Convergence (economics)

Engineering

Artificial intelligence

Economics

Temporal logic

Physics

Electromagnetic shielding

Economic growth

Petrology

Geology

Reinforcement learning

Shield

Quantum mechanics

Action (physics)

Computer science

Theoretical computer science

Electrical engineering

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.

Safe Reinforcement Learning via Shielding