Towards Estimating the Stiffness of Soft Fruits using a Piezoresistive Tactile Sensor and Neural Network Schemes

Measuring the ripeness of fruits is one of the key challenges to enable optimal and just-in-time strategies across the fruit supply chain. In this paper, we study the performance of a tactile sensor to estimate the ground truth of the stiffness of fruits, with kiwifruit as a case study. Our sensor configuration is based on a three-beam cantilever arrangement with piezoresistive elements, enabling the stable acquisition of sensor readings over independent trials. Our estimation scheme is based on the com-pact feed-forward neural networks, allowing us to find effective nonlinear relationships between instantaneous sensor readings and the ground truth of stiffness of fruits. Our experiments using several kiwifruit specimens show the competitive performance frontiers of stiffness approximation using 25 compact feed-forward neural networks, converging to MSE loss at 10-5 across training-validation-testing in most of the cases, and the utmost predictive performance of a pyramidal class of feed-forward architectures. Our results pinpoint the potential to realize robust fruit ripeness measurement with intelligent tactile sensors.