Reasoning on Navigation
100Success RateWith parent pointer
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| With parent pointerTraining=GRPO2026.05 | 100 | 14.28 | |
| BFSTraining=pretrained2026.05 | 100 | 20.27 | |
| BFSTraining=RL2026.05 | 100 | 16.99 | |
| With parent pointerTraining=SFT2026.05 | 95.2 | 15.63 | |
| No parent pointerTraining=GRPO2026.05 | 94.9 | 14.8 | |
| No parent pointerTraining=SFT2026.05 | 90.6 | 15.39 |