Signage-based navigation on LSI-Dataset unseen (val)
80Success RateSTART
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| STARTBackbone=ViT-B/32, Hint Encoder=ResNet-50, Architecture=Spatial-Temporal Aware Transformer2026.03 | 80 | 74 | 67 | |
| ViNTTraining=retrained on LSI-Dataset, Modification=replaced sub-goal encoder with hint image encoder2026.03 | 68 | 69 | 57 | |
| Rule-basedMapping=hand-crafted rules, Requirements=online odometry2026.03 | 67 | 73 | 56 | |
| VLM-basedModel=Qwen32026.03 | 30 | 54 | 25 |