ResearchBenchmarksCarGoal on Safety GymFollow0.44Normalized RewardReference policy (πref)-0.0280.09350.2150.3365May 20, 2026Evaluation ResultsMethodMethodLinksNormalized RewardNormalized CostReference policy (πref)2026.050.443.69PREFINE2026.050.380.3SafeDICE2026.050.320.44BC2026.050.280.31CPQ2026.050.280.2PPL2026.05-0.010.4