Intelligent Vehicle Group Chasing Based On Multi-agent Reinforcement Learning

With the rapid development of machine learning and intelligent robot research,reinforcement learning algorithms are increasingly outstanding in artificial intelligence technology,and the application direction is expanding.At the same time,robots have taken the place of human beings in many aspects,such as dangerous work,simple but a lot of repetitive work,battlefield operations and so on.Especially for the future war,people want to use military robots to replace human beings as much as possible to reduce casualties,so it is important to use robots to accomplish task independently in wars.This paper mainly uses reinforcement learning algorithms,combined with deep learning and multi-agent system,and three kinds of simulation chasing experiments are designed to study from single agent to complete a specific task,to multi-agents to complete more complex tasks,and Deep Q Network(DQN)algorithm and Multi-Agent Deep Deterministic Policy Gradient(MADDPG)method are applied respectively.In terms of simulation environment,the three experimental environments are built based on the Gym library,and appropriate reward functions are proposed respectively.In addition,in view of the situation that most reinforcement learning research only remains in the simulation stage,and considering the possible chasing task in military wars,the simulation results of three kinds of experiments are applied to hardware devices to provide technical support for them.Hardware devices are demonstrated by the Dashgo D1 intelligent chassis and the KR-MKNM01 intelligent car,and the location of the agents are captured in real time by Opti Track positioning system.From the experimental results,the three experiments have achieved the expected chasing effect,which proves the rationality of our experimental design and the accuracy of the reward function.