English

Rethinking VLM Representation for VLA Initialization

Computer Vision and Pattern Recognition 2026-05-26 v1

Abstract

Vision-Language-Action (VLA) models widely adopt pretrained Vision-Language Models (VLMs) as policy backbones, yet it remains unclear what kind of pretrained VLM representation is useful as a VLA initialization. In this paper, we study VLA initialization as a controlled representation-design problem along three axes: capability-level embodied VQA supervision, parameter-update strategy, and robot-data pretraining. Our experiments show that the original pretrained VLM representation is a key source of action performance. However, embodied VQA adaptation does not yield uniform gains: its benefit depends on downstream bottlenecks, and gains from different capability domains are not simply additive. For update strategy, LoRA provides a more reliable initialization than Full Finetune, indicating that overly reshaping the pretrained representation can weaken VLA initialization. Robot-data pretraining further improves VLA initialization, with the strongest variant obtained by staged LoRA-based training. Together, these findings suggest that effective VLM-to-VLA adaptation should inject action-relevant embodied and robot-trajectory signals while preserving the pretrained VLM representation that remains useful for action learning.

Keywords

Cite

@article{arxiv.2605.25802,
  title  = {Rethinking VLM Representation for VLA Initialization},
  author = {Weifeng Lin and Siyuan Huang and Hao Li and Tingwei Chen and Ruichuan An and Xinyu Wei and Jianbo Liu and Hongsheng Li},
  journal= {arXiv preprint arXiv:2605.25802},
  year   = {2026}
}

Comments

9 main-text pages, 5 appendix pages, 4 figures