机器人视觉示教学习方法研究  

Research on Learning Method of Robot Visual Teaching

作　　者：王鹏[1] 朱建新[1,2] 朱振新 罗南安[1] WANG Peng;ZHU Jian-xin;ZHU Zhen-xin;LUO Nan-an(Central South University,State Key Laboratory of High Performance Complex Manufacturing,Hu’nan Changsha410083,China;Hu’nan Sunward Intelligent Equipment Co.,Ltd.,Hu’nan Changsha410000,China)

机构地区：[1]中南大学高性能复杂制造国家重点实验室,湖南长沙410083 [2]湖南省山河智能装备股份有限公司,湖南长沙410100

出　　处：《机械设计与制造》2019年第11期242-245,共4页Machinery Design & Manufacture

基　　金：国家科技支撑计划项目(2014BAA04B01);湖南省科技重大专项(2011XK4001)

摘　　要：针对机器人在非结构化环境的人机交互过程中环境感知与目标识别问题,提出一种基于视觉的机器人示教学习方法,机器人通过视觉感知人机交互过程中用户的单次示教获得任务操作技能。采用视觉系统捕获用户示教操作动作图像,通过图像处理与特征匹配算法SIFT提取目标对象特征。机器人复现阶段,识别出新任务环境中目标对象且向机器人输入其坐标位置。在Matlab仿真环境中模拟机器人经始末坐标逆解出机器人任务操作的运动轨迹,通过Barrett WAM机器人实验平台实现操作一组多孔积木的装配任务。研究结果表明:在单次示教与极少的环境先验知识的情况下,机器人能够学习到用户的顺序任务操作技能,并且在视觉范围内有较好的泛化能力。According to the robot in the process of human-computer interaction in unstructured environments in environmental perception and target recognition problem,we propose a vision based robot teaching and learning method,the robot through the user perception of human-computer interaction in the process of single teaching job skills.The visual system is used to capture the user’s teaching action image,and the target object features are extracted by image processing and feature matching algorithm SIFT.In the stage of robot reappearance,the target objects in the new task environment are identified and their coordinate positions are input to the robot.In the Matlab simulation environment,the robot simulates the trajectory of the robot’s task operation through the inverse coordinates,and implements a set of porous building blocks assembly tasks through the Barrett WAM robot experimental platform.The results show that,under the condition of single teaching and very little prior knowledge of environment,the robot can learn the sequence operation skills of users,and have better generalization ability in visual range.

关 键 词：机器人 示教方法 SIFT 运动轨迹 

分 类 号：TH16[机械工程—机械制造及自动化] TP242.6[自动化与计算机技术—检测技术与自动化装置]