Distributed Circle Formation Control For Multi-agent Systems With Communication Constrain

In recent years,numerous research works on control of multi-agent systems(MAS)have been witnessed due to both its practical potential in variety of applications,such as environment monitoring,rescue,and source location.As an important problem of cooperative control of MAS,formation control has attracted a great interest for its widespread applications in various areas,which aims at guiding each agent's state such that a predetermined geometry can be formed and maintained.In many application situations of MAS where the components are often physically distributed in a wide range of areas,then the majority of attention has been devoted to the distributed framework in the design of formation control strategies,and less attention has been paid to the limited capacity of communication and together with the poor operating environment.Toward this end,the main contents of this thesis are given as follows:Firstly,this paper investigates distributed circle formation problems of MAS subject to limited information communication under a class of weight-unbalanced directed graphs,in which the communication topology of MAS contains a directed spanning tree and each agent can only perceive the distance from itself to the nearest neighbor in counterclockwise direction as well as the counterpart in the clockwise direction through communication.Toward this end,a new algorithm combined with encoder-decoder technique has been proposed.We show that,under the proposed policy,the resulting network executions can drive the states of all mobile agents to converge to some expected equilibrium point with an exponential convergence rate.Numerical simulation results have been given to demonstrate the feasibility of the proposed algorithm.Secondly,this paper proposes distributed event-triggered algorithmic solutions to circle formation problems of first-order and second-order MAS,where the communication topology of networks is described by weight-unbalanced,strongly connected digraph.In particular,the controller updates and communications between neighboring agents considered here are event-triggered,depending on the specified threshold of a certain measurement error.We show that,under the proposed policy,the resulting network executions can drive the states of all mobile agents to converge to some expected equilibrium point.Furthermore,we show that the Zeno behavior can be avoided under the proposed control laws.Numerical simulation results are given to illustrate the effectiveness of the proposed methods.Finally,consumptions of communication and computing capabilities were key factors for investigating reinforcement learning formation for multi-agent,which have to be considered seriously.To this end,a novel algorithm combined with event-triggered mechanism has been proposed,where agent has not needed for updating action-decisions periodically,but for depending on the event-triggered condition.Concretely speaking,both the sum of discounted total reward and the variance of current rewards have been considered in the process of designing event-triggered condition.In addition,a joint optimization strategy has been obtained by exchanging information among learning world.Numerical simulation results have been given to demonstrate the effectiveness of the proposed algorithm and provide comparison with the traditional reinforcement learning algorithm.