Data-driven modeling of isotropic haptic textures using frequency-decomposed neural networks

This paper presents a new approach to data-driven modeling of isotropic haptic textures using frequency-decomposed neural networks from the contact acceleration data that are captured when a stylus is scanned on a textured surface with diverse scanning velocities and normal forces. We first describe a motorized texture scanner that was developed for accurate and easy data collection under a wide variety of conditions. We then propose two neural network models with different topologies: a unified model that feeds all of acceleration data, scanning velocity, and normal force as input variables to a single large neural network and a decomposed model that consists of a number of smaller neural networks each of which is trained with the acceleration data for a pair of scanning velocity and normal force. An experiment with real samples showed that the unified model has better cross-validation ability in terms of spectral rms errors and its performance is comparable to the best available in the literature. We also present some preliminary results of anisotropic texture modeling achieved by extending the unified model.