ResearchBenchmarksSafe Reinforcement Learning on DSRL MediumSparseFollow242.75RewardCDCP50.818100.6465150.475200.3035Jul 4, 2026Evaluation ResultsMethodMethodLinksRewardCostCDCP2026.07242.7520.68CMTDiff2026.07214.6222.44CMTDD2026.0758.221.27