The Collaborative Combat of Heterogeneous Multi-UAVs Based on MARL

In the course of collaborative combat of multi-UAVs, there appear so many unpredictable fluctuations in the combat situation, and the mission assignment and decisionmaking of the combat are sequential and game-oriented, rendering it hard to build its dynamic and accurate model of operational missions. This paper breaks through the traditional research idea of mission modeling and optimization solution of collaborative operations of UAVs, and applies multi-agent reinforcement learning to the decision-making of multi-UAV collaborative combat, builds a verification environment of collaborative air-to-ground penetration and attacks of heterogeneous UAVs by designing MARL algorithm architecture, situation model, decision model and reward and punishment model, realizes the autonomous evolutionary learning of mission assignment and decision-making, and finally analyzes and evaluates the effectiveness and robustness of different MARL algorithms.