| With the rapid development of UAV technology and its increasingly complex application scenarios,it has become a trend for multi-UAV to cooperate to complete tasks.In different scenarios,how to assign tasks and give full play to the advantages of each UAV to improve the task completion rate is the key for multi-UAV cooperation to complete tasks.At present,the research of task assignment algorithm focuses on heuristic algorithm and reinforcement learning algorithm.Heuristic algorithms are mostly used in static scenes.Although this type of algorithm is simple and intuitive,it is easy to fall into local optimum.Reinforcement learning algorithms can solve the task allocation problem in dynamic environments,but most of the existing research results are aimed at single UAV.It is difficult to directly apply to multi-UAV task assignment scenarios.In view of the above problems,this paper researches and improves the multi-UAV cooperative task assignment algorithm in different scenarios.The specific research work is as follows:(1)Firstly,the task allocation of multi-UAV cooperative transportation and delivery in static scenes is studied.In this scenario,with the goal of maximizing the UAV task completion rate and minimizing the total cost,balanced and unbalanced task assignment models are established respectively.Aiming at the defect that particle swarm optimization is easy to fall into local optimum,this paper proposes an improved particle swarm optimization algorithm incorporating mutation operation to solve the task assignment model.It can be seen from the simulation results that the proposed algorithm can effectively solve the problem of multi-UAV cooperative task assignment in this scenario.Finally,the results are compared with other algorithms to verify that the algorithm has better performance in multi-UAV transportation and delivery scenarios.(2)The task allocation problem of multi-UAV in dynamic scenes with unknown information is studied.In a dynamic environment,the motion of the UAV exacerbates the non-stationarity of the environment,which makes the single-agent reinforcement learning algorithm unable to solve it effectively.In response to this problem,the paper first describes the dynamic process of multi-UAV task assignment as a stochastic game process,and elaborates the process in detail from four aspects:state,action,reward and state transition probability.Then,a multi-agent deep reinforcement learning algorithm MA-SAC(Multi-agent Soft Actor-Critic)algorithm that combines the SAC(Soft Actor-Critic)algorithm and the centralized training and decentralized execution framework is proposed to solve the process.The simulation results show that the proposed MA-SAC algorithm has certain advantages in the final convergence results when solving the multi-UAV cooperative task assignment problem in a dynamic environment. |
