A Collaborative Unmanned System Assignment Algorithm Based on Deep Reinforcement Learning

In unmanned systems, the objective of resource assignment is to optimally distribute different types of friendly resources among target objects in order to maximize the total expected threat value to the enemy targets. Exact methods can only solve small-scale problems in a reasonable time. Although many heuristic methods have been studied in the literature to solve WTA, their multiple iterations and long solving time lead to poor practical performance. Therefore, a scheme based on deep reinforcement learning algorithm has been proposed to solve the assignment problem. An intelligent assignment framework based on Markov process is constructed, and a damage probability threshold constraint is introduced to prevent weapon resource waste. The actor-critic network is used to predict state value and improve solution quality. To test the effectiveness of the proposed algorithm, 150 problem instances (up to 100 weapons and 40 targets) have been randomly generated. The numerous simulation experiments show that the reinforcement learning algorithm outperforms Hungarian algorithm and Monte Carlo algorithm in terms of performance and the solution speed is two orders of magnitude faster than the genetic algorithm while ensuring the quality of the solution.