Fruit Recognition Through Active Grasping of Two-Fingered Gripper Using a Soft Skin Sensor

Recognizing the fruit is a critical function to achieve for an autonomous harvesting system. Since fruit commonly grows in complex environments, it is challenging for the vision system to accurately identify the fruit. Humans, on the other hand, can easily recognize fruits by using their hands only. This paper proposes a method for fruit recognition based on active grasping with two-fingered grippers equipped with a soft skin sensor. The soft skin sensor has a soft silicone that is attached to the fingertip gripper. This sensor provides a 3-axis in normal and shear force based on a hall effect sensor. Active grasping by squeezing was implemented to sense and record time-series data of fruit properties. The evaluation was performed using several machine learning classification methods, such as Random Forest, Support Vector Machines (SVM), and K-Neighbors. The gripper can recognize fruit types with 88% accuracy using the Random Forest classifier. These results confirm the gripper's effective fruit recognition ability using active grasping to mimic human hand capabilities.