UAV Reconnaissance Task Allocation with Reinforcement Learning and Genetic Algorithm

Unmanned air vehicle (UAV) reconnaissance task allocation is important in a total military combat system. The typical Genetic Algorithm (GA) is a common effective means to deal with the UAV task allocation problem. But when face with a large number of targets, the initial population has a huge influence on the performance of GA algorithms, which leads to instability on the solution accuracy. To overcome this limitation of heuristics algorithms, we propose a new algorithm combing reinforcement learning (RL) and the GA algorithms, named GA-RL. The RL is used to fast provide an initial population for GA, and then the GA algorithms further optimize this initial population to get the solution. Finally, the numerical simulation tests show that this algorithm can hugely improve the solving accuracy, especially in large tasks allocation problems.