多节点集群系统高精度高鲁棒性导航需求，给传统面向单节点导航定位技术带来了难题。协作定位通过节点间协作与信息交互提高整体定位能力，为集群系统任务完成提供了新的导航定位解决方案，进而，也相应地在通信开销、实时性、功耗等方面，为算法设计带来了新挑战。本课题瞄准集群协作导航定位核心算法设计，着重进行了以下三方面的研究。首先，对集群协作导航定位概念进行了定义，并从高、低动态集群特性出发，分析相应导航定位需求，给出了相应的算法设计出发点、目标、及约束条件。第二，针对现有非参数化消息传递算法通信开销大、计算复杂度高的问题，提出了基于高斯置信度表征与传递的参数化消息传递协作导航定位算法。具体来说，在置信度表征与传递中，采用均值和方差来表征置信度并以广播的方式进行传递，降低了通信开销；在消息与置信度更新中，采用基于统计线性回归的线性化方法对涉及的非线性函数进行线性化，进而得到基于均值和方差的闭式更新表达式，降低了计算复杂度。仿真表明，在高斯观测噪声假设下，所提出的算法在低动态集群协作导航定位中性能更稳定、通信开销更小、计算复杂度更低。第三，基于高斯期望传播和统计线性化方法，设计了一种基于期望传播的迭代协作定位算法，信息交互与迭代运算机制灵活可变，可适应非同步低动态集群协作导航定位系统不同精度及收敛性的需求。 最后，面向高动态集群系统，针对现有协作算法需要在一个时隙内反复进行信息交互，难以满足高实时性需求的问题，提出了基于 MMSE 滤波的、非迭代序贯协作导航定位算法。首先，给出了集中式 MMSE 集群导航定位滤波器并对其进行分布式解耦，得到相应的分布式 MMSE 滤波器；之后，基于该分布式滤波器，设计了相应的序贯协作导航定位算法架构，该架构下，在一个时隙内，节点间只需要进行一次信息交互，从而降低了通信开销，提高了实时性，同时，由于该框架由集中式 MMSE 滤波解耦得到，故精度和稳定性得到了保证；最后，分别利用一阶线性化和统计线性化方法，对基于该架构的协作导航定位算法进行了低复杂度实现，并将其与现有迭代算法进行了比较。仿真表明，在高斯观测噪声假设下，提出的算法在保持高精度和高鲁棒性的同时，提高了实时性并降低了通信开销。
The high precision and robustness requirements of cluster navigation bring challenges to conventional single-agent oriented localization techniques, which can be well settled via the use of cooperation and information interaction. However, in cooperative localization, all agent in a cluster need to cooperate with each other to improve the overall localization performance, which in turn bring challenges to algorithm design under constraints such as low communication power overheads, real-time capabilities, and high precision requirements. In this paper, this work focus on designing key cooperative localization and navigation algorithms for cluster systems, and conduct several works as follows.Firstly, this work gives out the definition of cluster localization and navigation, analyizes the localization and navigation requirements for low and high dynamic cluster systems respectively, and put forward the key elements of algorithm design.Secondly, aiming to solve the problem that the existing nonparametric MP based cooperative algorithms usually have high communication and computation costs, and thus can be hardly used for real applications, propose a parametric one based on Gaussian belief representations. Specifically, this work represents and propagates the beliefs by their means and covariances to reduce the communication overhead; linearize the nonlinear functions therein using a linearization method that based on statistical linear regression to reduce the computation complexity; design two cooperation and information interaction mechanisms based on belief propagation and expectation respectively to accommodate the change of environment and different accuracy/convergence requirements.Cluster localization simulations show that under the Gaussian measurement noise assumptions and given the same accuracy requirements, our algorithms enjoy more robustness, lower communication and computation costs. Thirdly, to accommodate with different accuracy and convergence demands of loclaization and navigation for asynchrony culster systems, this work proposes an iterative cooperative localization algorithm based on expectation propagation, where the information interaction and iterative calculation strategies are more flexible.Finally, in view of that existing algorithms usually need lots of information intersections per time slot, which fails to meet the real-time requirements of high dynamic cluster systems, the work proposes a cooperative localization algorithm based on minimum mean square error (MMSE) filtering. This work first provide a MMSE filter for centralized localization and then decouple it into a set of sub-ones to obtain a distributed MMSE filter; upon that, this work designs a cooperative localization and navigation framework in which the information interaction in per time slot reduces to only one and what is more, since this distributed framework is obtained by decoupling the centralized one, its accuracy and robustness can be guaranteed; finally, the work presents a low complexity algorithm implementation by employing the first-order and statistical linearization techniques. Simulation shows that under the Gaussian measurement noises assumptions, the proposed algorithm not only has high accuracy and robustness, but also enjoys more efficiency and better real time performance.
