Safe Reinforcement Learning via Shielding

• DOI: 10.1609/aaai.v32i1.11797
• Publication Date: 2022-06-21T20:48:40Z
• Authors: Mohammed Alshiekh; Roderick Bloem; Rüdiger Ehlers; Bettina Könighofer; Scott Niekum; Ufuk Topcu
• Keywords: FOS: Computer and information sciences; Computer Science - Logic in Computer Science; Computer Science - Machine Learning; 0209 industrial biotechnology; Artificial Intelligence (cs.AI); Computer Science - Artificial Intelligence; 02 engineering and technology; Logic in Computer Science (cs.LO); Machine Learning (cs.LG)

Reinforcement learning algorithms discover policies that maximize reward, but do not necessarily guarantee safety during learning or execution phases. We introduce a new approach to learn optimal policies while enforcing properties expressed in temporal logic. To this end, given the temporal logic specification that is to be obeyed by the learning system, we propose to synthesize a reactive system called a shield. The shield monitors the actions from the learner and corrects them only if the chosen action causes a violation of the specification. We discuss which requirements a shield must meet to preserve the convergence guarantees of the learner. Finally, we demonstrate the versatility of our approach on several challenging reinforcement learning scenarios.