# 256参数微调OpenVLA-OFT实现机器人行为风格精准调控

> 原文链接：https://robotics.hellobike.com/papers/2026-11-03-2611-00522-b28abc8a
> 频道：papers | 更新日期：2026-11-03 | 标签：auto, arxiv
> 核心摘要：研究证实冻结OpenVLA-OFT 7B模型后，仅需256个可训练参数的TinyLoRA适配器即可将物体搬运高度提升120mm且成功率达92%，通过重缩放更新机制可在不重训下平衡风格迁移与预训练能力保留。

# 256参数微调OpenVLA-OFT实现机器人行为风格精准调控

::: v-pre
> 本文由 AI 根据公开内容整理，可能有误，请以原文为准：[Behavioral Steering of Vision-Language-Action Models in the Low-Parameter Regime](https://arxiv.org/abs/2611.00522)（arXiv 2611.00522）

## 核心内容

针对预训练视觉-语言-动作（VLA）模型的行为导向难题，研究团队在冻结OpenVLA-OFT 7B策略网络的前提下，验证了极低参数容量下的行为调控可行性。实验显示，利用仅含256个可训练参数的TinyLoRA适配器，机器人在已训练任务中将物体搬运高度提升120mm且保持92%成功率，在未见过的测试任务中亦提升74mm并达80%成功率，证明了超低维更新对VLA执行风格的有效干预能力。

技术层面采用TinyLoRA适配器配合GRPO策略梯度或CMA-ES进化搜索进行优化。研究发现可通过重缩放学习到的更新向量，在不重新训练的情况下动态调节风格增益与基础能力保留的权衡；增加适配器容量通常能改善该权衡。优化器选择与参数量强相关：低参数区间CMA-ES表现优异，而随着容量增加GRPO提供更优的权衡曲线，为端侧VLA轻量化适配提供了明确工程路径。

## 关键看点

- **极低参数量效**：256参数TinyLoRA使搬运高度升120mm，训练/测试任务成功率分别达92%和80%
- **免重训权衡机制**：通过重缩放更新向量动态调节风格增益与预训练能力保留，无需二次训练
- **优化器选型规律**：低参数CMA-ES占优，高参数GRPO权衡更佳，适配容量决定算法选择
- **基座模型验证**：基于OpenVLA-OFT 7B冻结策略验证，确认大模型行为导向的低秩可塑性

## 工程与产业观察

该成果确立了VLA行为微调的参数效率下限，证明256级参数即可实现具身风格迁移。对于端侧部署与多场景适配，TinyLoRA结合更新重缩放机制提供了低成本定制方案；同时揭示了优化器与参数量的耦合关系，指导工程师根据算力约束精准选择RL或进化搜索策略，显著降低具身大模型落地门槛。

## 英文摘要 (Abstract)

Reinforcement learning (RL) is increasingly used to adapt pretrained vision-language-action models (VLAs). We investigate the problem of behavioral steering: adapting a VLA's execution style to preferences such as carrying objects higher. To investigate how little trainable capacity such steering requires and how much pretrained capability can be retained, we freeze an OpenVLA-OFT 7B policy and optimize TinyLoRA adapters with 8 to 384 trainable parameters using either policy-gradient RL (GRPO) or evolutionary search (CMA-ES). With just 256 trainable parameters, we raise carry height by 120 mm on trained tasks at 92% success and by 74 mm on a held-out task at 80% success. By rescaling the learned update, we can tune the trade-off between style gains and capability retention without retraining. Increasing adapter capacity generally improves this trade-off, retaining more pretrained capability at comparable style gains. The choice of optimizer also depends on capacity, with CMA-ES being competitive at low parameter counts, while GRPO offers better trade-offs as capacity grows. These findings show that extremely low-dimensional updates can steer pretrained VLA behavior, with adapter capacity, optimizer choice, and update scaling offering practical control over adaptation and retention.

作者：James Tcheng, Feiyu Gavin Zhu
:::

## 原始参考来源 (Sources)

- [arXiv 2611.00522：Behavioral Steering of Vision-Language-Action Models in the Low-Parameter Regime](https://arxiv.org/abs/2611.00522)