Research On Distributed Source Localization And State Estimation In Multi-agent Systems

Multi-agent systems(MAS)have played a critical role in many applications,e.g.,space explo-ration,military,industrial and agriculture engineering,intelligent transportation,due to its natural distribution,strong robustness and scalability,and efficient computing performance.Before de-signing the control strategy for the MAS,we always need the real-time localization and the current state vector of the agent.It's one of the fundamental problems in MAS to localize the agent and estimate the agent its state as precisely as possible.This thesis would present the distributed local-ization algorithm for mobile multi-agent systems by the research on source localization problem with local measurement information;furthermore,we aim to solve the state vector estimation prob-lem in distributed network systems by the study on state estimation in relative sensing networks.Due to the movement of the agent and the variety of the network topology,the approaches in static network localization seems not applicable in mobile multi-agent system localization.Hence,it's of critical importance to study mobile multi-agent system localization problem.The authors present several distributed observers to estimate the position of source dynamically,and provide the-conditions for the observers to achieve exponentially convergence.Compared with static localiza-tion algorithms,the dynamical Besides,for general sensing networks,the core problem lies in how to precisely estimate the state of the nodes under complex topology changing and measurement missing.Our main contribution of the thesis can be summarized as follows:Firstly,a distributed localization algorithm is designed to cope with the multi-agent source lo-calization problem under switching topology,for which the goal is to estimate the relative position of the silent source in each agent's local frame.Sampled local measurements and local measure-ment information are transmitted among neighbors(including the source).Collaboration among agents is carried out to overcome measurement failures and topology constraints.The algorithm in this paper includes the followings:The algorithm for relative localization between pairwise a-gents can achieve exponential convergence to the true value if the relative motion of every pair of neighboring agents satisfies the persistent excitation condition;The consensus-based fusion algo-rithm for the estimate of the source's position in each agent's local frame would be asymptotically convergent as long as each agent in the sensing and communication graph is uniformly jointly reachable from the source.Secondly,we aim to solve the distributed source localization problem based on bearing mea-surements.We consider the nonholonomic unicycle model,which has been employed in many studies of distributed robotic systems to model a differentially driven mobile robot and aerial ve-hicle.Differing from the point-mass model assumed in the aforementioned works,the local frame on each unicycle-type agent rotates according to its kinematic motion,which causes extra chal-lenges in relative localization,in particular in the scenarios where no GPS and no compass are available to provide each agent its own absolute position and orientation.However,we do succeed in developing a simple and provably convergent distributed source localization algorithm.It is shown that if the graph describing the communication and sensing relationship among the agents and the source is connected and if the relative motion of any pair of neighboring agents satisfies a persistent excitation condition,then the estimate by every agent can asymptotically converge to the true relative coordinate of the source in its local frame.The novelty of our work is in the devel-opment of a distributed estimation scheme that takes into consideration of the more complicated nonholonomic-constraint motion and does not require a commonly known coordinate system.Finally,we research on the distributed Kalman filtering problem in relative sensing networks with rigorous analysis.Inspired by the previous work on localization,we focus on more general state estimate problem in relative sensing networks.The relative sensing network is modeled by an undirected graph while nodes in this network are running homogeneous dynamical models.The sufficient and necessary condition for the observability of the whole system is given with detailed proof.By local information and measurement communication,a novel distributed suboptimal esti-mator based on the Kalman filtering technique is designed to make comparison with the centralized optimal estimator.Sufficient conditions for its convergence are presented with respect to the topol-ogy of the network and the numerical solutions of n LMI equations combining system parameters.The distributed algorithm reduces the communication rate,improves the estimate scalability of the system,but still combines the basic idea of Kalman filter.It can intuitively perform quite well but the convergence of it seems also challenging because it's comprised of no standard Kalman filter form and also affected by the system topology and parameters.Hence we design a novel method to prove the convergence of the distributed algorithm,which effectively solve this problem while the traditional analysis methods fail.Detailed theoretical proof for the presented algorithms and the estimation methods are given in corresponding chapters of this dissertation.The numerical simulations are given in each chapter to illustrate the results.