基于强化学习的多Agent路径规划方法研究

以复杂任务下多个智能体路径规划问题为研究对象,提出一种基于强化学习的多Agent路径规划方法。该方法采用无模型的在线Q学习算法,多个Agent不断重复“探索学习利用”过程,积累历史经验评估动作策略并优化决策,完成未知环境下的多Agent的路径规划任务。仿真结果表明,与基于强化学习的单Agent路径规划方法相比,该方法在多Agent避免了相碰并成功躲避障碍物的前提下,减少了17.4%的总探索步数,形成了到达目标点的最短路径。

MULTI-AGENT PATH PLANNING BASED ON REINFORCEMENT LEARNING

Taking multiple agents path planning problems under complex tasks as the research object,we proposed a multi-agent path planning method based on reinforcement learning.The method adopted a model-free online Q learning algorithm.In this method,a model-free online Q-learning algorithm was adopted.Many agents repeated the process of "exploration- learning-utilization",accumulated historical experience,evaluated action strategies and optimized decision-making,and completed the task of multi-agent path planning in unknown environment.The simulation results show that compared with the single agent path planning method based on reinforcement learning,this method reduces the total exploration steps by 17.4% and forms the shortest path to the target point on the premise that multi-agent avoids collision and successfully avoids obstacles.