Waypoint Regression on CosFly-Track (evaluation)
95.6Success Rate @ 1mQwen3.5-9B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen3.5-9BParams=9B, Evaluation Protocol=SFT fine-tuned on CosFly-Track2026.05 | 95.6 | 1.213 | 95.86 | 0.268 | |
| GLM-4.6V-FlashParams=4.6B, Evaluation Protocol=SFT fine-tuned on CosFly-Track2026.05 | 95.48 | 1.231 | 95.77 | 0.274 | |
| Qwen3-VL-8BParams=8B, Evaluation Protocol=SFT fine-tuned on CosFly-Track2026.05 | 95.22 | 1.247 | 91.88 | 0.312 | |
| Qwen3.5-2BParams=2B, Evaluation Protocol=SFT fine-tuned on CosFly-Track2026.05 | 95.08 | 1.271 | 93.12 | 0.298 | |
| Qwen3-VL-2BParams=2B, Evaluation Protocol=SFT fine-tuned on CosFly-Track2026.05 | 94.8 | 1.315 | 88.45 | 0.352 | |
| Qwen3.5-0.8BParams=0.8B, Evaluation Protocol=SFT fine-tuned on CosFly-Track2026.05 | 94.5 | 1.298 | 87.61 | 0.334 | |
| Gemma-4-E4BParams=4B, Evaluation Protocol=SFT fine-tuned on CosFly-Track2026.05 | 78.34 | 1.527 | 72.33 | 0.461 | |
| Qwen3-VL-8BParams=8B, Evaluation Protocol=Zero-shot (no fine-tuning)2026.05 | 33.47 | 2.601 | 40.78 | 1.527 | |
| Qwen3.5-0.8BParams=0.8B, Evaluation Protocol=Zero-shot (no fine-tuning)2026.05 | 25.17 | 2.872 | 40.41 | 1.731 | |
| Qwen3.5-2BParams=2B, Evaluation Protocol=Zero-shot (no fine-tuning)2026.05 | 25.17 | 2.872 | 40.41 | 1.731 | |
| Qwen3-VL-2BParams=2B, Evaluation Protocol=Zero-shot (no fine-tuning)2026.05 | 25.17 | 2.872 | 40.41 | 1.731 |