A Simple Robotic Fingertip Sensor Using Imaging and Shallow Neural Networks

Measurements of geometrical information (e.g., contact position and normal as well as the curvature of an object) and physical information (e.g., contact force) at fingertips are essential for a multifingered robot hand to perform dexterous grasping and manipulation tasks. This paper presents a low-cost fingertip sensor that can simultaneously measure the contact position and normal, the 3-D contact force, and the local radius of curvature (ROC) of the object in contact with the fingertip. The fingertip is a hemisphere made of silicone, and there are marker points on its internal surface. A camera is installed beneath the fingertip to capture the markers' displacements, from which the aforementioned geometric and physical quantities are learned using artificial neural networks. Experiments on a handmade prototype show that, with such a simple setting, errors in the measured contact position and contact force can be as low as 1 mm and 0.3 N, respectively, and the ROC measurement can be very accurate.