Multi-Agent Path Finding on 3-agent 7x7 (evaluation)
46Valid RateQwen3-4B + GRPO + Ours
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Qwen3-4B + GRPO + OursBackbone=Qwen3-4B, Training=GRPO + Adaptive-weight reward2026.06 | 46 | 26 | |
| Qwen3-4B + GRPO (random)Backbone=Qwen3-4B, Training=GRPO (random)2026.06 | 38.67 | 21.33 | |
| Kimi-K2.52026.06 | 38 | 22.67 | |
| Grok-4.22026.06 | 29.33 | 12 | |
| GPT-5.42026.06 | 20.67 | 12.67 | |
| Gemini-3.1-Pro2026.06 | 18.67 | 9.33 | |
| Qwen3-4B (base)Backbone=Qwen3-4B2026.06 | 8 | 8 |