FabriVLA: A Lightweight Vision-Language-Action Model for Precise Multi-Task Manipulation

FabriVLA: A Lightweight Vision-Language-Action Model for Precise Multi-Task Manipulation

|Authors: Shiyuan Yang, Borong Zhang, Jizheng Zhang, Zhijia Tao, Junfei Guo, Donglai Ran, Xu Bian, Qingbiao Li

|Venue: arXiv preprint | Year: 2026

|arXiv: 2607.08575v1

Abstract

We present FabriVLA, a lightweight Vision-Language-Action model for Precise Multi-Task Manipulation. FabriVLA combines an InternVL3.5 vision-language backbone with a flow-matching action head featuring gated self-attention across action tokens and shallow VLM layer fusion for enriched spatial context. The model is trained via single stage joint optimization from a pretrained VLM and randomly initialized action head. On the Meta-World MT50 benchmark spanning 50 diverse manipulation tasks, FabriVLA achieves a tier-average success rate of 90.0%, demonstrating that a compact VLA built on a 1B scale VLM can achieve strong performance without relying on multi billion parameter VLA backbones.

Key Contributions

• We present FabriVLA, a lightweight Vision-Language-Action model for Precise Multi-Task Manipulation
• FabriVLA combines an InternVL3.5 vision-language backbone with a flow-matching action head featur…
• The model is trained via single stage joint optimization from a pretrained VLM and randomly initi…
• On the Meta-World MT50 benchmark spanning 50 diverse manipulation tasks, FabriVLA achieves a tier…

Topics

• manipulation
• vision
• vla
• reinforcement-learning
• benchmark

Code & Data

No code repository linked in paper metadata.

BibTeX

@article{Yang2026_260708575v1,
  title     = {FabriVLA: A Lightweight Vision-Language-Action Model for Precise Multi-Task Manipulation},
  author    = {Shiyuan Yang and Borong Zhang and Jizheng Zhang and Zhijia Tao and Junfei Guo and Donglai Ran and Xu Bian and Qingbiao Li},
  year      = {2026},
  eprint    = {2607.08575v1},
  archivePrefix = {arXiv},
  primaryClass  = {cs.RO},
  url       = {https://arxiv.org/abs/2607.08575v1}
}