Vision-and-Language Navigation on IR2R (val-seen)
9.4Trajectory Length (TL)TourHAMT
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| TourHAMTPH=true, TH=true, PHI=true, IW=true2024.03 | 9.4 | 5.8 | 56 | 59 | 45 | 43 | 43 | |
| TourHAMTPH=true, TH=true, PHI=true, IW=true2025.10 | 9.4 | 5.8 | — | 59 | 45 | 43 | 45 | |
| OmnigraphPH=false, TH=false, PHI=false, IW=false2024.03 | 9.9 | 3.7 | 70 | 73 | 65 | 63 | 62 | |
| OVER-NAV2025.10 | 9.9 | 3.7 | — | 73 | 65 | 63 | 62 | |
| HAMTPH=false, TH=false, PHI=false, IW=false2024.03 | 10.1 | 4.2 | 70 | 71 | 63 | 61 | 58 | |
| HAMT2025.10 | 10.1 | 4.2 | — | 71 | 63 | 61 | 58 | |
| TourHAMTPH=true, TH=true, PHI=true, IW=false2024.03 | 10.5 | 6 | 60 | 58 | 45 | 43 | 42 | |
| TourHAMTPH=true, TH=true, PHI=true2025.10 | 10.5 | 6 | — | 58 | 45 | 43 | 42 | |
| TourHAMTPH=true, TH=true, PHI=false, IW=false2024.03 | 10.6 | 6 | 61 | 58 | 45 | 42 | 42 | |
| TourHAMTPH=true, TH=true2025.10 | 10.6 | 6 | — | 58 | 45 | 42 | 42 | |
| TourHAMTPH=true, TH=false, PHI=false, IW=false2024.03 | 10.9 | 6.1 | 60 | 58 | 45 | 42 | 41 | |
| TourHAMTPH=true2025.10 | 10.9 | 6.1 | — | 58 | 45 | 42 | 41 | |
| Memoir (w/o retrieval) on DUETPre-training protocol=default protocol, Retrieval mechanism=w/o retrieval2025.10 | 11.2 | 2.3 | — | 81.2 | 79.4 | 75.8 | 72.3 | |
| Memoir (Ours) on DUETPre-training protocol=default protocol2025.10 | 11.5 | 2.6 | — | 78.9 | 77.1 | 72.8 | 68 | |
| Memoir (Ours) on ScaleVLNPre-training protocol=environmental augmentation2025.10 | 11.6 | 2.5 | — | 78.7 | 76.1 | 72.3 | 67.1 | |
| Memoir (Ours) on GR-DUETPre-training protocol=full navigation graph2025.10 | 11.8 | 3 | — | 74.1 | 72.2 | 66.7 | 61.9 | |
| Memoir (w/o retrieval) on ScaleVLNPre-training protocol=environmental augmentation, Retrieval mechanism=w/o retrieval2025.10 | 12.4 | 2.4 | — | 79.2 | 79.1 | 74.1 | 67.4 | |
| DUETPre-training protocol=default protocol2025.10 | 12.5 | 2.2 | — | 79.8 | 79.8 | 74.5 | 69.1 | |
| GR-DUETPre-training protocol=full navigation graph2025.10 | 12.5 | 4.3 | — | 65.2 | 61.1 | 55.1 | 49 | |
| ScaleVLNPre-training protocol=environmental augmentation2025.10 | 12.8 | 2.2 | — | 79.6 | 79.5 | 74.1 | 67 | |
| Memoir (w/o retrieval) on GR-DUETPre-training protocol=full navigation graph, Retrieval mechanism=w/o retrieval2025.10 | 12.9 | 2.8 | — | 75.6 | 76.7 | 70.1 | 63.9 |