Predicting Perceptual Haptic Attributes of Textured Surface from Tactile Data Based on Deep CNN-LSTM Network

This paper introduces a framework to predict multi-dimensional haptic attribute values that humans use to recognize material properties using physical tactile signals (acceleration) generated when a textured surface is stroked. To this end, two spaces are established: a haptic attribute space and a physical signal space. A five-dimensional haptic attribute space is established through human adjective rating experiments with 25 real texture samples. The physical space is constructed using tool-based interaction data from the same 25 samples. A mapping is modeled between these spaces using a newly designed CNN-LSTM deep learning network.

A prediction algorithm is implemented that takes acceleration data and returns coordinates in the haptic attribute space. A quantitative evaluation was conducted to inspect the reliability of the algorithm on unseen textures, showing that the model outperformed other similar models.

Download (PDF)

@inproceedings{awan2023predicting,
  title={Predicting Perceptual Haptic Attributes of Textured Surface from Tactile Data Based on Deep CNN-LSTM Network},
  author={Awan, Mudassir Ibrahim and Hassan, Waseem and Jeon, Seokhee},
  booktitle={Proceedings of the 29th ACM Symposium on Virtual Reality Software and Technology},
  pages={1--9},
  year={2023}
}
  

For more information, explore the full paper on ResearchGate, ACM, or Google Scholar.