All Tools
S
AgentsFreeOpen Source
SAFETY-GYMNASIUM
Unified benchmark for safe reinforcement learning
Apache-2.0
ABOUT
Safe RL needs shared tasks that report costs, not only reward, or constraint methods cannot be compared. Safety-Gymnasium provides Gymnasium environments with safety costs so researchers can train constrained policies on a common NeurIPS 2023 benchmark.
INSTALL
pip install safety-gymnasiumINTEGRATION GUIDE
1. Train a constrained PPO agent that respects Safety-Gymnasium cost limits
2. Benchmark safe RL algorithms on navigation and locomotion constraint tasks
3. Compare constraint-satisfaction methods on a shared Gymnasium API
4. Evaluate whether a policy stays inside safety-critical regions during rollout
TAGS
pythonreinforcement-learningsafetygymnasiumconstraintsbenchmark