Understanding whisker array sensing using interpretable machine learning

Equipped with their whisker arrays, seals have exquisite sensitivity and intelligence for tactile and hydrodynamic sensing. While research efforts have improved the understanding of the hydrodynamically optimized whisker morphology, much less is known about how the signals from the whiskers are mapped to the environmental information being sensed. In this study, a one-way flow-structure interaction simulation setup is used to generate training data for a large-language-model based machine learning model. Whisker bending signals are collected using finite element simulations for whiskers in an array subjected to flow loading in the wake of various shapes. The flow features and their spatiotemporal correlation with whisker signals are analyzed. Arrays of whisker sensors are also being fabricated, calibrated, and assembled to generate training data to augment the training of the model. The model is trained to predict the shape and position of the upstream object based on whisker array signals. The trained model allows interpretability to gain insights into the sensing mechanism of the whisker array. It can be used in the future to power a locomotive agent for underwater applications such as trail tracking and environment mapping.