随着无线传感技术的高速发展以及多节点协同网络的不断演进，多节点网络的定位与编队技术在环境勘探、自动驾驶、应急搜救、仓储物流等众多领域得到了广泛的应用，尤其是物联网、无人集群作业等多机系统的出现，更加使得面向多节点网络的协作定位与编队控制技术受到学术界与工业界的高度重视。然而，大规模网络、动态复杂运动场景下的定位与控制问题依然存在着较大的挑战。针对这一问题，本文建立了多机协作定位框架，在动态场景中将导航与控制联合优化提高定位与编队性能，对于复杂环境下的编队避障任务，引入深度强化学习策略给出可行方案，并通过仿真实验与实物平台分别验证各算法的有效性。主要的研究工作与创新点总结为如下三个方面：首先，针对多节点分布式定位，提出了主干-侦听相对定位架构，并设计了分布式几何合并与反向校准算法进一步提高系统的定位稳定性与精度。基于费希尔信息矩阵分析，本文揭示了节点测量噪声以及位置分布对于定位性能的影响，以此指导我们对所设计系统进行性能优化，同时作为性能基准仿真验证所设计算法的性能。其次，针对动态场景的导航与编队，提出了航位推算滤波方案完成空时协作，设计主干节点激活策略减少由节点拓扑畸形而引起的性能损失，通过反向校准融合邻居节点的测量和定位信息提高时空基准精度，将定位阶段的统计信息同步给编队控制阶段，借助模型预测算法完成定位与编队性能的联合优化。最后，针对复杂场景下的多机编队与避障任务，设计基于深度强化学习的编队避障算法，通过课程学习、策略蒸馏等算法保证训练过程的收敛性与泛化性，除了数值仿真，我们将上述算法均在实物测试平台上部署验证，利用实测数据证明了所设计算法的性能优势。综上所述，本文围绕多节点网络的协作相对定位与编队问题，提出了分布式协作相对定位框架，给出了动态场景导航与编队联合优化的方案，设计了基于深度强化学习的编队避障策略，并在实物平台上将各算法进行部署验证，大量的实测数据为系统的有效性提供了坚实的保障。相关研究工作既可以为多机系统的定位与编队提供理论支撑，亦可以进行方法指导。

With the rapid development of wireless sensing technology and the continuous evolution of multi-agent collaborative networks, cooperative localization and formation control technologies for multi-agent systems have encompassed a vast range of applications in civil and military contexts, such as environmental exploration, autonomous driving, emergency search and rescue, warehousing and logistics. In particular, the emergence of the Internet of Things and unmanned swarm combat platforms has made the cooperative localization and formation control technologies highly valued by academia and industry.However, there are still great challenges in the localization and control of large-scale networks and complex dynamic scenes. To this end, this paper builds a multi-agent cooperative localization framework, and improves localization and formation performance in dynamic scenarios by jointly optimizing navigation and formation control. For obstacle avoidance tasks in complex environments, deep reinforcement learning strategies are introduced to give feasible solutions. The effectiveness of each algorithm is verified not only by simulations, but also by experiments on a self-built physical platform. The main contributions are specified as follows.First, we establish a backbone-listener relative localization architecture for distributed multi-agent systems, where a distributed geometric merging strategy and a back calibration algorithm are proposed to further improve the localization stability and accuracy of the system. Based on the Fisher information matrix analysis, we characterize the effects of agents’ measurement noise and location distribution on localization performance, and at the same time use this performance error lower bound to serve as a benchmark to validate the performance of the designed algorithms.Second, we propose a dead reckoning filtering scheme for the navigation and formation in dynamic scenarios, where spatio-temporal cooperation is accomplished by information fusion. A node activation strategy is designed to reduce the performance loss caused by node topology deformity, and the measurement and positioning information of neighbor agents are fused by a back calibration algorithm to improve the accuracy of the space-time reference. With the help of the model prediction algorithm, the joint optimization of navigation and formation control is completed by sharing the statistical information of the localization stage.Third, we propose a distributed formation and obstacle avoidance method based on multi-agent reinforcement learning. The convergence and generalization of the training process are ensured through strategies such as curriculum learning and policy distillation.In addition to numerical simulations, the above algorithms are deployed and verified on the physical test platform, and the outperformance of the designed algorithms is proved by the actual measured data.In conclusion, this paper focuses on the cooperative relative localization and formation control of multi-agent systems. We establish a cooperative relative localization framework for distributed networks, and propose a joint optimization scheme for navigation and formation control. To solve multi-agent formation as well as obstacle avoidance tasks, we design a multi-agent reinforcement learning strategy, and verify each algorithm on the physical platform. Extensive simulation and real-world experiments validate the effectiveness of the proposed algorithms. This work not only provides theoretical support for the localization and formation control of multi-agent networks, but also providemethodological guidance.