Graphene-based tactile sensors and its texture recognition system

Abstract

Tactile recognition is an enormous challenge for development of touch interface between human and machine to attain an artificial intelligence describing touch feeling which is one of five senses. As described by David Katz in the duplex theory of tactile texture perception, texture recognition is very complex process requiring sensitive detection of both pressure and vibration to mediate spatial encoding of coarse textures arising from the geometrical properties of the texture and vibrotactile encoding of fine textures. Although emulating surface texture recognition has been continuously studied, there are two major difficulties in current technology that is far away from achievement of comparable level to human tactile perception system. One is to find a proper sensing device providing equivalent capability to human mechanoreceptors with flexible architecture in order to represent information of complex texture on the contacted objects. Currently available approach, especially developed for robotics, has applied the accelerometer to generate touch feeling by detection of vibrations induced by the interacting surfaces. However it doesn’t only be able to represent fine texture, but also be limited on integration to soft electronics because of its rigid nature. On the other hand several demonstrations have recently been proposed for the force sensor exploiting the electro-mechanical character in specially designed polymers or nano-materials. The results of these studies make expectation that it may be possible to find a proper sensor device for emulating tactile sensation according to human perception. Even if the sensor acting as the mechanoreceptor is accessible, the other issue is how to classify the practical textures that human recognizes through experience. To address, in this doctoral thesis, I demonstrated graphene-based tactile sensors and its texture recognition system. Tactile sensors were fabricated by introducing single-layer graphene (SLG) or graphene flake (GF) with unique architectures. For the SLG tactile sensor, the force sensor coated with SLG as piezoresistive material on flexible substrate was first fabricated to convert the structural surface deformation to a resistance change. The resistance changes due to local deformation of a local area of the single layer graphene are reflected in the resistance of the entire sensor. The generic feature of SLG representing a continuous film with single atomic thickness has a great advantage for conserving the piezo-resistive response due to the local deformation in the entire SLG film. By introducing microstructures inspired by human finger prints, surface texture was successfully defined through fast Fourier transform analysis, and spatial resolution was easily achievable. This work provides a simple method utilizing single sensor for surface texture recognition at the level of human sensation without using matrix architecture which requires high density integration technology with force and vibration sensor elements. Furthermore, it allows the tactile sensor for applicable to the electronic skin with sufficient texture resolution corresponding to recognition capability of human skin. In another case, a flexible force sensor was designed with thin film of GF which was introduced as a piezo-resistive material to generate resistance output for detection of force itself and difference in force since electrical connections between GFs sensitively varies with structural deformation of the film by compressive force. A ridge structure, which is named as artificial finger-print structure (AFPS), was introduced on Polyethylene terephthalate (PET) by lithography to obtain periodic line patterns with a width of 300 μm, height of 70 μm, and spacing of 300 μm. The sensor was completed with assembling the AFPS tissue on the GF film with electrodes. With a naturally formed porous structure of GF film, the sensor shows not only sensitive response for low pressure (< 10 kPa), but also response time as fast as 2 ms for deformation and 1 ms for restoration, which are critical factors for tactile sensing. Using the above mentioned characteristics, the sensor attempted to recognize texture and softness that are major functions for tactile recognition. The electrical texture signals detected using graphene-based tactile sensors were quantified to recognize and classify the contacted textures. Texture was recognized by detecting interacting vibrations induced by physical contact due to rubbing motion with texture material on the sensor assembled with the AFPS. Softness was included in the sensor output with differences of resistances and response times, depending on stiffness of contacted object. Like this, texture with periodic roughness is easily recognized or classified by the analysis of the power spectrum in frequency domain converted from the sensor outputs by taking discrete time Fourier transform (DTFT). Contrast to the periodic roughness, recognition and classification for complex texture with aperiodic roughness in amplitude and spatial interval requires a new tool for quantifying the texture, and I here introduce a support vector machine (SVM) which is one of machine learning techniques. A method of data collection for building up the feature vector for training- and test-sets of SVM was developed to extract texture-dependent inherent features using area under curve (AUC)-based feature selection technique. 100% of classification accuracy was obtained with comparison of only first principal 15 features for 4 different testing samples which can be apparently distinguishable tactile patterns by human sense. By expanding, 12 testing fabrics that human tactile sense actually classified each fabric with 57.8 % accuracy were completely classified by using the 18 features. This complete classification capability demonstrates that the senor can recognize the tactile information of contacted object which is distinguishable at least for the testing samples. Finally, prediction for unknown fabric was carried out through comparison with learned features using SVM for the experienced 12 fabrics, and the results prove that the sensor classifies the fabrics as similar to human tactile perception. To realize touch interface devices, the electrical output signals of the sensor were attempted to convert to the signals that human can perceive. Such a converting is based on the neural spike signals triggered with mechanical stimulation on biological skin. The sensor signals were transformed to neural signals by considering the response characteristics of slow adaptive (SA) and fast adaptive (FA) receptors, thus the converting system consists of neural stimulator. It was confirmed that the transformed neural signals were transmitted normally between nerves of rat without signal distortion. With the tactile recognition system that is composed of tactile sensor, neural stimulator, and bioelectrical stimulation system, the pressure and texture information signals of objects on the sensor were finally succeeded in stimulating the neuron electrically as well as in transmitting the stimulus signals, in real time. Then, this system is to demonstrate a tactile recognition system.