A UAV COOPERSTIVE PLANNING METHOD BASED ON MULTI-AGENT DEEP REINFORCEMENT LEARING
Human-machine cooperative control is an important way of multi-UAV task planning.Considering the requirements of cooperative interpretation of multi-UAV task environment and consistency of strategy control,we propose a UAV cooperative planning method based on multi-agent deep reinforcement learning.According to the task knowledge and behavior state,it constructed a task planner based on hierarchical agent to generate the interdependence of human-machine interaction.It designed a deep learning reinforcement method to solve the optimal strategy and cooperative control method of group behavior,and used the mixed-initiative behavior selection mechanism to evaluate the learning strategy.Experimental results show that,as an example of human-machine interaction,the proposed method can make the group perform better in the global joint action through deep reinforcement learning,and the learning speed and stability can be better than the deterministic strategy gradient method.The flight path and task of UAV can be controlled better in modes of the following,autonomous and mixed-initiative,which provides intelligent decision basis for the implementation of UAV cluster tasks.
万方数据
维普
