采用神经网络算法的多指机械手织物抓取规划  

Fabric grasp planning for multi-fingered dexterous hand based on neural network algorithm

作　　者：张蕾[1] 韦攀东 李鹏飞[1] 王晓华[1] 刘秀平[1] 

机构地区：[1]西安工程大学电子信息学院,陕西西安710048

出　　处：《纺织学报》2017年第1期132-139,共8页Journal of Textile Research

基　　金：陕西省教育厅科研计划项目资助项目(14JK1306);陕西省科技计划项目(2016GY-136);西安工程大学学科建设经费资助项目(107090811)

摘　　要：针对纺织服装行业织物自主抓取环节依靠人工操作而导致的生产效率低的问题,利用机械手进行织物抓取。设计了多指机械手,通过描述手指各连杆之间的变换关系对其进行运动学分析;采用RBF(径向基函数)神经网络方法对其抓取模式进行规划,通过识别织物的几何特征并根据抓取的任务要求进行自主抓取;在抓取运动过程中,采用关节空间轨迹规划和笛卡尔空间轨迹规划相结合的方式,保证机械手各手指能够稳定和准确地到达抓取点;最后利用Mat Lab/Robotics Toolbox对多指机械手和抓取规划进行建模仿真。结果表明,所设计的机械手各个关节参数设置合理,机械手织物抓取规划满足要求。For fabric autonomous grasp of textile and garment industry, low production efficiency will be caused by manual operation. Fabric is grasped by dexterous hand in this paper. Firstly, multi-fingered dexterous hand was designed and a method of kinematics analysis was used by describing coordinates transformation relation of fingers＇ connecting rod. Grasp mode planning was programmed by using Radial Basis Function （RBF） neural network method. By identifying the fabric＇s geometric feature and according to the requirements of the grasp tasks autonomous grasp is realized. In the process of grasp movement, the joint space trajectory planning and Cartesian space trajectory planning were combined to ensure the dexterous hand fingers can stably and accurately reach to the grasp point. Finally, multi- fingered dexterous hand and grasp planning were simulated by using MatLab/Robotics Toolbox, and the simulation results show that the design of the dexterous hand joint parameters setting is reasonable, and fabric grasp planning meets the requirements.

关 键 词：多指机械手 抓取模式规划 RBF神经网络 轨迹规划 织物抓取 

分 类 号：TP241[自动化与计算机技术—检测技术与自动化装置] TS103.8[自动化与计算机技术—控制科学与工程]