基于触觉传感器阵列的机械手抓取分类方法研究

Research on Classification Method of Manipulator Grasp Based on Tactile Sensor Array

为了丰富机器人的感知能力,弥补触觉信息感知不足、解决触觉识别物体存在准确率不高的问题,基于磁性霍尔元件设计制作了可测量三轴正交接触力的触觉传感器,并将其集成为两种形状的触觉阵列安装在机械手上实现交互。搭建触觉识别系统对15种日常生活用品进行数据采集,采用卷积神经网络和注意力机制的触觉识别算法训练学习触觉数据,应用在识别日常生活物品分类场景中。实验结果表明,传感器的测量精度和最大重复性误差约为10-4N和1.2%,具备良好的稳定性和灵敏度;设计的触觉识别算法的平均识别准确率最高可达94.49%,能够对抓取到的物体进行准确分类,为机器人后续的探索交互任务奠定基础。

In order to enrich the perceptual ability of the robot,make up for the lack of tactile information perception and solve the problem of low accuracy of tactile object recognition,a tactile sensor is designed and fabricated based on the magnetic Hall element,which can measure the three-axis orthogonal contact force.The tactile array is integrated into two shapes and mounted on the robot hand to realize the interaction.The haptic recognition system is built to collect data of 15 kinds of daily articles.The haptic recognition algorithm of convolutional neural network and attention mechanism is used to train and learn the haptic data,and the haptic data is applied to recognize the classification scene of daily articles.The experimental results show that the measurement accuracy and maximum repeatability error of the sensor are about 10-4 N and 1.2%,and the sensor has good stability and sensitivity.The average recognition accuracy of the designed tactile recognition algorithm is up to 94.49%,which can accurately classify the captured objects and lay a foundation for the subsequent exploration and interaction tasks of the robot.