ResearchBenchmarksSafe Reinforcement Learning on Safety Gymnasium HalfCheetah (threshold=5)Follow2,619ReturnPPO-Lag2,062.62,207.052,351.52,495.95Mar 24, 2026Evaluation ResultsMethodMethodLinksReturnCostPPO-LagType=OracleType=Oracle2026.032,6194.82PPO-BT2026.032,4945.58PbCRL2026.032,3674.66RLSF2026.032,0843.26