Safe Reinforcement Learning via Shielding

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.

Authors

• Mohammed Alshiekh
• Roderick Bloem
• Rüdiger Ehlers
• Bettina Könighofer Graz University of Technology
• Scott Niekum University of Massachusetts Amherst
• Ufuk Topcu The University of Texas at Austin

Keywords

No keywords are indexed for this paper.