Vision-Language Navigation on HA-VLN Retrained (val-seen)
3.93NE (w/o human)Rec (PREVALENT)
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Rec (PREVALENT)Training=Retrained2024.06 | 3.93 | 63 | 4.95 | 41 | 25.9 | -34.9 | |
| AirbertTraining=Retrained2024.06 | 4.01 | 62 | 3.98 | 50 | -0.7 | -19.4 | |
| Rec (OSCAR)Training=Retrained2024.06 | 4.29 | 59 | 4.67 | 42 | 8.9 | -28.8 | |
| DRAEhuman_presence=w/ human2025.07 | 5.1 | 58 | — | — | — | — | |
| DRAERetrained=true2025.07 | 5.3 | 52 | 5.1 | 58 | -3.8 | 11.5 | |
| DRAEhuman_presence=w/o human2025.07 | 5.3 | 52 | — | — | — | — | |
| Speaker-FollowerTraining=Retrained2024.06 | 6.62 | 35 | 5.58 | 34 | -15.7 | -2.9 | |
| DRAEmeasurement=difference (w/ human vs w/o human)2025.07 | — | — | — | — | -380 | 1,150 |