基于角色变换和强化学习的多机器人协同仿真  

Multi-Robot Coordination Based on Role Transformation and Reinforcement Learning

作　　者：樊建[1,2] 郑昌陆[1] 费敏锐[1] 高志年[2] 

机构地区：[1]上海大学上海市电站自动化技术重点实验室,上海200072 [2]南京陆军指挥学院作战实验中心,南京210045

出　　处：《系统仿真学报》2009年第21期6964-6967,共4页Journal of System Simulation

基　　金：上海大学创新基金(A.10-0109-08-016);曙光计划跟踪项目(06GG10);上海市优秀学科带头人计划项目(08XD14018)

摘　　要：征对多移动机器人协同问题,将角色变换与强化学习相结合,采用集中式控制结构,并提出了距离最近原则,将距离障碍物最近的机器人作为主机器人并指挥其它从机器人运动,同时采用了基于行为的多机器人协同方式,在提出的基于强化学习的行为权重基础上,通过与环境交互使机器人行为权重趋向最佳,并利用基于最大行为值的协调策略来规划机器人避碰行为。通过在动态环境下多机器人协同搬运仿真实验,表明在使用了角色变换和强化学习后,有效减少了多机器人与障碍物发生碰撞的次数,成功的实现了协同搬运,具有良好的学习效果。Focusing on multi-robot coordination problem, role transformation and reinforcement learning method were combined, and the distance nearest rule was proposed which means that the nearest robot ranges from obstacles is the leader robot to control other robots movement by using centralize control framework. Behavior based multi-robot coordination was proposed, the robot behavior weight was more and more optimized through altering with environment, and the coordination policy based on maximum behavior value was used to plan the collision avoidance behavior of robot on the foundation of reinforcement learning behavior weight. Through multi-robot cooperative carrying simulation experiment in dynamic environment, it＇s shown that the collision number between robots and obstacles is effective decreased after using role changing and reinforcement learning method, cooperative carrying is successfully achieved and great learning effect is made.

关 键 词：强化学习 角色变换 多机器人 协同搬运 

分 类 号：TP42[自动化与计算机技术]