arXiv：cs.AI· Haihao Lin, Xiangsheng Huang, Xiao Yang, Weibang Zhou, Yiqi Zhang, Bo Yang, Simin Zeng, Jiawei Yang, Zhengyang Wang, Jiahui Du·· 5 小时前
FiberTune：在视觉-语言-动作微调中保留动作纤维视觉残差
FiberTune: Preserving Action-Fiber Visual Residuals in Vision-Language-Action Fine-Tuning
AI 导读
FiberTune 是一种训练时目标，可在不增加推理开销的前提下保留教师模型结构化的视觉残差，缓解视觉-语言-动作（VLA）策略微调中的残差视觉坍缩。它通过在线动作探针从中间视觉 token 表示中筛选动作预测方向，并用冻结视觉教师的中心化方向对齐与有效秩正则化保留这些残差。
正文
Abstract:Action-supervised fine-tuning of vision-language-action (VLA) policies can compress visual structure that is weakly constrained by the action objective. We formalize this as residual visual collapse along local action fibers and propose FiberTune, a training-time objective that preserves teacher-structured visual residuals without adding inference-time overhead. FiberTune uses an online action probe to filter action-predictive directions from intermediate visual-token representations. It preserves the residuals through centered directional alignment with a frozen visual teacher and effective-rank regularization. Under identical training conditions, FiberTune improves over task-loss-only fine-tuning in every one of six simulation settings spanning two benchmarks and two architectures ($\pi_{0.5}$ and OpenVLA-OFT), as well as on physical SO-101 pick-and-place; representative gains include +10.7 pp SR(5) on long-horizon CALVIN ABC$\rightarrow$D and physical SO-101 task success rising from 72.7% to 78.1%. Representation analyses support the preservation of visual information in the residuals.
| Comments: | Accepted at CoRL 2026. Project page: this https URL Code: this https URL |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Robotics (cs.RO) |
| Cite as: | arXiv:2606.08653 [cs.CV] |
| (or arXiv:2606.08653v4 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2606.08653 arXiv-issued DOI via DataCite |
Submission history
From: Haihao Lin [view email]
[v1]
Sun, 7 Jun 2026 14:41:22 UTC (470 KB)
[v2]
Wed, 9 Sep 2026 03:51:29 UTC (470 KB)
[v3]
Sun, 13 Sep 2026 13:53:46 UTC (470 KB)
[v4]
Tue, 13 Oct 2026 10:38:05 UTC (474 KB)
来源：arXiv：cs.AI · arxiv.org