The sense of surface orientation — A new sensor modality for humanoid robots

Abstract

This paper introduces a novel sensor concept for surface exploration in robotic applications. We present an IMU-based haptic sensor that accurately estimates the orientation of contact points. One of the biggest advantages of our proposed sensor concept is the extraction of absolute surface orientation without being affected by inaccuracies in the forward kinematics. Conventional sensors require additional steps to compute surface normals, whereas the proposed concept directly generates normals on contact, leading to accurate surface reconstruction. The introduced sensor is based on off-the-shelf components and costs much less than conventional tactile sensors. We attached our sensor to a 4-DOF robotic arm and conducted several experiments with different objects. Experimental results show that the sense of surface orientation significantly improves surface reconstruction of unknown objects with a method based on Gaussian Implicit Surfaces (GPIS), that is also presented in detail.