Local navigation target prediction on Landmark-RxR full (val)
57.72Geo AccuracyBEACON
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| BEACONInput=RGB-D, Output=BEV heatmap2026.03 | 57.72 | 60.17 | 2.13 | 58.5 | |
| VLM + point headInput=RGB, Output=BEV point2026.03 | 41.25 | 50.15 | 19.81 | 47.5 | |
| RoboRefer-8B-SFTInput=RGB-D, Output=image point2026.03 | 38 | 44.65 | 15.97 | 42.47 | |
| RoboPoint-13B (best point)Input=RGB, Output=image point2026.03 | 35.86 | 46.96 | 27.63 | 45.5 | |
| RoboPoint-13BInput=RGB, Output=image point2026.03 | 20.86 | 32.59 | 39.34 | 30.96 | |
| ChatGPT-4o (oracle-view)Input=RGB, Output=image point2026.03 | 15.97 | 30.79 | 48.2 | 28.28 | |
| ChatGPT-4oInput=RGB, Output=image point2026.03 | 9.69 | 20.65 | 57.25 | 18.94 |