Multi-Agent Reinforcement Learning on SMAC 3s5z map (heterogeneous setup)
61.56Win Rate (%)[4]
Evaluation results
| Method | Links | ||
|---|---|---|---|
| [4]Soft Logit (SL)=Yes, Hard Logit (HL)=No, Relation Loss (IR)=Yes, Coordination Loss (COR)=No, Agent Types=Attention+RNN2026.04 | 61.56 | 31,700,000 | |
| KD-MARLSoft Logit (SL)=Yes, Hard Logit (HL)=Yes, Relation Loss (IR)=Yes, Coordination Loss (COR)=Yes, Agent Types=RNN2026.04 | 58.17 | 13,000,000 | |
| [28]Soft Logit (SL)=Yes, Hard Logit (HL)=Yes, Relation Loss (IR)=No, Coordination Loss (COR)=No, Agent Types=Attention+RNN2026.04 | 54.92 | 30,500,000 | |
| [36]Soft Logit (SL)=Yes, Hard Logit (HL)=No, Relation Loss (IR)=No, Coordination Loss (COR)=No, Agent Types=RNN2026.04 | 46.08 | 11,300,000 |