Vision-Language Navigation on VLN-CE (test)
4.98Navigation Error (NE)StreamVLN
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| StreamVLNSetting=Supervised2026.03 | 4.98 | — | 64.2 | 56.9 | 51.9 | |
| BEVBertSetting=Supervised2026.03 | 5.13 | 13.63 | 57 | 60 | 53.41 | |
| ETPNavSetting=Supervised2026.03 | 5.15 | 11.08 | 58 | 52 | 52.18 | |
| Open-Nav (GPT4)Setting=Zero-Shot, Input=RGB-D, Policy Backbone=LLM/VLM policy, Extra Spatial Prior=No, Waypoint Predictor=Yes, Sensing Regime=Panoramic2026.03 | 6.7 | 7.68 | 23 | 19 | 16.1 | |
| CMASetting=Supervised2026.03 | 6.92 | 11.08 | 45 | 37 | 32.17 | |
| SmartWaySetting=Zero-Shot, Input=RGB-D, Policy Backbone=VLM policy, Extra Spatial Prior=No, Waypoint Predictor=Yes, Sensing Regime=Panoramic2026.03 | 7.01 | 13.09 | 51 | 29 | 22.46 | |
| Open-Nav (Llama3.1)Setting=Zero-Shot, Input=RGB-D, Policy Backbone=LLM/VLM policy, Extra Spatial Prior=No, Waypoint Predictor=Yes, Sensing Regime=Panoramic2026.03 | 7.25 | 8.07 | 23 | 16 | 12.9 | |
| EmergeNavSetting=Zero-Shot, Input=RGB, Policy Backbone=Qwen3-VL-32B, Extra Spatial Prior=No, Waypoint Predictor=No, Sensing Regime=Role-separated2026.03 | 7.6 | 19.22 | 58 | 37 | 21.33 | |
| Fast-SmartWaySetting=Zero-Shot, Input=RGB-D, Policy Backbone=VLM policy, Extra Spatial Prior=No, Waypoint Predictor=No, Sensing Regime=Frontal View2026.03 | 7.72 | 12.56 | — | 27.75 | 24.95 | |
| InstructNav (GPT4)Setting=Zero-Shot, Input=RGB, Policy Backbone=LLM-guided policy, Extra Spatial Prior=Multi-sourced value maps, Waypoint Predictor=No, Sensing Regime=Egocentric2026.03 | 7.74 | — | 6.89 | 31 | 24 | |
| DiscussNav (GPT4)Setting=Zero-Shot, Input=RGB, Policy Backbone=Multi-expert discussion policy, Extra Spatial Prior=No, Waypoint Predictor=No, Sensing Regime=Panoramic2026.03 | 7.77 | 6.27 | 15 | 11 | 10.51 | |
| MapGPT-CE(GPT4o)Setting=Zero-Shot, Input=RGB, Policy Backbone=Map-guided GPT policy, Extra Spatial Prior=Online linguistic map, Waypoint Predictor=No, Sensing Regime=Panoramic2026.03 | 8.16 | 12.63 | 22 | 7 | 5.04 | |
| CA-NavSetting=Zero-Shot, Input=RGB-D, Policy Backbone=GPT/VLM-based, Extra Spatial Prior=Constraint-aware prior, Waypoint Predictor=No, Sensing Regime=Egocentric2026.03 | 8.32 | — | 39.4 | 23 | 11 | |
| EmergeNavSetting=Zero-Shot, Input=RGB, Policy Backbone=Qwen3-VL-8B, Extra Spatial Prior=No, Waypoint Predictor=No, Sensing Regime=Role-separated2026.03 | 8.38 | 19.5 | 48 | 30 | 21.26 |