# RT-VLA：知识蒸馏实现VLA模型推理提速44.8倍

> 原文链接：https://robotics.hellobike.com/papers/2026-06-12-2606-14010-5569e1fb
> 频道：papers | 更新日期：2026-06-12 | 标签：auto, arxiv, cs.RO
> 核心摘要：针对VLA模型推理延迟高难题，提出RT-VLA轻量化架构。通过多级监督蒸馏迁移SimLingo能力，视觉模式推理提速44.8倍，视觉语言模式提速7.9倍，兼顾实时控制与事后可解释性分析。

# RT-VLA：知识蒸馏实现VLA模型推理提速44.8倍

::: v-pre
> 本文由 AI 根据公开内容整理，可能有误，请以原文为准：[RT-VLA: Real-Time Vision-Language-Action Models via Knowledge Distillation](https://arxiv.org/abs/2606.14010)（arXiv 2606.14010）

## 核心内容

针对Vision-Language-Action（VLA）模型因大参数骨干网络导致推理延迟高、难以部署的问题，研究提出RT-VLA轻量化模型。该方法采用多级监督蒸馏策略，将SOTA模型SimLingo的驾驶决策与语言推理能力迁移至紧凑学生模型中。实验表明，RT-VLA在保持闭环驾驶性能与语言推理竞争力的同时，纯视觉模式推理速度提升44.8倍，视觉+语言模式提升7.9倍，验证了蒸馏技术在构建实时可解释VLA模型中的有效性。

RT-VLA核心创新在于解耦实时控制与语言推理负载。系统保留语言模块用于安全关键场景的离线事后分析，避免在实时控制回路中引入额外延迟。相比传统端到端方案，该架构通过知识蒸馏压缩了视觉感知与动作预测模块的计算开销，使学生模型在资源受限条件下仍能继承教师模型的复杂推理能力。这种“在线轻量执行+离线语言审计”的设计范式，为具身智能体在算力约束下实现高性能与可解释性的平衡提供了工程参考。

## 关键看点

- **推理加速**：纯视觉模式推理耗时降低44.8倍，视觉语言联合模式提速7.9倍，满足实时控制需求
- **蒸馏策略**：采用多级监督蒸馏将SimLiao教师模型的驾驶与推理能力迁移至轻量级学生模型
- **可解释机制**：保留语言模块支持安全关键时刻的离线事后分析，不增加实时控制链路延迟
- **性能保持**：在大幅压缩计算量的同时，维持与教师模型相当的闭环驾驶及语言推理性能

## 工程与产业观察

RT-VLA证明了多级监督蒸馏是解决VLA模型落地延迟瓶颈的有效路径。其将语言推理从实时控制环中剥离进行离线审计的架构设计，为具身智能在边缘侧部署提供了重要参考：即在保证核心动作生成实时性的前提下，通过异步语言分析兼顾系统可解释性与安全性，平衡了算力约束与功能完备性。

## 英文摘要 (Abstract)

Vision-Language-Action (VLA) models have shown strong potential for end-to-end autonomous driving by jointly modeling visual perception, language reasoning, explainability and action prediction. However, their large vision-language backbones and reasoning modules introduce substantial inference latency and thereby prevent their deployment in the unforgiving reality of the road networks. We propose RT-VLA, a lightweight, distilled VLA model that transfers the driving and reasoning capabilities of the state-of-the-art SimLingo model into a compact student through multi-level supervised distillation. RT-VLA preserves language-based reasoning and supports post-hoc explanation through offline language analysis of safety-critical driving moments without adding latency to real-time control. Compared to the SimLingo teacher, RT-VLA maintains competitive closed-loop driving and language reasoning performance while reducing inference time by 44.8X in vision-only mode and 7.9X in vision+language mode. These results suggest that supervised distillation is a practical approach for building real-time, explainable VLA-style autonomous driving models.

作者：Xiangyu Huang, Zhenlin Hua, Han Zhou, Shounak Sural, Ragunathan Rajkumar
:::

## 原始参考来源 (Sources)

- [arXiv 2606.14010：RT-VLA: Real-Time Vision-Language-Action Models via Knowledge Distillation](https://arxiv.org/abs/2606.14010)