ResearchBenchmarksOffline Reinforcement Learning on Ant Extreme Gravity-ShiftFollow70.8Ant-m (m)TCE(OG)30.34440.84751.3561.853May 13, 2026Evaluation ResultsMethodMethodLinksAnt-m (m)Ant-m (e)Ant-m-r (m)Ant-m-r (e)Ant-m-e (m)Ant-m-e (e)Total ScoreTCE(OG)variant=Only Generated...variant=Only Generated (strictly utilize target-consistent transitions)2026.0570.886.944.370.970.182.5425.5TCE(SM)variant=Source Mixing...variant=Source Mixing (retrieving source transitions)2026.0564.749.341.123.968.656303.6OTDF2026.0534.538.224.823.13533.7189.3IQL*2026.0531.931.318.618.630.131.6162.1