基于多机器人系统的避碰规划再励学习研究-计算机技术专业论文.docxVIP

扬州大学硕士学位论文 扬州大学硕士学位论文 2 Abstract With the development of tlle multi-robot system technology,the ability of robots continuously is improved．and its application fields are expanded．When柚environment of cooperative mobile robots bc；c‘蛐窿s inoce complex，muhiple mobile robots systcllls which enmposed of many robots are used to complete the work which a single robot cannot complete．The mobile robots may work in the unknov／n and complex environment．Robot motion control tm．gh on reinforcement learning need not precise information of robot movements，but destination location of robots in the task．Reinforcomem Ie盯ning is丑kind ofreal—time and online learning丑Ig‘蛹lllm．Through machine learning，robots Call gain useful麟耐a鹳，and mas灯柚optimal strategy and knowledge step by step．The study in this thesis has strong application demand and application value to research of cooperation between multi-robots． For reinforuemem learning dt嘣aty il啦目疆贮姆with envimnmcm and need not environment model and a咖knowledge，it is employed in eoope咖mobile robots system．The rcinforcment function is designed with behavior-based and model—free．The movable robots in the mobile robot system avoid collision with the dynamic and static obstacles．The main contribution in this thesisFirsay,the枷ties,researchis given as follows． status,and dcvclopamm tmmtmcfive of multi—robot systems躺introduced systematically． Secondly,a modified artificial field algorithm based Oil robot spccd marginal function re(u) is presented,which is applied in coo删ve mobile robo伍systems．Tim minfmr．cmcnt function is verified in the simulmiOn tests． Thirdly,all improved Q—learning algorithm with variable attenuation factor is proposed,which combines advantages of the 8-Greedy strategy and the traditional Q-learning algorithm．The simulations show the effectiveness ofthe p|．o|posed scheme． Key words：multi·robot systems，reinforcement learning,robot learning 唐英：基丁．多机器人系统的避碰规划再励学习研究 唐英：基丁．多机器人系统的避碰规划再励学习研究 75 扬州大学学位论文原创性声明和版权使用授权书 学位论文原创性声明 本人声明：所呈交的学位