Learning to Feel Textures: Predicting Perceptual Similarities From Unconstrained Finger-Surface Interactions

Area of Science:

• Neuroscience
• Haptics
• Robotics

Background:

• Human tactile perception relies on complex sensory information from finger-surface interactions.
• Previous research often overlooked the dynamic aspects of touch, limiting understanding of texture perception.

Purpose of the Study:

• To develop an interpretable model for predicting perceptual similarity of surfaces based on interaction dynamics.
• To investigate how the brain aggregates sensory information during active touch for texture representation.

Main Methods:

• Utilized physical signals: finger motion, contact force, and fingernail acceleration during unconstrained interactions.
• Compared probability distributions of features extracted from short time windows.
• Developed an interpretable modeling method to predict perceptual similarity.

Main Results:

• The model successfully predicted individual participants' surface similarity judgments (max Spearman's correlation of 0.7).
• Identified that different individuals weight interaction features differently in their judgments.
• Demonstrated the importance of interaction dynamics in texture perception.

Conclusions:

• The findings offer novel insights into human texture perception during active touch.
• The developed approach has potential applications in haptic surface assessment, robotic tactile perception, and haptic rendering.