Work

Shields for Safe Reinforcement Learning

Year: 2025

Type: article

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.

Source: Communications of the ACM

DOI: 10.1145/3715958

Authors: Bettina Könighofer, Roderick Bloem, Nils Jansen, Sebastian Junges, Stefan Pranger

Institutions: Graz University of Technology, Radboud University Nijmegen, Ruhr University Bochum

Language: English

Topic: Reinforcement Learning in Robotics

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: hybrid

Funders: Austrian Science Fund, Nederlandse Organisatie voor Wetenschappelijk Onderzoek, Amt der Steiermärkischen Landesregierung, HORIZON EUROPE European Research Council

Awards: Data-Driven Verification and Learning Under Uncertainty, Logical Methods in Computer Science