Pipeline Inspection with Autonomous Swarm Robotics

Underground water pipelines require frequent inspection to prevent decay and avoid costly repairs. The use of robots to inspect pipelines is well documented but the uniquely hostile environment of subterranean water networks means pipes require continual inspection. The idea of autonomous inspection robots that can provide continuous coverage has shown promise, but existing methods do not proactively aim to overcome a variety of diverse challenges. Specifically, extreme variability in the pipe conditions and the dense surrounding earth limit the communication capabilities of the robots, while dynamic water flows and power issues have a detrimental effect on their movement. This thesis presents the implementation of a range of path planning algorithms of varying levels of autonomy as governing swarm behaviours, each with a focus on overcoming some of the specific challenges inherent in underground water networks, with the goal of improving the efficiency with which the swarm can inspect a network. The Greedy Walk uses stochastic processes to plan a locally optimal path, the novel Ad Hoc algorithm aims to provide cyclic coverage, with robots moving as a fluid net throughout the network, and the k-Chinese Postman Problem solution explicitly plans optimal paths round subsections of a network. The thesis examines the performance of these behaviours against existing methods and anticipates obstacles in their real world implementation. The thesis then presents tailored versions of the path planning behaviours that include the introduction of more advanced methods focused on circumventing these issues. Specifically, the algorithms are developed to incorporate Gaussian Process Regression models to analyse strong water flows and use the data to plan intelligently, mitigating the detrimental effects of the flow. The flow analysis also provides a platform from which a novel Simultaneous Localisation and Mapping algorithm is presented, alongside a Multi-Objective Genetic Algorithm with the focus of increasing inspection frequency and conserving robot charge. The thesis shows evidence that an approach to pipeline inspection with autonomous swarm robotics based in path planning algorithms can help overcome the likely physical limitations of real world implementation.