Assessing fruit hardness in robot hands using electric gripper actuators with tactile sensors

Abstract

Fruit hardness is an indispensable attribute for appraising fruit quality, with notable implications for nondestructive robotic grasping, ripeness determination, and classification. This article introduces an tactile predictive recognition approach that leverages an adaptive capsule network to assess fruit hardness through manual interaction. This methodology serves as the foundation for an all-encompassing robotic system, integrating grasping and tactile acquisition capabilities. Within this framework, a vision module captures images and generates optimal grasping configurations, while an actuator endowed with tactile fingertips applies controlled pressure to collect tactile data from the fruit’s surface. It is noteworthy that the capsule model surpasses conventional convolutional neural networks in its adept utilization of temporal information. Additionally, a dedicated fruit grasping dataset is meticulously curated and employed, online grasping experiments performed on representative fruit samples, The empirical results unequivocally establish the superior accuracy of the capsule network model in hardness recognition, surpassing existing Time Series models. Consequently, our proposed scheme holds promise for engendering novel insights into the automation of fruit sorting processes and the advancement of non-destructive fruit picking methodologies.