# RageVLA以0.21B参数在LIBERO-PRO刷新VLA鲁棒性SOTA

> 原文链接：https://robotics.hellobike.com/papers/2026-11-19-2611-23517-87882629
> 频道：papers | 更新日期：2026-11-19 | 标签：auto, arxiv
> 核心摘要：针对VLA模型泛化瓶颈，RageVLA通过数据重排与感知动作解耦，以0.21B参数在LIBERO-PRO取得0.72成功率，大幅超越现有最优模型。消融实验证实其策略执行严格依赖结构化感知输入而非图像记忆。

# RageVLA以0.21B参数在LIBERO-PRO刷新VLA鲁棒性SOTA

::: v-pre
> 本文由 AI 根据公开内容整理，可能有误，请以原文为准：[RageVLA: Augmentation-Trained Perception and Action for State-of-the-Art Robustness on LIBERO](https://arxiv.org/abs/2611.23517)（arXiv 2611.23517）

## 核心内容

针对VLA模型在LIBERO-PRO基准中因物体位移或指令变更导致性能骤降的问题，研究提出RageVLA架构。该模型仅含0.21亿参数，却在LIBERO-PRO上取得0.72成功率，显著优于此前最佳模型的0.53，大参数版本达0.73。核心突破在于将“做什么”与训练时的场景布局及特定措辞解耦，解决了现有策略对训练分布过拟合的痛点。

技术层面采用双重解耦：数据端对LIBERO演示进行物体旋转、环境迁移及指令改写增强；模型端基于冻结DINOv3特征构建感知引擎，输出物体位置列表与夹爪视角紧凑编码，结合TinyCLIP嵌入，由从零训练的Flow-matching Diffusion Transformer生成动作。消融显示当输入物体位置被随机打乱时，成功率从0.965跌至0.215；目标物体位移28cm时仍保持0.865成功率，验证了其对结构化感知的强依赖。

## 关键看点

- **SOTA性能**：0.21B参数RageVLA在LIBERO-PRO获0.72成功率，较前代最优提升36%，大模型版达0.73
- **数据增强**：对演示数据实施物体旋转、跨环境迁移及指令同义改写，强制切断策略对特定布局的过拟合
- **感知解耦**：基于冻结DINOv3提取物体位置列表与夹爪视角编码，替代原始像素作为动作模型核心条件输入
- **鲁棒性验证**：目标物体位移28cm下成功率0.865，远超$π_{0.5}$的0.105，证实非图像记忆驱动

## 工程与产业观察

RageVLA证明VLA鲁棒性瓶颈不在规模而在表征解耦。通过结构化感知中间层替代端到端像素映射，小参数模型即可实现强泛化。这为具身智能提供了低成本、高可靠的技术路径，提示工程落地应优先优化感知-动作接口设计而非盲目堆叠算力。

## 英文摘要 (Abstract)

Vision-language-action (VLA) models look saturated on LIBERO, yet most of them fall apart on LIBERO-PRO as soon as an object moves or the instruction asks for something new. We argue that what is missing is not scale but decoupling. In these policies, what to do is tied to where things were during training, to the scene around them and to the wording of the instruction. We cut these ties in two places. In the data, LIBERO demonstrations are re-arranged by general rules: every object and fixture is moved and rotated, tasks are carried into other environments and onto other surfaces, and every instruction is paraphrased. In the policy, a perception engine on frozen DINOv3 features, trained on this augmented data, hands the action model the scene in decoupled form: a list of objects with their positions, and a compact code of what the gripper is looking at, next to image and instruction embeddings from TinyCLIP. A flow-matching diffusion transformer, trained from scratch, acts on this view. We call this combination of Robust Augmentation and Grounded Encoding RageVLA. With 0.21 billion parameters, RageVLA sets a new state of the art on LIBERO-PRO among VLA policies, scoring 0.72 against 0.53 for the best published model, and a larger variant reaches 0.73. Two further tests on LIBERO-Object show that the policy acts on what it perceives. When the object positions it is given are shuffled while the camera images stay correct, its success falls from 0.965 to 0.215. When the target object is displaced by 28 cm, it still succeeds in 0.865 of the episodes, where $\pi_{0.5}$, run through the same harness, succeeds in 0.105.

作者：Amrit Lal Singh, Smarak Kanjilal
:::

## 原始参考来源 (Sources)

- [arXiv 2611.23517：RageVLA: Augmentation-Trained Perception and Action for State-of-the-Art Robustness on LIBERO](https://arxiv.org/abs/2611.23517)