Task Assignment with Minimum Cost for Multi-UAV System via Reinforcement Learning

This paper addresses the task assignment problem for multi-UAV system in pursuit-evasion game via reinforcement learning. The targets are assigned for agents based on the principle of minimizing the total execution cost of multiple tasks. The cost and the corresponding optimal control policy of agent executing each task are solved before the task assignment process. Reinforcement learning algorithm without knowledge of the agent dynamics is proposed to solve the problems arising from high nonlinearities of agent model and parameter uncertainties caused by external disturbances. Simulation results are given to verify the effectiveness of the proposed task assignment algorithm.