# OpenFlyV2：端侧VLA模型实现无人机10Hz闭环导航与零样本迁移

> 原文链接：https://robotics.hellobike.com/papers/2026-10-29-2610-39287-8e1c0a4e
> 频道：papers | 更新日期：2026-10-29 | 标签：auto, arxiv, hf-daily-papers
> 核心摘要：OpenFlyV2统一视觉语言动作架构在Jetson Orin NX端侧以10Hz生成航点，构建7万条轨迹仿真数据集，四旋翼零样本迁移实飞导航成功率达48%，验证资源受限平台单机闭环能力。

# OpenFlyV2：端侧VLA模型实现无人机10Hz闭环导航与零样本迁移

::: v-pre
> 本文由 AI 根据公开内容整理，可能有误，请以原文为准：[OpenFlyV2: An onboard vision-language-action model for aerial perception and navigation](https://arxiv.org/abs/2610.39287)（HF Daily Papers · arXiv 2610.39287）

## 核心内容

研究团队发布OpenFlyV2，一种面向空中机器人的统一机载视觉-语言-动作（VLA）架构，旨在解决现有系统依赖离线计算或功能割裂问题。该模型基于共享表征支持语义理解、空间推理与连续控制，并开源代码至AeroBrain仓库。配套构建了包含约7万条导航轨迹及6万对视觉语言指令响应的仿真数据集，为具身航空智能提供标准化训练基准。

技术上采用紧凑动作专家模块将多模态上下文转为短视域航点，结合运动感知时序记忆与动作先验降低闭环算力开销。系统在NVIDIA Jetson Orin NX上以10Hz频率输出含12个航点的动作块，无需真实飞行数据即可零样本迁移至物理四旋翼。在四个真实环境中测试，导航成功率达48%，证实了宽泛VLA能力与实时闭环控制在资源受限端侧平台的共存可行性。

## 关键看点

- **端侧推理性能**：在NVIDIA Jetson Orin NX上以10Hz频率实时生成包含12个航点的动作块，满足空中机器人闭环控制时延要求
- **零样本迁移验证**：仅依赖仿真数据训练，无真实飞行微调即在四个物理环境实现48%导航成功率，验证Sim-to-Real泛化能力
- **专用数据集构建**：发布含7万条导航轨迹与6万对视觉语言指令对的仿真数据集，支撑航空VLA模型语义理解与空间推理训练
- **轻量化架构设计**：引入紧凑动作专家与运动感知时序记忆机制，在单机资源受限条件下实现感知、推理与控制统一闭环

## 工程与产业观察

OpenFlyV2证明在Jetson Orin NX等嵌入式平台上，通过动作专家与时序记忆优化，VLA模型可兼顾语义泛化与10Hz实时控制。其48%零样本成功率虽待提升，但确立了航空具身智能端侧部署的技术基线，为后续 Sim-to-Real 数据管线与轻量化策略研究提供了关键参照。

## 英文摘要 (Abstract)

Human cognition flexibly draws on a shared perceptual foundation to understand environments, attend to referenced entities, reason about spatial relationships, and act. Endowing aerial robots with an analogous functional organization remains challenging, as existing systems are often specialized for individual capabilities, dependent on offboard computation, or unable to sustain closed-loop operation on resource-constrained platforms. We present OpenFlyV2, a unified onboard vision-language-action architecture that uses a shared vision-language representation to support semantic understanding, language-grounded perception, spatial reasoning, and continuous aerial control within a single model. A compact action expert converts the shared multimodal context into short-horizon flight waypoints, while motion-aware temporal memory and action priors reduce the computational cost of closed-loop execution. We construct , a simulation-based dataset comprising approximately 70,000 navigation trajectories and 60,000 vision-language instruction--response pairs. OpenFlyV2 generates action chunks, each comprising 12 waypoints, entirely onboard at 10 Hz using an NVIDIA Jetson Orin NX. It transfers zero-shot from simulation to physical quadrotor flight, achieving a 48\% navigation success rate across four real-world environments without real-world aerial training data. Extensive results demonstrate that broad vision-language competence and continuous closed-loop flight can coexist within a single model on a resource-constrained aerial platform. Code will be made publicly available at https://github.com/Ideal-ljl/AeroBrain.

Hugging Face Daily Papers 热度：3 upvotes
作者：Junli Liu, Zhongrui You, Chenhui Li, Zhen Li, Yuyang Tian, Qizhen Chen, Delin Qu, Haoming Song, Dong Wang, Xuelong Li, Zhigang Wang, Bin Zhao
:::

## 原始参考来源 (Sources)

- [HF Daily Papers · arXiv 2610.39287：OpenFlyV2: An onboard vision-language-action model for aerial perception and navigation](https://arxiv.org/abs/2610.39287)