Numerical and experimental study of a bio-inspired whisker sensor for tactile surface perception in robotic applications
Abstract
This study presents the design and evaluation of a bio-inspired piezoelectric whisker sensor system for tactile surface-profile classification and contact-height estimation through dynamic tactile interaction and data-driven analysis. Drawing inspiration from mammalian vibrissae, the sensor integrates piezoelectric transducers with a flexible whisker structure via a silicone coupler, mimicking the biological follicle. Finite-element analysis (FEA) was performed using COMSOL Multiphysics software to investigate stress distribution, sensitivity, and coupling performance across different materials and whisker geometries. The fabricated sensor was characterized for key performance metrics including sensitivity, hysteresis, repeatability, response time, drift, and fatigue life. A custom-built test stand equipped with a laser Doppler vibrometer (LDV) enabled controlled evaluation across different excitation frequencies and displacement conditions. Experimental results showed acceptable repeatability, stable drift behavior within ±8%, and sensitivity values up to 15.04 V/mm, with the most favorable sensor response within the tested parameter range obtained at 12 Hz using a 12 mm whisker. A convolutional neural network (CNN) hybrid with a long short-term memory (LSTM) and feedforward neural network (FNN) model was trained on tactile piezoelectric voltage time-series data to simultaneously classify contour geometry and estimate contact height. Under the controlled measurement conditions used for model development, the system achieved contour-classification accuracy above 98% with sub-millimeter contact-height estimation error. Robustness was further evaluated using an independent noisy test dataset comprising 200 recordings. Under noisy sensing conditions, the model achieved an overall contour-classification accuracy of 81.5% and a contact-height mean absolute error (MAE) of 0.504 mm, although prediction errors increased for larger contact heights. The proposed design demonstrates a proof-of-concept pathway toward AI-integrated tactile perception systems for autonomous navigation and tactile surface perception in robotic applications where conventional optical or vision-based sensing is constrained.
