TexSenseGAN: A User-Guided System for Optimizing Texture-Related Vibrotactile Feedback Using Generative Adversarial Network

This repository contains the code for the paper: TexSenseGAN: A User-Guided System for Optimizing Texture-Related Vibrotactile Feedback Using Generative Adversarial Network

The opendataset used this paper: LMT Haptic Texture Database (108 surface materials, SoundScans, Movement)

To obtain the preprocessed dataset, run the notebook preprocess.ipynb. In this study, we selected 14 classes to build a training dataset.

Run the TactileCAAE/train.py to train the model. The dictionary of the trained model parameters are saved in TactileCAAE. After loading the trained parameters, the model can be used directly for the user optimization.

Run the DSS_Experiment_UserInitialization.py to start the optimization with the user initialization. Run the DSS_Experiment.py to start the optimization directly.

Citation

If you find this repo is helpful, please cite:

@ARTICLE{10891204,
  author={Zhang, Mingxin and Terui, Shun and Makino, Yasutoshi and Shinoda, Hiroyuki},
  journal={IEEE Transactions on Haptics}, 
  title={TexSenseGAN: A User-Guided System for Optimizing Texture-Related Vibrotactile Feedback Using Generative Adversarial Network}, 
  year={2025},
  volume={},
  number={},
  pages={1-15},
  keywords={Vibrations;Optimization;Generative adversarial networks;Vectors;Generators;Deep learning;Training;Human in the loop;Haptic interfaces;Aerospace electronics;Haptic display;Human-computer interaction;Optimization;Deep learning;Autoencoder;Generative adversarial networks},
  doi={10.1109/TOH.2025.3542424}}

Acknowledgments

This code is based on the implementations of Difference-Subspace-Search.