DeMaVLA: A Vision-Language-Action Foundation Model for Generalizable Deformable Manipulation
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866910272115965952 |
|---|---|
| author | Su, Taiyi Zhu, Jian Wang, Tianjian He, Youzhang Huang, Zitai Zhang, Jianjun Ma, Chong Wang, Hanyang Zhang, Tianjiao Yin, Munan Ding, Weihao Xu, Yi |
| author_facet | Su, Taiyi Zhu, Jian Wang, Tianjian He, Youzhang Huang, Zitai Zhang, Jianjun Ma, Chong Wang, Hanyang Zhang, Tianjiao Yin, Munan Ding, Weihao Xu, Yi |
| contents | Real-world household robots require Vision-Language-Action (VLA) foundation models that can acquire reusable manipulation skills across diverse objects, task conditions, and household environments. Deformable-object folding is a representative challenge, requiring robots to handle clothing items from random initial states across varying categories, geometries, materials, and scenes. However, existing VLA systems commonly train separate policies for different object categories, while naively mixed multi-task training often suffers from task interference and degraded performance. To move beyond category-specific folding policies, we introduce DeMaVLA, a VLA foundation model for generalizable Deformable Manipulation. DeMaVLA adopts a VLM backbone with an action expert and formulates continuous action generation using flow matching. To improve efficiency, the action expert is constructed by pruning every other transformer layer while preserving layer-wise alignment with the VLM backbone, reducing training and inference cost. DeMaVLA is first pre-trained on approximately 5,000 hours of selected real-world dual-arm demonstrations to acquire general manipulation priors. It is then post-trained on mixed folding data that aggregates self-collected demonstrations and corrective trajectories from real-robot failures across multiple folding tasks through a human-in-the-loop Data Aggregation~(DAgger) pipeline. Experiments show that DeMaVLA achieves competitive performance on RoboTwin and strong real-world results on our household folding benchmark. These results highlight the value of scalable real-world data, efficient action generation, and corrective learning for general-purpose VLA policies in deformable-object manipulation. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_31286 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | DeMaVLA: A Vision-Language-Action Foundation Model for Generalizable Deformable Manipulation Su, Taiyi Zhu, Jian Wang, Tianjian He, Youzhang Huang, Zitai Zhang, Jianjun Ma, Chong Wang, Hanyang Zhang, Tianjiao Yin, Munan Ding, Weihao Xu, Yi Robotics Artificial Intelligence Real-world household robots require Vision-Language-Action (VLA) foundation models that can acquire reusable manipulation skills across diverse objects, task conditions, and household environments. Deformable-object folding is a representative challenge, requiring robots to handle clothing items from random initial states across varying categories, geometries, materials, and scenes. However, existing VLA systems commonly train separate policies for different object categories, while naively mixed multi-task training often suffers from task interference and degraded performance. To move beyond category-specific folding policies, we introduce DeMaVLA, a VLA foundation model for generalizable Deformable Manipulation. DeMaVLA adopts a VLM backbone with an action expert and formulates continuous action generation using flow matching. To improve efficiency, the action expert is constructed by pruning every other transformer layer while preserving layer-wise alignment with the VLM backbone, reducing training and inference cost. DeMaVLA is first pre-trained on approximately 5,000 hours of selected real-world dual-arm demonstrations to acquire general manipulation priors. It is then post-trained on mixed folding data that aggregates self-collected demonstrations and corrective trajectories from real-robot failures across multiple folding tasks through a human-in-the-loop Data Aggregation~(DAgger) pipeline. Experiments show that DeMaVLA achieves competitive performance on RoboTwin and strong real-world results on our household folding benchmark. These results highlight the value of scalable real-world data, efficient action generation, and corrective learning for general-purpose VLA policies in deformable-object manipulation. |
| title | DeMaVLA: A Vision-Language-Action Foundation Model for Generalizable Deformable Manipulation |
| topic | Robotics Artificial Intelligence |
| url | https://arxiv.org/abs/2605.31286 |