ResearchBenchmarksReinforcement Learning on MuJoCo Halfcheetah epsilon=0.15 (test)Reinforcement Learning on MuJoCo Halfcheetah epsilon=0.15 (test)Follow5,048Natural ReturnPPO4,7104,797.754,885.54,973.25Dec 1, 2025Evaluation resultsMethodMethodLinksNatural ReturnRandom Attack ReturnCritic Attack ReturnMAD Attack ReturnRS Attack ReturnSA-RL Attack ReturnPA-AD Attack ReturnWorst ReturnRobustness ScorePPO2025.125,0484,4633,2819181,049-213-69-213-1.042BAR2025.124,8374,7414,8034,7294,2653,8943,1813,181-0.342SA2025.124,7804,9835,0353,7592,7271,4431,5511,443-0.698RADIAL2025.124,7394,6424,5462,9611,3271,5221,9681,327-0.72WocaR2025.124,7234,7984,8464,5433,3022,2702,4982,270-0.519