DynVLA: Learning World Dynamics for Action Reasoning in Autonomous Driving
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866917337451462656 |
|---|---|
| author | Shang, Shuyao Zhan, Bing Yan, Yunfei Wang, Yuqi Li, Yingyan An, Yasong Wang, Xiaoman Liu, Jierui Hou, Lu Fan, Lue Zhang, Zhaoxiang Tan, Tieniu |
| author_facet | Shang, Shuyao Zhan, Bing Yan, Yunfei Wang, Yuqi Li, Yingyan An, Yasong Wang, Xiaoman Liu, Jierui Hou, Lu Fan, Lue Zhang, Zhaoxiang Tan, Tieniu |
| contents | We propose DynVLA, a driving VLA model that introduces a new CoT paradigm termed Dynamics CoT. DynVLA forecasts compact world dynamics before action generation, enabling more informed and physically grounded decision-making. To obtain compact dynamics representations, DynVLA introduces a Dynamics Tokenizer that compresses future evolution into a small set of dynamics tokens. Considering the rich environment dynamics in interaction-intensive driving scenarios, DynVLA decouples ego-centric and environment-centric dynamics, yielding more accurate world dynamics modeling. We then train DynVLA to generate dynamics tokens before actions through SFT and RFT, improving decision quality while maintaining latency-efficient inference. Compared to Textual CoT, which lacks fine-grained spatiotemporal understanding, and Visual CoT, which introduces substantial redundancy due to dense image prediction, Dynamics CoT captures the evolution of the world in a compact, interpretable, and efficient form. Extensive experiments on NAVSIM, Bench2Drive, and a large-scale in-house dataset demonstrate that DynVLA consistently outperforms Textual CoT and Visual CoT methods, validating the effectiveness and practical value of Dynamics CoT. Project Page: https://yaoyao-jpg.github.io/dynvla. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_11041 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | DynVLA: Learning World Dynamics for Action Reasoning in Autonomous Driving Shang, Shuyao Zhan, Bing Yan, Yunfei Wang, Yuqi Li, Yingyan An, Yasong Wang, Xiaoman Liu, Jierui Hou, Lu Fan, Lue Zhang, Zhaoxiang Tan, Tieniu Computer Vision and Pattern Recognition Robotics We propose DynVLA, a driving VLA model that introduces a new CoT paradigm termed Dynamics CoT. DynVLA forecasts compact world dynamics before action generation, enabling more informed and physically grounded decision-making. To obtain compact dynamics representations, DynVLA introduces a Dynamics Tokenizer that compresses future evolution into a small set of dynamics tokens. Considering the rich environment dynamics in interaction-intensive driving scenarios, DynVLA decouples ego-centric and environment-centric dynamics, yielding more accurate world dynamics modeling. We then train DynVLA to generate dynamics tokens before actions through SFT and RFT, improving decision quality while maintaining latency-efficient inference. Compared to Textual CoT, which lacks fine-grained spatiotemporal understanding, and Visual CoT, which introduces substantial redundancy due to dense image prediction, Dynamics CoT captures the evolution of the world in a compact, interpretable, and efficient form. Extensive experiments on NAVSIM, Bench2Drive, and a large-scale in-house dataset demonstrate that DynVLA consistently outperforms Textual CoT and Visual CoT methods, validating the effectiveness and practical value of Dynamics CoT. Project Page: https://yaoyao-jpg.github.io/dynvla. |
| title | DynVLA: Learning World Dynamics for Action Reasoning in Autonomous Driving |
| topic | Computer Vision and Pattern Recognition Robotics |
| url | https://arxiv.org/abs/2603.11041 |