*Result*: 农业多机器人全覆盖作业关键技术研究进展与展望.

*[Significance] With the deepening of intelligent agriculture and precision agriculture, the agricultural production mode is gradually transforming from traditional manual experience based operations to a modern model driven by data, intelligent decisionmaking, and autonomous execution. In this context, improving agricultural operation efficiency and achieving large-scale continuous and seamless operation coverage have become key requirements for promoting the modernization of agriculture. The multi-robot full coverage operation technology, with its significant advantages in operation efficiency, system robustness, scalability, and resource utilization efficiency, provides practical and feasible intelligent solutions for key links such as sowing, plant protection, and harvesting in large-scale farmland. This technology, through the collaborative work of multi-robot systems, can not only effectively reduce the repetition rate of tasks and avoid omissions, but also achieve efficient and accurate continuous operations in complex and dynamic agricultural environments, greatly improving the automation and intelligence level of agricultural production. [Progress] Starting from the global perspective of systems engineering, an integrated closed-loop technology framework of "perception-decision-execution" is constructed. It systematically sorts out and deeply analyzes the technological development status and research methods of each key link in the full-coverage operations of agricultural multi robot. At the level of perception and recognition, it focus on exploring the application of multi-source information fusion and collaborative perception technology. By integrating multi-source sensor data, multi-level fusion of data level, feature level, and decision level is achieved, and a refined global environment model is constructed to provide accurate crop status, obstacle distribution, and terrain information for the robot system. Especially in the field of multi-robot collaborative perception, research has covered advanced models such as distributed simultaneous localization and mapping (SLAM) and ground to ground collaboration. Through information sharing and complementary perspectives, the system's perception ability and modeling accuracy in wide area, unstructured agricultural environments have been improved. At the decision-making and planning level, three key aspects are analyzed: task allocation, global path planning, and local path adjustment. Task allocation has evolved from traditional deterministic methods to market mechanisms, heuristic algorithms, and intelligent methods that integrate reinforcement learning and graph neural networks to address the challenges of dynamic and complex resource constraints in agricultural scenarios. The global path planning system analyzes the characteristics of geometric decomposition, grid method, global planning, and learning methods in terms of path redundancy, computational efficiency, and terrain adaptability. Local path planning emphasizes the combination of real-time perception in dynamic environments, using methods such as graph search, sampling optimization, model predictive control, and end-to-end reinforcement learning to achieve real-time obstacle avoidance and trajectory smoothing. At the control execution level, the focus is on model-based trajectory tracking and control technology, aiming to accurately convert planned paths into robot motion. Traditional control methods such as PID, LQR, sliding mode control, etc. are continuously optimized to cope with terrain undulations and system disturbances. In recent years, intelligent methods such as fuzzy control, neural network control, reinforcement learning, and multi machine collaborative strategies have been gradually applied, further improving the control accuracy and collaborative operation capability of the system in dynamic environments. [Conclusions and Prospects] The closed-loop technical framework is systematically constructed for agricultural multi-robot full coverage operations, and in-depth analysis of key modules is conducted, providing some understanding and suggestions, and providing theoretical references and technical paths for related research. However, the technology still faces many challenges, including perceptual uncertainty, dynamic changes in tasks, vast and irregular work areas, unpredictable dynamic obstacles, communication and collaboration barriers, and energy endurance issues. In the future, this field will further strengthen the integration with artificial intelligence, the Internet of Things, edge computing and other technologies, focusing on promoting the following directions, including the development of intelligent dynamic task allocation mechanism; optimize global and local path planning algorithms to enhance their efficiency and adaptability in large-scale complex scenarios; enhance the real-time perception and response capability of the system to dynamic environments; promote software hardware collaboration and intelligent system integration to achieve efficient communication and integrated task management; develop high-efficiency power systems and intelligent energy consumption strategies to ensure long-term continuous operation capability. Through these efforts, agricultural multi-robot systems will gradually achieve higher levels of precision, automation, and intelligence, providing key technological support for the transformation of modern agriculture. [ABSTRACT FROM AUTHOR]*

*［目的/意义］随着智慧农业和精准农业的快速发展, 多机器人全覆盖作业技术在农业领域得到了广泛关 注。该技术在作业效率、覆盖范围、系统鲁棒性与可扩展性等方面展现出显著优势, 能够支持机器人集群高效完 成播种、收割等大范围覆盖任务, 是实现农业生产智能化的关键。［进展］本文系统梳理了农业多机器人全覆盖作 业系统的整体技术框架, 构建了 "感知-决策-执行"一体化闭环框架, 并深入分析了各模块关键技术的发展与应 用现状。在感知识别层面, 重点探讨如何融合多源信息与协同感知技术以构建精确的全局环境模型; 在决策规划 层面, 集成任务分配、全局路径规划和局部路径调整等关键环节, 优化机器人集群作业任务与路径策略。在控制 执行层面, 则聚焦于基于模型的轨迹跟踪控制技术, 将规划指令精准转化为农机运动。［结论/展望］总结当前农 业复杂场景下的主要挑战, 包括任务数量多且需求变动、作业空间大且具有不规则性、动态环境中存在不可预知 的障碍物等, 并展望未来技术发展方向, 为推进农业智能化和现代化提供参考。 [ABSTRACT FROM AUTHOR]*