Soft Finger-tip Sensing Probe Based on Haptic Primary Colors

Abstract

This paper describes a novel tactile sensing probe based on haptic primary colors (HPCs) and a tactile classifying system. We developed a finger-type soft tactile probe incorporating a sensor to measure three physical quantities: force, vibration, and temperature. We also constructed a tactile probe sliding system on the surface of the material repeatedly. The tactile fluctuation obtained from the tactile probe was recorded, and a frequency analyzed image was generated. In the evaluation experiments, the tactile images were generated by sliding the tactile probe on seven materials (ray fish skin, aluminum plate, rusting hemp fabric, MDF board, tatami mat fabric, acrylic board and rubber sheet). A convolutional neural network (CNN) was constructed and its classification performance was evaluated. In addition, we used tactile images to clarify the classification performance through TLAlexnet (transfer learned Alexnet). Pre-trained TLAlexnet was generated by domain adaptation using the tactile images. The results of TLAlexnet showed the great performance to be 85.0%, 91.7%, and 85.7% with respect to single primary colors of force, vibration, and temperature, respectively, and it improved to 96.4% when using three HPCs. In addition, the classification performance of the proposed seven-layered another CNN that was trained with the obtained tactile images was 98.2% of the CNN constructed using common filtering parameters. Thus, highly accurate classification was realized by using three HPCs elements.