Trajectory Prediction on Impromptu (test)
1.34ADE (m)OneVL
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| OneVLModel Size=4B, Backbone=Qwen3-VL, Interpretability=Vision + Language2026.04 | 1.34 | 3.7 | 4.02 | |
| AR CoT+AnswerModel Size=4B, Backbone=Qwen3-VL, Interpretability=Language2026.04 | 1.42 | 3.96 | 6.84 | |
| AR AnswerModel Size=4B, Backbone=Qwen3-VL, Interpretability=-2026.04 | 1.46 | 4.03 | 4.24 | |
| COCONUTModel Size=4B, Backbone=Qwen3-VL, Interpretability=-2026.04 | 1.49 | 4.07 | 5.27 | |
| Impromptu VLAModel Size=3B, Interpretability=-2026.04 | 1.6 | 4.28 | 6.1 | |
| CODIModel Size=4B, Backbone=Qwen3-VL, Interpretability=-2026.04 | 1.86 | 5.18 | 5.24 | |
| SIM-CoTModel Size=4B, Backbone=Qwen3-VL, Interpretability=Language2026.04 | 2.43 | 6.1 | 5.09 |