Visual Navigation on AI2-THOR Unseen Scenes (L >= 5) (test)
13.91SPLSAVN
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SAVNLoss type=learned self-supervised loss2018.12 | 13.91 | 2,870 | |
| SAVN (diversity)Loss type=hand-crafted diversity loss2018.12 | 13.38 | 2,766 | |
| SAVN (prediction)Loss type=hand-crafted prediction loss2018.12 | 12.61 | 2,641 | |
| No adaptation (A3C)interaction-gradient=false2018.12 | 11.69 | 2,144 | |
| Scene Priors2018.12 | 11.37 | 2,225 | |
| Nearest neighbor (NN)Backbone=ResNet features2018.12 | 1.38 | 166 | |
| Visual Navigation with Spatial AttentionArchitecture=A2C2021.04 | 0.1594 | 32.63 | |
| Visual Navigation with Spatial AttentionArchitecture=A3C2021.04 | 0.1551 | 31.71 | |
| SAVN2021.04 | 0.1391 | 28.7 | |
| Scene Prior2021.04 | 0.1137 | 22.25 | |
| Random agent2018.12 | 0.1 | 28 |