ResearchBenchmarksSafe Reinforcement Learning on Safety Gymnasium Push (threshold=35)Follow4.77ReturnPPO-Lag0.9221.9212.923.919Mar 24, 2026Evaluation ResultsMethodMethodLinksReturnCostPPO-LagType=OracleType=Oracle2026.034.7736.17PPO-BT2026.034.0941.85PbCRL2026.033.234.86RLSF2026.031.0720.05