Counterexample-Guided Repair of Reinforcement Learning Systems Using Safety Critics
This addresses safety issues in reinforcement learning systems, which is crucial for real-world applications, but it appears incremental as it builds on existing repair and safety critic methods.
The paper tackles the problem of unsafe behavior in trained Deep Reinforcement Learning agents by proposing a counterexample-guided repair algorithm that uses safety critics and gradient-based constrained optimization to fix the agents without retraining.
Naively trained Deep Reinforcement Learning agents may fail to satisfy vital safety constraints. To avoid costly retraining, we may desire to repair a previously trained reinforcement learning agent to obviate unsafe behaviour. We devise a counterexample-guided repair algorithm for repairing reinforcement learning systems leveraging safety critics. The algorithm jointly repairs a reinforcement learning agent and a safety critic using gradient-based constrained optimisation.