Design And Research Of Multi-functional Humanoid Finger Tactile Sensor Sensing System

In unstructured environment,tactile information is indispensable for dexterous manipulation of robot astronauts,while tactile sensor is an important source of tactile information.However,the existing robot tactile sensors have some shortcomings,such as low reliability and lack of unified calibration methods.This paper is devoted to the development of a micro tactile sensor sensing system which can achieve multi-functional detection,easy calibration and maintenance.It can provide more reliable tactile information for the dexterous operation of robots astronaut.Firstly,the advantages and disadvantages of various tactile sensors are analyzed,and a humanoid finger tactile sensor sensing system based on conductive liquid is proposed.The tactile sensor system includes tactile sensor,signal acquisition system and tactile data fusion algorithm.Then,according to the biomedical structure of human hand,the structure optimization design of tactile sensor and the electrical design of sensitive elements are completed.The tactile sensor in this paper includes independent flexible skin,electrode array attached to the surface of phalanx and conductive liquid between flexible skin and phalanx.It can measure the physical information such as temperature,three-dimensional force and micro-vibration at the same time.On this basis,the simple fabrication and efficient assembly of tactile sensors are studied.Afterwards,a signal acquisition system including hardware system,software system and calibration tooling is built.After calibrating the modal information,independent experiments of hardness identification,thermal conductivity identification,surface roughness and texture fineness identification,and object local size identification are completed.Finally,based on Tensor Flow deep learning framework,a convolution neural network(CNN)classification model including four layers of convolution layer,two layers of pooling layer and two layers of full connection layer is established.Based on the five-mode tactile signal of tactile sensor raw data,a 32*32 haptic image is constructed.The CNN model achieves the classification of object categories by extracting various features of multi-mode haptic image.Experiments show that under sparse samples,Adam optimizer,getting a higher F1 score,has better recognition effect than SGD optimizer,which means the CNN model designed in this paper can realize the classification and recognition of targets well.