ResearchBenchmarksSingle-Agent Spatial Puzzles on Sokoban (In-domain)Follow77.3Success RateMAGE31.85243.65155.4567.249Mar 4, 2026Evaluation ResultsMethodMethodLinksSuccess RateMAGECategory=Meta-RLCategory=Meta-RL2026.0377.3GiGPOCategory=RLCategory=RL2026.0371.9LAMERCategory=Meta-RLCategory=Meta-RL2026.0368.8GRPOCategory=RLCategory=RL2026.0360.2ReflexionCategory=In-Context Le...Category=In-Context Learning2026.0343.8ReActCategory=In-Context Le...Category=In-Context Learning2026.0338.3A-MEMCategory=Memory-basedCategory=Memory-based2026.0336.7MementoCategory=Memory-basedCategory=Memory-based2026.0333.6