空间机械臂多模态视觉感知与操作技术综述

Review of multimodal visual perception and manipulation technologies for space manipulator

摘要:

为研究空间机械臂多模态视觉感知与操作的发展现状及亟须解决的关键技术难题,对目前该领域的文献进行了分析与总结。多模态视觉感知是指融合可见光、红外、深度相机与激光雷达等异构传感器及多源数据形式实现的视觉感知方式,空间操作是指面向在轨任务,利用机械臂等执行机构实施接近、抓取、装配与维护等作业的过程。首先,回顾了目前国内、外已服役的典型空间机械臂系统,总结其发展脉络与应用特点。其次,文中以“感知-规划-控制”为主线,系统梳理了自主在轨服务所需的3项关键技术:一是多模态视觉感知技术,重点介绍多源异构数据融合以及多模态视觉位姿估计技术；二是复杂约束下的轨迹规划技术,涵盖基于模型、优化与学习的空间机械臂轨迹规划方法,并探讨其在自由漂浮基座和强耦合动力学条件下的适用性；三是面向漂浮运动目标的柔顺抓取技术,以保障操作过程中的安全性。最后,总结了当前空间机械臂在自主在轨服务中面临的计算资源受限、在轨数据匮乏、多模态协同困难,以及运行生命周期短等核心挑战,并展望了其在硬件、算法与系统层面的未来发展趋势。研究表明:空间机械臂的自主在轨服务技术尚未成熟,关键技术环节与应用落地存在多处瓶颈,多模态视觉传感器与智能图像处理算法、基于学习的轨迹规划与柔顺抓取控制,将成为提升空间机械臂自主性能的核心发展方向。

Abstract:

To investigate the current development of multimodal visual perception and manipulation for space manipulators and the pressing technical challenges, this paper conducts an analysis and summary of the existing literatures. Multimodal visual perception refers to a vision approach that integrates heterogeneous sensors and multi-source data, including visible, infrared, depth cameras, and LiDAR. Space manipulation is on-orbit activities conducted with robotic manipulators and other actuators, encompassing approach, grasping, assembly, and maintenance. This paper first reviews representative space manipulator systems that have been deployed domestically and internationally, summarizing their developmental path and application characteristics. Building on this foundation, we adopt a perception-planning-control framework to systematically review three technologies essential for autonomous on-orbit servicing. We first address multimodal visual perception, focusing on heterogeneous data fusion and multimodal pose estimation. We then examine trajectory planning under complex constraints, covering model-based, optimization-based, and learning-based methods and their applicability to free-floating bases and strongly coupled dynamics. Last, we discuss compliant grasping for free-floating moving targets to ensure operational safety. Finally, the paper highlights major challenges faced by space manipulators in autonomous on-orbit servicing, including limited onboard computational resources, scarcity of on-orbit data, difficulties in multimodal coordination, and the need for long-term reliability. Future directions are then outlined from the perspectives of hardware, algorithms, and system-level integration. Research indicates that autonomous on-orbit servicing for space manipulators is still immature, with multiple bottlenecks persisting in key technical components and practical deployment. Advances in multimodal vision, learning-based trajectory planning, and compliant grasping control will be critical to enhancing autonomous performance.