基于BP神经网络的仿人两指末端执行器抓握模式预测  

Prediction of humanoid two-finger end-effector grasp type based on BP neural network

作　　者：陈小静[1] 彭培成 张高峰[1] 王裕清[1] CHEN Xiaojing;PENG Peicheng;ZHANG Gaofeng;WANG Yuqing(School of Mechanical and Power Engineering,Henan Polytechnic University,Jiaozuo 454000,Henan,China)

机构地区：[1]河南理工大学机械与动力工程学院

出　　处：《河南理工大学学报（自然科学版）》2020年第2期97-102,共6页Journal of Henan Polytechnic University（Natural Science）

基　　金：国家自然科学基金资助项目(U1261115)

摘　　要：为了得到手部特征及物体特征与两指抓握模式之间的非线性映射关系,以便对仿人两指末端执行器的抓握模式进行预测,采用5554次人手拇指-食指成功抓握试验的数据作为训练样本,构建基于L-M算法的BP神经网络两指抓握模式预测模型,进行仿人两指末端执行器的抓握模式预测。结果表明:该神经网络模型的预测准确率达90%,预测值与实测值的相关系数为0.83,能够快速有效地预测仿人两指末端执行器的抓握模式;对于等效直径较小且质量较轻的目标物,多选择精密捏;对于等效直径较大且质量较重的目标物,多选择强力握。研究结果可为仿人两指末端执行器的稳定抓握控制提供重要的决策依据。The nonlinear mapping relationship between hand and object characteristics and thumb-index finger grasp type was studied to predict the grasp type for humanoid two-finger end-effector.The data from 5554 thumb-index finger success grasping trails were used as the training samples,the predicting model of two-finger grasp type was established by BP neural network with L-M algorithm,and the grasp type for humanoid two-finger end-effector was predicted.The results showed that the accuracy of BP neural network model with L-M algorithm was 90%,and the correlation coefficient between predicted value and observed value was 0.83.The grasp type for humanoid two-finger end-effector could be predicted fast and effectively:the precision-pinch was more likely to be chosen for the small equivalent diameter and light objects,otherwise the power-grasp was more likely to be chosen for the large equivalent diameter and heavy objects.The study provided a helpful reference of decision-making for the stable grasp control of humanoid two-finger end-effector.

关 键 词：抓握模式 L-M算法 BP神经网络 两指末端执行器 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]