Development of a Navigation System for a Snake Robot Locomoting in Pipes

Abstract

In this report outlines a partial navigation system with a time-of-flight camera, a laser range finder and a stereo camera rig. The sensors have been evaluated with regard to domain, advantages, error sources and overall performance. The time-of-flight camera and the laser range finder gave adequate results out-of-the-box. The stereo camera rig, on the other hand, did not work efficient in the pipe environment, even though the intrinsic parameters and distortion parameters of the cameras were estimated and compensated for. This was because of the limited lighting conditions and absence of structure in the pipeline environment. A topological world representation is also presented. Pipe line junctions and other possible landmarks are nodes in a graph. Each node has a list of attributes which is determined by the system depending on the condition of the pipe junction or landmark. This relies greatly on the ability of the system to recognize and extract features from the sensor data. The idea is to fit a cylinder to the 3D dataset given from the time-of-flight camera or stereo camera, estimate the properties as radius and direction, and match this to a database to find the matching node. Large deviations from the expected values can be assumed to be anomalies, and obstacles. The proposed feature extraction algorithms did not work according to expectations due to no efficient dataset segmentation was proposed. For the 2D dataset from the laser range finder the results were more promising. Lines were fitted to the dataset employing histogram segmentation, to extract the pipeline walls, and worked adequately.