Exploiting symmetry to generalize biomimetic touch

Abstract

We introduce a method for generalizing tactile features across different orientations and locations, inspired by recent studies demonstrating tactile generalization in humans. This method is applied to two 3d-printed bioinspired optical tactile sensors. Internal pins acting as taxels are arranged with rotational and translational symmetry in these sensors. By rotating or translating a small sample of tactile images, we are able to generalize tactile stimuli to new orientations or locations along the sensor respectively. Applying these generalization methods in combination with active perception leads to the natural formation of a fovea of accurate real tactile data surrounded by moderately less accurate generalized data used to focus the sensor’s tactile perception.