Multi-agent coordination on Maze Structure Map
81.15Final Cumulative Win Rate (FW)MAPPO
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| MAPPOTraining steps budget=2.0M2026.03 | 81.15 | 81.15 | 2.01 | 3.1 | |
| EuclidTraining steps budget=2.0M2026.03 | 77.89 | 77.89 | 1.99 | 6.3 | |
| SOG(Vision)Training steps budget=2.0M2026.03 | 72.73 | 72.73 | 1.99 | 15 | |
| QMIXTraining steps budget=2.0M2026.03 | 71.98 | 71.98 | 2 | 3.1 | |
| CommFormerTraining steps budget=2.0M2026.03 | 70.4 | 70.4 | 1.99 | 15.1 | |
| DPPTraining steps budget=2.0M2026.03 | 69.63 | 69.63 | 2 | 15.2 | |
| SOG(RL-Vision)Training steps budget=2.0M2026.03 | 62.68 | 62.68 | 2 | 3.8 |