基于深度信息的多目标抓取规划方法研究  

Research on grasp planning of multi-objects based on depth data

作　　者：颜培清 何炳蔚[1] 雷阿唐 张立伟[1] 韩国强[1] 

机构地区：[1]福州大学机械工程及自动化学院,福州350116

出　　处：《电子测量与仪器学报》2016年第9期1342-1350,共9页Journal of Electronic Measurement and Instrumentation

基　　金：国家自然科学基金(61473090)资助项目

摘　　要：针对复杂场景中的多目标物抓取规划问题,提出结合抓取顺序、抓取姿势与目标物可操作性的抓取规划方法。该方法通过获取场景三维点云构建抓取空间和物体空间以此分析目标物的可操作性,然后利用所提出的抓取评价函数计算出所有抓取顺序与姿势下的目标物的可操作性评价值,以最高评价值对应的抓取方案作为机器人对场景中多目标物的最佳抓取策略。最后,以华硕Xtion Pro Live和仿人机器人Nao为该研究方法的实施平台,进行多目标抓取规划的实验测试,结果表明所提出方法能够有效解决场景中存在目标物干涉情况下的多目标物抓取规划问题,并提高机器人对多目标场景的处理效率。A new grasp planning method based on grasp sequence, grasp posture and manipulability of objects for the multi-targets in the complex scene is proposed. Manipulability of objects was analyzed by building actual grasp space and object space according to point cloud. The evaluation value of objects＇ manipulability in all of grasp sequence and posture were computed using grasp evaluation function. A grasp solution that has the highest score in the grasp evaluation was chosen as the grasp strategy of robot in clutter. Our approach was tested with experiments on ASUS Xtion Pro Live and Nao humanoid robot. The results show that the introduced method could solve the problem of interference between objects and increase the efficiency of the robot.

关 键 词：多目标 抓取顺序与姿势 抓取空间 物体空间 可操作性 

分 类 号：TN29[电子电信—物理电子学]