ResearchBenchmarksSafe Reinforcement Learning on DSRL HardSparseFollow257.98RewardCDCP59.7248111.1949162.665214.1351Jul 4, 2026Evaluation ResultsMethodMethodLinksRewardCostCDCP2026.07257.9824.94CMTDiff2026.07227.9431.06CMTDD2026.0767.3521.9