ResearchBenchmarksSafe Reinforcement Learning on DSRL Average across 9 tasksFollow270.45RewardCDCP55.2428111.1139166.985222.8561Jul 4, 2026Evaluation ResultsMethodMethodLinksRewardCostCDCP2026.07270.4523.06CMTDiff2026.07246.5730.26CMTDD2026.0763.5221.53