Research On Multi-agent Coverage Control Under Communication Connectivity Maintenance

Multi-agent coverage control theory is widely used in complex mission scenarios such as environmental monitoring,personnel rescue and battlefield reconnaissance.It is a research hotspot in the field of multi-agent cooperative control.According to the environment information,the coverage control will drive the agents to disperse as much as possible to achieve better spatial coverage effect and realize the optimal monitoring of the area cooperatively.In this process,the behavior coordination between agents depends on information data exchange,that is,the wireless communication network needs to be connected.However,in the complex electromagnetic environment,the attenuation of signal propagation and strong electromagnetic interference will limit the communication of agents,which means the communication ranges of the agents are limited.Therefore,the spatial dispersion behavior caused by coverage control will lead to the interruption of communication links between some agents.This may cause the communication network to lose connectivity and fail the task.As the basis of cluster information interaction,the purpose of communication connectivity maintenance is to plan the trajectories of the cluster so that it can still maintain the connectivity of the network in motion.Therefore,communication connectivity maintenance is the key to successfully implement the coverage control task in the complex electromagnetic environment.Taking algebraic connectivity as connectivity index,this thesis focuses on the communication connectivity maintenance in coverage control tasks under known environment model and unknown environment model respectively.The main research works are summarized as follows:(1)In the existed coverage control researches with known environment model,most of them realize the coverage control under connectivity maintenance by integrating coverage and communication connectivity maintenance control law,but fail to consider the influence caused by connectivity maintenance control to the coverage effect.Therefore,this thesis firstly proposes a distributed bounded connectivity maintenance control law and compensates the connectivity estimation error.On this basis,a piecewise control method based on critical agent identification is proposed to reduce the influence of connectivity maintenance control to coverage control effect.Finally,aiming at the "deadlock" phenomenon that the agents fall into the local optimal solution,an automatic deadlock detection and elimination mechanism are proposed in this thesis.Simulation results show the effectiveness of the proposed methods.(2)In the existed coverage control researches with unknown environment model,multi-agent reinforcement learning method is mostly used to solve the coverage control law with communication connectivity constraints,but it needs to constantly learn through trial and error,and there is no security guarantee.In this thesis,communication connectivity maintenance and collision avoidance are regarded as security constraints,and the shielding mechanism in safe multi-agent reinforcement learning is applied.Firstly,aiming at the controller design problem with discrete decision space,a multi-agent Qlearning method is proposed,in which the shielding mechanism monitors the learning process in real time,rejects the unsafe actions as well as gives punishment.And the method of fixed sparse representation is used to reduce the spatial complexity of the algorithm.Secondly,for the controller design problem with continuous decision space,combined with the control barrier function,a multi-agent deep deterministic policy gradient control method is proposed,which seeks a compromise between target search and security constraints.Simulation results show that the proposed methods can maintain the security of learning and achieve the optimal coverage.(3)Aiming at the experimental verification requirements of communication connectivity maintenance algorithm,this thesis combines the robot operating system with the motion capture system,takes the four rotor UAV as the controlled object,and builds a multi-agent semi-physical verification platform.This thesis shows the implementation method of the algorithm in the actual coverage scene,and verifies the effectiveness of the algorithm through the experimental results.