Design And Analysis Of Hand Pressure Tactile Acquisition And Perception System

As one of the five basic sensations,tactile is an important aspect of the somatosensory system.It provides various information about the detected object under exploration,including its shape,size,deformability,softness,and texture,which is essential for humans to detect and learn the properties of different objects.The acquisition,analysis,and recognition of tactile signals are of great significance in the research of tactile perception mechanisms,the research of robotic grasping strategies,and the manufacturing of prostheses.At present,most of the systems that can be applied in the above-mentioned fields have an insufficient resolution.Therefore,this paper designs high-resolution hand tactile pressure acquisition system,uses dimensionality reduction methods to reveal the high-level properties and relationships contained in tactile signals,and gains an understanding of hand tactility.Use neural network as a black box to simulate the human perception process to realize the classification and recognition of interactive tasks.Firstly,a real-time collection system of the dynamic pressure of human hands is designed based on the human’s tactile perception mechanism.A resistive pressure sensor covering the hairless skin of the entire hand is designed.STM32 is used as the lower computer of the acquisition circuit,and design the array signal scanning circuit to collect pressure data with higher spatial resolution in real-time.And design a serial communication module to realize real-time communication between the lower computer and the upper computer.The upper computer completes the functions of reading and writing data,dimensionality reduction analysis,visual display,and classification recognition.Secondly,a collection experiment is designed,and the pressure-tactile collection system is used to obtain the changes in hand pressure distribution during the process of human grasping and sliding objects.The shape and posture trajectories in the process of grasping the object are decoded using the PCA algorithm.The results show that changes in pressure characteristics can be used to characterize the grasping process under different grasp postures.Based on TCA,the experimental factors,sensor factors,and time factors are decoded during the process of interacting with the objects.Thus we can describe the tactile interaction processes with a linear combination of an appropriate number of low-order components.Finally,in order to achieve accurate object interaction recognition based on pressure tactile information,different deep learning-based classification methods are proposed for different interactive tasks.On the one hand,an improved multi-channel CNN is designed for the classification of grasping tasks with the obvious spatial distribution.On the other hand,a TCN is designed to recognize the slide texture tasks with obvious temporal dynamic changes.The results show that the models proposed for the two types of interactive tasks finally converge after training,achieving positive classification results compared with traditional classification methods.The system designed in this thesis has broad application prospects in the field of robot grasping strategy research,it provides technical support for robots with tactile feedback,and at the same time provides a new platform for the study of tactile perception mechanisms.