Machine Learning–Based Multi-Point Load Sensing for Smart Skins

Artificial skin with tactile capabilities can greatly improve robotic interaction and task performance. Traditional touch sensors, which detect touch positions through changes in resistance or capacitance, are unsuitable for artificial skin for large deformations and fail to accurately detect force magnitudes. A more effective approach is monitoring the strain field induced by touch on a surface. While complete strain field data would enable easy touch detection, the real challenge lies in achieving this with limited data. In this study, a machine learning approach is introduced that predicts touch positions and magnitudes using sparse strain data. This approach has been validated both numerically and experimentally, demonstrating accurate prediction of single and multiple touch points on a flat surface. Additionally, the machine learning model is extended to predict touch on a flexible, shape-changing surface under bending and torsion. The proposed machine learning-based touch prediction technique has promising applications in flexible touch screens, robotic smart skin, and micro-scale touch sensors.