ResearchBenchmarksHopperVelocity on Safety GymFollow93Normalized RewardReference policy (πref)12.9233.7154.575.29May 20, 2026Evaluation ResultsMethodMethodLinksNormalized RewardNormalized CostReference policy (πref)2026.05938.24PREFINE2026.05660.33BC2026.05605.22SafeDICE2026.05512.12PPL2026.05381.19CPQ2026.05162.1