ResearchBenchmarksReinforcement Learning on MuJoCo Walker2d epsilon=0.05 (test)Follow4,875Natural ReturnSA2,437.243,070.123,7034,335.88Dec 1, 2025Evaluation ResultsMethodMethodLinksNatural ReturnReturn (Random Attack)Return (Critic Attack)Return (MAD Attack)Return (RS Attack)Return (SA-RL Attack)Return (PA-AD Attack)Worst ReturnRobustness ScoreSA2025.124,8754,9075,0294,8332,7753,356997997-0.795BAR2025.124,7324,7184,7274,7504,6464,2852,6992,699-0.436PPO2025.124,6624,6284,5844,5071,062719336336-0.928WocaR2025.124,2264,3474,3424,3733,3582,3851,0641,064-0.752RADIAL2025.122,5312,1702,0632,3161,2394261,353426-0.832