A Local Filtering Technique for Robot Skin Data

Abstract

Algorithms for tactile data processing heavily rely on the generation of tactile images representing the contact pressure distribution. This approach allows to exploit existing algorithms for image processing, but requires the integration of the tactile elements on a flat surface. Robot skin technologies introduce challenges related to the non-regular distribution of the tactile elements and the non-planar surface over which they are integrated. In this paper, we present a method to address these challenges by developing a local filtering technique directly applicable on large-area tactile sensing systems. The proposed filter can process the contact distribution without the need of intermediate steps that are required in the typical method of generating a tactile image. We particularly focus on the design of a filter to detect sharp variations in the contact distribution, i.e. edges. The approach is validated in a task of planar contour following performed using a robot equipped with two different end-effectors (planar and non-planar) sensorized with large-area tactile sensing technology. Additional experiments have been performed to evaluate strengths and limitations of the proposed approach with respect to tactile image-based data processing techniques.