ResearchBenchmarksSafe Reinforcement Learning on DSRL MediumDenseFollow237.75RewardCDCP46.23495.9545145.675195.3955Jul 4, 2026Evaluation ResultsMethodMethodLinksRewardCostCDCP2026.07237.7518.75CMTDiff2026.07205.6720.55CMTDD2026.0753.618.86