforcement Learning: An Overview2017
- Safety-Constrained Reinforcement Learning for MDPs2016
- Shield Synthesis:2015
- Uppaal Stratego2015