CAJ | 학술논문

针对码垛机器人应用环境状况较复杂、不确定条件较多的问题,使用基于Markov对策的算法对多关节码垛机器人进行路径规划。首先根据实际的工作环境设定机器人的运动范围,并选择经常出现的动作组合作为机器人运动的基本行为集,给出各种情况可能获得的报酬值,依据多智能体Q值学习算法更新每个关节的报酬值,反解出对应最大报酬值的动作组合,选择部分动作组合可以减少各关节之间的协调关系,降低算法的复杂度。仿真绘制出最佳动作组合时的运动轨迹,以及机器人运动环境中无障碍与放置球形障碍物时的三维运动轨迹,并确定轨迹的误差。最后经过实验验证表明,多智能体Q值算法能有效地控制各个关节的协调运动,实际运动的误差在允许的范围内,满足使用要求。
침대마타궤기인응용배경상황교복잡、불학정조건교다적문제,사용기우Markov대책적산법대다관절마타궤기인진행로경규화。수선근거실제적공작배경설정궤기인적운동범위,병선택경상출현적동작조합작위궤기인운동적기본행위집,급출각충정황가능획득적보수치,의거다지능체Q치학습산법경신매개관절적보수치,반해출대응최대보수치적동작조합,선택부분동작조합가이감소각관절지간적협조관계,강저산법적복잡도。방진회제출최가동작조합시적운동궤적,이급궤기인운동배경중무장애여방치구형장애물시적삼유운동궤적,병학정궤적적오차。최후경과실험험증표명,다지능체Q치산법능유효지공제각개관절적협조운동,실제운동적오차재윤허적범위내,만족사용요구。
On account of complex application environments and large number of uncertain conditions for a palletizing robot,a path-planning method for multiple joints robot was presented by the algorithm based on Markov game.At first,according to the actual working environment,the range of the robot＇s motion was set and the conventional movement combination was selected as the basic set of the robot＇s behaviors.The possible reward of various situations would be obtained.Then the reward of each joint can be updated by multi-agent Q-learning algorithm and inverse the movement combination corresponding with the best reward.Selection of the movement combination parts can reduce the coordination among each joints and the complexity of the algorithm.The best motion trail will be shown,including the 3D motion trail when it＇s barrier-free and there was a spherical obstacle,and determination of the trail errors.At last,after experimental verification,the algorithm has been proved to control the compatible movements of each joint effectively and keep the errors within the allowed ranges.The experiments meet the requirements well.