Abstract: The robot does not fully understand the intention of the operator in the collaboration process.Aiming at this problem,the flexible tactile sensor was used to convert the hand grasping pressure information to tactile image information.Based on the operators operating habits,a series of grasping gestures were defined to realize the flexible adjustment of robot pose.A Convolutional Neural Network(CNN)classifier suitable for tactile image recognition was constructed to realize the accurate recognition of operation intention,and a variable damping admittance control method was constructed based on grip looseness information.Through the human-machine collaboration experiment,the accuracy of intention recognition reacheed 98.80%,and the proposed variable damping admittance control could complete the human-machine cooperation task more efficiently.

Key words: tactile perception, man-machine collaboration, convolutional neural network, intention recognition, admittance control

摘要： 针对机器人在协作过程中对意图理解不充分的问题,自主设计了一种基于触觉感知的手柄作为意图理解的接口,采用柔性触觉传感器将手部抓握压力信息转换成触觉图像信息,并依据操作者的操作习惯定义一系列抓握手势实现对机器人位姿的灵活调整。构建了一种适用于触觉图像识别的卷积神经网络(CNN)分类器,并基于抓握松紧信息构建一种变阻尼导纳控制方法。最后,采用实验室自主研制的幕墙安装机器人并搭载触觉手柄进行人机协作实验,结果表明此类触觉手柄及CNN方法对操作者抓握意图的识别准确率达到98.80%,所提出变阻尼导纳控制能够更加高效的完成人机协作任务。

关键词: 触觉感知, 人机协作, 卷积神经网络, 意图识别, 导纳控制