ResearchBenchmarksReinforcement Learning on MuJoCo Ant epsilon=0.15 (test)Reinforcement Learning on MuJoCo Ant epsilon=0.15 (test)Follow5,381Natural ReturnPPO4,016.524,370.764,7255,079.24Dec 1, 2025Evaluation resultsMethodMethodLinksNatural ReturnRandom Attack ReturnCritic Attack ReturnMAD Attack ReturnRS Attack ReturnSA-RL Attack ReturnPA-AD Attack ReturnWorst ReturnRobustnessPPO2025.125,3815,3294,6961,7681,097-1,398-3,107-3,107-1.577SA2025.125,3675,2175,0125,1144,3964,2272,3552,355-0.539BAR2025.125,0244,9794,7774,8434,1713,3672,8252,825-0.438RADIAL2025.124,3584,3093,6284,2053,7422,3643,2612,364-0.462WocaR2025.124,0693,9113,9783,6893,1761,8681,8301,830-0.55