Research map / Reinforcement learning / Visual and representation RL
Safety Gymnasium: A Unified Safe Reinforcement Learning Benchmark
Jiaming Ji, Borong Zhang, Jiayi Zhou, Xuehai Pan, Weidong Huang, Ruiyang Sun, Yiran Geng, Yifan Zhong, Josef Dai, Yaodong Yang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Lagrangian constraint handling to PPO | PPO · Velocity tasks | episodic cost / reward: 98% cost reduction, 45% reward decrease | Mixed |
| add a PID controller to Lagrangian multiplier update | PPO-Lag · Navigation tasks | safety (fewer Strongly Unsafe states) | Improved |
| use projection-based PCPO instead of CPO | CPO · Navigation and velocity tasks | average cost and reward | Hurt |
| use CPO instead of PPO-Lag (projection vs Lagrangian) | constrained policy optimization · training cost curves | oscillation around cost limit | Improved |
| use SafePO implementation instead of Safety Starter Agents | CPO · 10 safety envs | AvgR/AvgC: 0.27 -> 0.56 | Improved |
| use SafePO implementation instead of Safety Starter Agents | TRPO-Lag · 10 safety envs | AvgR/AvgC: 0.40 -> 0.51 | Improved |
| use SafePO implementation instead of Safety Starter Agents | PPO-Lag · 10 safety envs | AvgR/AvgC: 0.41 -> 0.64 | Improved |
| use SafePO implementation instead of original implementation | FOCOPS · 10 safety envs | AvgR/AvgC: 0.39 -> 0.52 | Improved |
| use safe MARL (MAPPO-Lag) instead of MAPPO | MAPPO · multi-agent velocity tasks | normalized cost: 39.12 -> 0.00 (2x4AntVel) | Improved |
Closest papers by meaning
- Safety Representations for Safer Policy Learning (ICLR 2025 · similarity 0.90)
- Balance Reward and Safety Optimization for Safe Reinforcement Learning: A Perspective of Gradient Manipulation (AAAI 2024 · similarity 0.90)
- SafeDreamer: Safe Reinforcement Learning with World Models (ICLR 2024 · similarity 0.90)
- Robust Gymnasium: A Unified Modular Benchmark for Robust Reinforcement Learning (ICLR 2025 · similarity 0.89)
- Enhancing Efficiency of Safe Reinforcement Learning via Sample Manipulation (NeurIPS 2024 · similarity 0.89)
- Long-Term Safe Reinforcement Learning with Binary Feedback (AAAI 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
