ResearchBenchmarksWalker2dVelocity on Safety GymFollow93Normalized RewardReference policy (πref)17.0836.7956.576.21May 20, 2026Evaluation ResultsMethodMethodLinksNormalized RewardNormalized CostReference policy (πref)2026.05939.81PREFINE2026.05750.03SafeDICE2026.05680.29BC2026.05653.21PPL2026.05610.39CPQ2026.05200.04