A Method for Selecting the Next Best Angle-of-Approach for Touch-Based Identification of Beam Members in Truss Structures

Abstract

A robot designed to climb truss structures such as power transmission towers is expected to have an adequate tactile sensing in the grippers to identify a structural beam member and its properties. Depending on how a gripper grasps a structural member, defined as the Angle-of-Approach (AoA), the extracted tactile data can result in erroneous identifications due to the similarities in beam cross-sectional shapes and sizes. In these cases, further grasps at favorable Angles-of-Approach (AoAs) are required to correctly identify the beam member and its properties. This paper presents an information-based method which uses tactile data to determine the next best AoA for the identification of beam members in truss structures. The method is used in conjunction with a state estimate of beam shape, dimension, and AoA calculated by a Random Forest classifier. The method is verified through simulation by using the data collected using a soft gripper retrofitted with simple tactile sensors. The results show that this method can correctly identify a structural beam member and its properties with a small number of grasps (typically fewer than 4). This method can be applied to other adaptive robotic gripper designs fitted with suitable tactile sensors, regardless of the number of sensors used and their layout.