ResearchBenchmarksSafe Reinforcement Learning on Safety Gymnasium Humanoid (threshold=5)Safe Reinforcement Learning on Safety Gymnasium Humanoid (threshold=5)Follow5,948ReturnPPO-BT4,770.725,076.365,3825,687.64Mar 24, 2026Evaluation resultsMethodMethodLinksReturnCostPPO-BT2026.035,9485.89PPO-LagType=OracleType=Oracle2026.035,8765.03PbCRL2026.035,2474.85RLSF2026.034,8164.05