Vibrotactile Signal Generation from Texture Images or Attributes using Generative Adversarial Network

TL;DR

This paper introduces a novel deep learning approach using conditional GANs to automatically generate vibrotactile feedback from texture images or attributes, enhancing haptic interaction realism.

Contribution

It is the first to leverage GANs for vibrotactile signal generation based on texture data, enabling automatic and realistic haptic feedback creation.

Findings

Users could not distinguish generated from real signals

Generated signals felt realistic to users

Model adapts to various textures and attributes

Abstract

Providing vibrotactile feedback that corresponds to the state of the virtual texture surfaces allows users to sense haptic properties of them. However, hand-tuning such vibrotactile stimuli for every state of the texture takes much time. Therefore, we propose a new approach to create models that realize the automatic vibrotactile generation from texture images or attributes. In this paper, we make the first attempt to generate the vibrotactile stimuli leveraging the power of deep generative adversarial training. Specifically, we use conditional generative adversarial networks (GANs) to achieve generation of vibration during moving a pen on the surface. The preliminary user study showed that users could not discriminate generated signals and genuine ones and users felt realism for generated signals. Thus our model could provide the appropriate vibration according to the texture images or the attributes of them. Our approach is applicable to any case where the users touch the various surfaces in a predefined way.