ResearchBenchmarksContinual Reinforcement Learning on Atari-20Continual Reinforcement Learning on Atari-20Follow42.7FT GainAMC0.47611.43822.433.362Apr 2, 2026Evaluation resultsMethodMethodLinksFT GainAP GainForgetting ReductionAP ScoreTasks Beating HumanAMCbaseline_comparison=st...baseline_comparison=strongest per-benchmark baseline2026.0442.77.522——AMC2026.0422.4—-9.720180PackNet2026.0415.7—-12.418770PER2026.048.3—-38.614245EWC2026.045.2—-41.213640VR2026.042.1—-47.312435