Smart Electronic Skin Having Gesture Recognition Function By LSTM Neural Network

In recent years,the rapid development of robotics and bionics has enabled robots to imitate and even surpass human behavior in auditory and visual aspects.Due to the complex and changeable application scenarios of robots,the need to improve the tactile signal sensing ability of robots is becoming increasingly prominent.Electronic skin can be simply defined as an electronic system consisting of different types of sensors.Robots or prostheses can interact with the outside world through electronic skin.Although electronic skin has made great achievements in tactile function,there are still some problems in practical application.The preparation of electronic skin sensor involves high design complexity and equipment cost,which is unreasonable for application scenarios where production capacity and cost are the main concerns.In addition,the existing research on electronic skin mainly embodies in the synthesis of polymer materials and the structural design of sensor units.The intelligent characteristics of electronic skin have received little attention.This thesis presents a simple,low-cost and intelligent electronic skin system,which consists of capacitive sensor array,data acquisition circuit and data processing.Polydimethylsiloxane(PDMS)thin films were prepared by hand.PDMS thin film is the dielectric layer of capacitive sensor array.The sensor array is designed to include 35capacitive touch units,each of which is a sandwich structure.Due to the flexible,bendable and stretchable properties of the PDMS film,the sensor array can be attached to an object surface with irregular shapes.In order to achieve the conversion of pressure/touch signals to electrical signals,designed data acquisition circuit module converts the capacitance change of the capacitive sensor array into a voltage signal.Capacitive touch sensor controller can detect changes in capacitance.Data transmission between the capacitive touch sensor controller and the data acquisition platform is accomplished via the I~2C bus.In order to endow the electronic skin with intelligence perception ability,the Long Short-Term Memory(LSTM)algorithm was introduced.LSTM is a neural network algorithm based on time series modeling.After reasonable LSTM neural network algorithm design,the training accuracy of the neural network reaches 100%.The designed electronic skin system can correctly recognize the four gestures.In order to investigate the stability of the gesture recognition rate of the electronic skin system,500 repetitive experiments were carried out.The experiment verified that the gesture recognition rate of the electronic skin system was stable above 80%.In addition,the effects of folding,curvature,tensile strength and temperature on the recognition rate of electronic skin system were investigated.The experimental results show that the external conditions have little influence on the gesture recognition rate of electronic skin system.