ResearchBenchmarksMaze Navigation on maze2d umaze v1Follow161.6D4RL ScoreInverter129.36137.73146.1154.47May 22, 2026Evaluation ResultsMethodMethodLinksD4RL ScoreNN Forward Passes per EpisodeTotal Wall Time per Episode (ms)InverterK (one-shot configurat...K (one-shot configuration)=128, Hardware=single A40 GPU, Software framework=PyTorch, Batch size=1, CUDA-synced=true2026.05161.6311.4SAC-N2026.05130.6——BC-10%Software framework=PyT...Software framework=PyTorch2026.05—300374.9