Formation Control Methods Of Networked Multi-agent Systems And Development Of Real-time Simulation Platform

Networked multi-agent formation refers to the formation of multiple single agents with input and output,and the agents communicate with each other through a certain topological structure,so that they can form a fixed or changing formation to cooperate with each other to complete a specific task.Compared with the traditional single agent control,it has the advantages of high flexibility,high reliability,low cost and can accomplish more complex tasks.Networked multi-agent formation control includes four elements: agent model,network performance index,formation control algorithm and algorithm verification platform.None of these four elements is indispensable,otherwise it is insufficient to fully describe the details of networked multi-agent formation control.In order to ensure the realization of multi-agent formation task based on network,effective networked predictive control algorithm and accurate system model are the key points.In this dissertation,some issues existing in multi-agent formation control are studied.First,in the research of multi-agent formation control,many research objects are selected with special characteristics,and it is often unable to compare with other methods due to the lack of model parameters and experimental details.In order to solve this problem,based on the ordinary commercial quadrotors as the research object of multi-agent formation control,the identification work was carried on for quadrotor systemss in detail.Furthermore,the model testing and screening was completed,and finally a standard networked prediction control model was obtained.On this basis,the contrast experiment analysis was designed which applies the the math model and real quadrotor with the same controller to verify the accuracy of the models.Second,in this dissertation a flight control platform based on Vicon visual positioning system is built.The double close-loop PID control algorithm and adaptive adjustment median potential control algorithm are used to control the quadrotors effectively,and the results are verified by specific experiments.In view of the forward channel delay,a PID predictive controller is designed and applied to the quadrotor control experiment.Based on the leader-follower control strategy,the analysis of multi-agent formation control,the design of multi-agent formation controller,the simulation of multi-agent formation control and the experiment of multi-agent formation control were completed.Third,the network performance index is a very important element,in which packet loss,network delay,data in the wrong order and other network characteristic problems are not easy to describe,often these parameters and specific control signals are bound together and sent as one packet.Therefore,this dissertation proposes a method for designing networked control toolbox,which can be used to simulate network delay,packet loss,network conflict,timestamp,delay compensation and so on.Because this method is based on standard C language S-function of Matlab software Simulink Function,it has good applicability and portability.In addition,the overall design idea,module function and the way of implementation of the toolbox are explained and illustrated in detail.Fourth,there are many practical problems in the experiment of quadrotor formation control.Sometimes in the process of multi-agent formation control experiments,the crashes may happen caused by malfunction or algorithm problem,which causes potential security risks.With the increase of the number of formation system individuals and the requirements of increasing the control precision,the equipment overhead of the quadrotor and controller will increase greatly.To solve these problems,this dissertation proposes a method for 3D virtual experiment equipment design,which can be used to replace the real experimental objects by semi physical simulation experiment.A virtual simulation experiment platform for multi-agent formation control is set up,and through the demonstration of networked multi-agent hierarchical control algorithm experiment,the effectiveness of network formation control algorithm validation on this platform is illustrated.Fifth,although the virtual experimental equipment has many advantages,virtual experimental equipment lacks sense of reality due to its disconnection from the real environment,and it is not easy for virtual equipment to compare with real equipment.There is still a gap between simulation results and real experimental results of many algorithms.To address this phenomenon,in this dissertation a networked multi-agent augmented reality simulation platform is developed,which combines 3D virtual simulation with real multi-agent experiment platform.The experiment is described by fusion of 3D virtual animation and information from real experiment platform and device captured by cameras.By putting the real equipment and the virtual one into the same world,the algorithm can be effectively compared in the difference between the virtual device and the real one.