AIR-VLA+: Decoupling Movement and Manipulation via Cascaded Dual-Action Decoders with Asymmetric MoE for Aerial Robots

📄 arXiv: 2606.12859v1 📥 PDF

作者: Jianli Sun, Bin Tian, Qiyao Zhang, Zijian Liu, Yutong Wang, Zhiyong Cui, Bai Li, Yisheng Lv, Yonglin Tian

分类: cs.RO

发布日期: 2026-06-11

💡 一句话要点

提出AIR-VLA+以解决无人机运动与操作耦合问题

🎯 匹配领域: 支柱一：机器人控制 (Robot Control) 支柱二：RL算法与架构 (RL & Architecture) 支柱九：具身大模型 (Embodied Foundation Models)

关键词: 无人机操作 级联解码器 混合专家 任务协调 运动控制

📋 核心要点

  1. 现有无人机操作系统在运动与操作之间存在耦合，导致控制效率低下和稳定性不足。
  2. 本文提出AIR-VLA+架构，通过级联双动作解码器和不对称MoE，解耦无人机运动与操作，提升任务协调性。
  3. 实验结果显示，AIR-VLA+在标准基准上平均得分达到48.0，相较于单头策略提升80.2%，显著改善了协调控制冲突。

📝 摘要（中文）

无人机操作系统在端到端控制中长期受到表示耦合的困扰，平台级无人机运动与末端执行器级臂操作在动作规模、动态和控制目标上存在显著差异。本文提出AIR-VLA+，一种专为无人机操作设计的流匹配动作生成架构，采用级联双动作解码器和不对称特征级混合专家（MoE）。通过构建级联的操作和运动解码器，允许无人机在运动过程中单向观察操纵器的意图，从而实现工作流协调，并隔离无人机运动信息对臂操作稳定性的影响。实验表明，该方法在标准化的AIR-VLA基准上全面超越所有基线，整体任务完成分数较单头策略提升80.2%。

🔬 方法详解

问题定义：本文旨在解决无人机在执行操作时运动与操作之间的耦合问题，现有方法在控制效率和稳定性上存在不足。

核心思路：通过设计AIR-VLA+架构，采用级联的运动和操作解码器，使无人机能够在运动过程中观察操纵器的意图，从而实现更好的工作流协调。

技术框架：整体架构包括输入特征增强模块、级联运动解码器和操作解码器，以及不对称MoE结构，允许不同的运动专家在训练过程中自适应地展现能力倾向。

关键创新：最重要的创新在于引入了不对称MoE架构，使得无人机运动能够根据任务阶段自适应调整，显著提高了任务阶段的适应性。

关键设计：在运动解码器中，采用隐式视觉抓取投影器来感知抓取器与物体之间的交互状态，并注入压缩的全局语义特征，增强输入特征的表达能力。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

实验结果表明，AIR-VLA+在标准化AIR-VLA基准上取得了48.0的平均得分，相比于单头策略$π_{0.5}$，任务完成分数提升了80.2%。该方法有效缓解了复合机器人协调控制中的异质性冲突，展现出优越的性能。

🎯 应用场景

该研究的潜在应用领域包括无人机的物流运输、灾害救援、农业监测等场景，能够有效提升无人机在复杂环境中的操作能力和任务执行效率。未来，该技术有望推动无人机在多任务协作中的广泛应用，提升自动化水平。

📄 摘要（原文）

Aerial manipulation systems have long suffered from representation coupling in end-to-end control, as platform-level Unmanned Aerial Vehicle (UAV) movement and end-effector-level arm manipulation differ substantially in action scale, dynamics, and control objectives. In this paper, we propose AIR-VLA+, a flow matching action generation architecture specifically designed for aerial manipulation, featuring cascaded dual-action decoders and an asymmetric feature-level Mixture of Experts (MoE). We construct cascaded manipulation and movement decoders, allowing the UAV to unidirectionally observe the manipulator's intent during movement to achieve workflow coordination, while isolating the impact of UAV movement information backpropagation on arm manipulation stability. Addressing the characteristic that UAV movement is highly dependent on high-level semantics and responsible for task state transitions in aerial manipulation, we design an input feature enhancement module for the UAV movement decoder. This module introduces an implicit visual grasp projector to perceive the interaction state between the gripper and the object, and injects compressed global semantic features. Within the UAV movement decoder, we deploy an implicit MoE architecture, enabling different movement experts to spontaneously exhibit capacity inclinations for various task stages during training. Through dense soft blending computation on the feature manifold, the UAV movement is endowed with stronger task-stage adaptability. Experiments on the standardized AIR-VLA benchmark demonstrate that our method comprehensively surpasses all baselines with an overall average score of 48.0. The overall task completion score improves by 80.2\% compared to the single-head $π_{0.5}$ policy, effectively mitigating the heterogeneous coordinated control conflicts of composite robots.