Novel Tactile-Texture Sensing System with Super-Human Sensing Performance via Deep Learning

In this study, we combined a highly sensitive tactile sensor with higher force resolution <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(&lt; 12\mu \mathrm{N})$</tex> and spatial resolution <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(&lt; 50\mu \mathrm{m})$</tex> than fingertip skin with deep learning, resulting in accurate identification of 14 types of tissue paper with 98.2% accuracy. We also developed a system that learns their market prices and predicts the in-market prices of unlearned samples. The regression model was a convolutional neural network (CNN). 32 measurement data were obtained for each sample, and the generalization performance for unknown tissue papers was evaluated by cross-validation. The results showed that the model was able to predict prices within an error range of approximately 7% at best and 30% overall with respect to market prices, and that market prices and prices predicted by the model captured market price trends with good accuracy. This sensing system is the first demonstration result showing that the combination of a high-performance tactile sensor and deep learning can realize superhuman tactile capabilities.