【作者】 刘昊； 王涛； 范伟； 彭光正； 王可；

【Author】 Liu Hao Wang Tao Fan Wei Peng Guangzheng Wang Ke (School of Automation,Beijing Institute of Technology,Beijing 100081,China)

【机构】 北京理工大学自动化学院；

【摘要】 抓持位形与被抓持物体的尺寸是仿人灵巧手抓持规划的重要约束。本文设计了一种复合径向基神经网络的分类器,集成概率神经网络的分类能力和泛化回归神经网络的曲线拟合能力的优点。该分类器应用在示教再现抓持规划中,融合数据手套上传感器的信息,实现对人手抓持类型的准确判断,以及对被抓持物体的关键尺寸的拟合计算。在测量范围内,抓持分类识别率达到97%,尺寸测量最大相对误差为4%。更多还原

【Abstract】 Grasping position and the size of grasped objects are two essential constraints for grasping planning of humanoid dexterous hand.A compound RBF neural network classifier is presented,which is integrated both classification ability of PNN and curve fitting ability of GRNN.In grasping planning of teaching and playback,by means of fusing sensors information of the data glove,the classifier implements accurate identification of human hand grasping model,simultaneously measuring the key size of grasped objects.In the measurement range,grasping identification rate is up to 97%and the maximum relative error of size measurement is only 4%.更多还原