Safety Gym

Safety Gym is a benchmark that contains various tasks designed to evaluate safe reinforcement learning algorithms by measuring their ability to minimize constraint violations while maintaining task performance.

Papers using Safety Gym (3)

Action-Conditioned Risk Gating for Safety-Critical Control under Partial Observability2026Constraint-Aware Reinforcement Learning via Adaptive Action Scaling2025Flipping-based Policy for Chance-Constrained Markov Decision Processes2024