Vega: Learning to Drive with Natural Language Instructions
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_ | 1866912987462238208 |
|---|---|
| author | Zuo, Sicheng Li, Yuxuan Zheng, Wenzhao Zhu, Zheng Zhou, Jie Lu, Jiwen |
| author_facet | Zuo, Sicheng Li, Yuxuan Zheng, Wenzhao Zhu, Zheng Zhou, Jie Lu, Jiwen |
| contents | Vision-language-action models have reshaped autonomous driving to incorporate languages into the decision-making process. However, most existing pipelines only utilize the language modality for scene descriptions or reasoning and lack the flexibility to follow diverse user instructions for personalized driving. To address this, we first construct a large-scale driving dataset (InstructScene) containing around 100,000 scenes annotated with diverse driving instructions with the corresponding trajectories. We then propose a unified Vision-Language-World-Action model, Vega, for instruction-based generation and planning. We employ the autoregressive paradigm to process visual inputs (vision) and language instructions (language) and the diffusion paradigm to generate future predictions (world modeling) and trajectories (action). We perform joint attention to enable interactions between the modalities and use individual projection layers for different modalities for more capabilities. Extensive experiments demonstrate that our method not only achieves superior planning performance but also exhibits strong instruction-following abilities, paving the way for more intelligent and personalized driving systems. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_25741 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Vega: Learning to Drive with Natural Language Instructions Zuo, Sicheng Li, Yuxuan Zheng, Wenzhao Zhu, Zheng Zhou, Jie Lu, Jiwen Computer Vision and Pattern Recognition Artificial Intelligence Robotics Vision-language-action models have reshaped autonomous driving to incorporate languages into the decision-making process. However, most existing pipelines only utilize the language modality for scene descriptions or reasoning and lack the flexibility to follow diverse user instructions for personalized driving. To address this, we first construct a large-scale driving dataset (InstructScene) containing around 100,000 scenes annotated with diverse driving instructions with the corresponding trajectories. We then propose a unified Vision-Language-World-Action model, Vega, for instruction-based generation and planning. We employ the autoregressive paradigm to process visual inputs (vision) and language instructions (language) and the diffusion paradigm to generate future predictions (world modeling) and trajectories (action). We perform joint attention to enable interactions between the modalities and use individual projection layers for different modalities for more capabilities. Extensive experiments demonstrate that our method not only achieves superior planning performance but also exhibits strong instruction-following abilities, paving the way for more intelligent and personalized driving systems. |
| title | Vega: Learning to Drive with Natural Language Instructions |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence Robotics |
| url | https://arxiv.org/abs/2603.25741 |