Abstract:
With the shortages experienced in the labor market, coupled with the growth of the agricultural industry, the need for autonomous and intelligent harvesting solutions has been steadily rising. Naturally, the field of robotics has ingrained itself into the agricultural sector by presenting the needed solutions. Soft robotics has recently begun to play a significant role, as its compliant, flexible structure, which is capable of delicately interacting with the environment, allows it to handle delicate objects such as ripe fruits and vegetables with ease. This work focuses on an artificially intelligent 3D printed soft robotic gripper with embedded pneumatic sensing chambers capable of categorizing tomatoes during harvesting. The ripeness identification process involves two stages: a data collection stage and a classification stage. In the first stage, a closed-loop pressure/force control is used to squeeze the tomato with the gripper, and the resulting pressure versus displacement data is recorded and fed to the custom-designed neural network (NN) in the second stage. The developed NN follows a layered structure based on a 1D convolutional neural network (CNN) architecture. The final model achieves a five-fold cross-validation accuracy of 85.87%, with real-time deployment an accuracy of 80.55%. This two-stage approach mimics human behavior of assessing the ripeness of fruit, which involves gently applying pressure to the fruit to identify its stiffness through touch and then handling the produce accordingly. This proposed gripper and the developed NN present a reliable and nondestructive solution for produce handling, both in the harvesting and quality control stages.
