ObjectNav on Gibson (val)
95.77Success RateGT Distance
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GT DistanceTraining Type=Privileged Training (GT Distance), Noise Level=No noise2025.12 | 95.77 | 61.03 | — | |
| GT DistanceTraining Type=Privileged Training (GT Distance), Noise Model=GeoNoise2025.12 | 90.91 | 55.47 | — | |
| GT DistanceTraining Type=Privileged Training (GT Distance), Noise Model=GeoNoise + Confidence2025.12 | 89.94 | 58.09 | — | |
| GAMapSetting=Zero-shot2026.03 | 85.7 | 55.5 | — | |
| TriHelperSetting=Zero-shot2026.03 | 85.2 | 43.1 | — | |
| RESTSetting=Zero-shot2026.03 | 85.1 | 53.5 | — | |
| VLFMSetting=Zero-shot2026.03 | 84 | 52.2 | — | |
| OVRL-V2Category=External Image-Based Nav Baselines (reported)2025.12 | 82 | 58.7 | — | |
| PONIMethod Type=Modular2022.01 | 73.6 | 41 | 1.25 | |
| VLDDeployment strategy=Swap: Replace Distance with VLD at Deployment, Policy Training Condition=GeoNoise2025.12 | 73.14 | 39.95 | — | |
| SemExpMethod Type=Modular2022.01 | 71.7 | 39.6 | 1.39 | |
| Predict-θMethod Type=Non-interactive2022.01 | 69.9 | 35.7 | 1.44 | |
| VLDDeployment strategy=Swap: Replace Distance with VLD at Deployment, Policy Training Condition=GeoNoise + Confidence2025.12 | 68.21 | 38.6 | — | |
| ANSMethod Type=Modular2022.01 | 67.1 | 34.9 | 1.66 | |
| Predict-xyMethod Type=Non-interactive2022.01 | 66.9 | 34.2 | 1.69 | |
| FBEMethod Type=Modular2022.01 | 64.3 | 28.3 | 1.78 | |
| ViNT-TunedDeployment strategy=Swap: Replace Distance with ViNT at Deployment, Policy Training Condition=GeoNoise2025.12 | 60.46 | 25.55 | — | |
| VLDTraining Mode=Policy trained end-to-end on VLD2025.12 | 56.64 | 42.27 | — | |
| VIP-NavDeployment strategy=Swap: Replace Distance with VIP at Deployment, Policy Training Condition=GeoNoise2025.12 | 27.87 | 11.48 | — | |
| DD-PPOMethod Type=End-to-end RL2022.01 | 15 | 10.7 | 3.24 | |
| Predict-AMethod Type=Non-interactive2022.01 | 14.7 | 13.6 | 3.45 | |
| BCMethod Type=Non-interactive2022.01 | 12.2 | 8.3 | 3.9 |