Trajectory Prediction on NAVSIM supervised fine-tuning setting
88.84PDM ScoreOneVL
Evaluation results
| Method | Links | ||
|---|---|---|---|
| OneVLModel Size=4B, Backbone=Qwen3-VL, Interpretability=Vision + Language2026.04 | 88.84 | 4.46 | |
| AR CoT+AnswerModel Size=4B, Backbone=Qwen3-VL, Interpretability=Language2026.04 | 88.29 | 6.58 | |
| AR AnswerModel Size=4B, Backbone=Qwen3-VL2026.04 | 87.47 | 4.49 | |
| LaST-VLAModel Size=8B2026.04 | 87.3 | — | |
| AdaThinkDriveModel Size=8B, Interpretability=Language2026.04 | 86.2 | — | |
| COCONUTModel Size=4B, Backbone=Qwen3-VL2026.04 | 84.84 | 5.93 | |
| SIM-CoTModel Size=4B, Backbone=Qwen3-VL, Interpretability=Language2026.04 | 84.21 | 10.86 | |
| CODIModel Size=4B, Backbone=Qwen3-VL2026.04 | 83.92 | 8.62 |