Li Zou

Decoding Human Texture Perception with Interpretable Machine Learning

When we explore surfaces through touch, our skin is stimulated by rich tactile signals that help us perceive their sensory attributes like smoothness, warmth, and softness. Yet, how these signals are transformed into perceptual experiences and material recognition remains poorly understood. In this talk, we present our work using interpretable machine learning to decode tactile signals from finger-surface interactions. We employed interpretable algorithms to investigate which signal features contribute to the elicitation of perceptual attributes and the classification of materials. Furthermore, we explored whether these perceptual attributes can reliably inform material recognition. Our findings reveal decision patterns that align with human perception, offering new insights into texture recognition and supporting the development of more perception-based haptic technologies.