VLA-Thinker: Boosting Vision-Language-Action Models through Thinking-with-Image Reasoning
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_ | 1866911518249975808 |
|---|---|
| author | Wang, Chaoyang Bao, Wenrui Gao, Sicheng Xu, Bingxin Tian, Yu Rawat, Yogesh S. Ge, Yunhao Shang, Yuzhang |
| author_facet | Wang, Chaoyang Bao, Wenrui Gao, Sicheng Xu, Bingxin Tian, Yu Rawat, Yogesh S. Ge, Yunhao Shang, Yuzhang |
| contents | Vision-Language-Action (VLA) models have shown promising capabilities for embodied intelligence, but most existing approaches rely on text-based chain-of-thought reasoning where visual inputs are treated as static context. This limits the ability of the model to actively revisit the environment and resolve ambiguities during long-horizon tasks. We propose VLA-Thinker, a thinking-with-image reasoning framework that models perception as a dynamically invocable reasoning action. To train such a system, we introduce a two-stage training pipeline consisting of (1) an SFT cold-start phase with curated visual Chain-of-Thought data to activate structured reasoning and tool-use behaviors, and (2) GRPO-based reinforcement learning to align complete reasoning-action trajectories with task-level success. Extensive experiments on LIBERO and RoboTwin 2.0 benchmarks demonstrate that VLA-Thinker significantly improves manipulation performance, achieving 97.5% success rate on LIBERO and strong gains across long-horizon robotic tasks. Project and Codes: https://cywang735.github.io/VLA-Thinker/ . |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_14523 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | VLA-Thinker: Boosting Vision-Language-Action Models through Thinking-with-Image Reasoning Wang, Chaoyang Bao, Wenrui Gao, Sicheng Xu, Bingxin Tian, Yu Rawat, Yogesh S. Ge, Yunhao Shang, Yuzhang Computer Vision and Pattern Recognition Artificial Intelligence Robotics Vision-Language-Action (VLA) models have shown promising capabilities for embodied intelligence, but most existing approaches rely on text-based chain-of-thought reasoning where visual inputs are treated as static context. This limits the ability of the model to actively revisit the environment and resolve ambiguities during long-horizon tasks. We propose VLA-Thinker, a thinking-with-image reasoning framework that models perception as a dynamically invocable reasoning action. To train such a system, we introduce a two-stage training pipeline consisting of (1) an SFT cold-start phase with curated visual Chain-of-Thought data to activate structured reasoning and tool-use behaviors, and (2) GRPO-based reinforcement learning to align complete reasoning-action trajectories with task-level success. Extensive experiments on LIBERO and RoboTwin 2.0 benchmarks demonstrate that VLA-Thinker significantly improves manipulation performance, achieving 97.5% success rate on LIBERO and strong gains across long-horizon robotic tasks. Project and Codes: https://cywang735.github.io/VLA-Thinker/ . |
| title | VLA-Thinker: Boosting Vision-Language-Action Models through Thinking-with-Image Reasoning |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence Robotics |
| url | https://arxiv.org/abs/2603.14523 |