Buried pipe networks present unique challenges for robotic localization and mapping due to their constrained environments and limited accessibility. Real-time localization and mapping are essential for enabling robots to autonomously navigate these networks for inspection and maintenance. This thesis introduces methods for robotic localization and topological mapping in buried pipe networks.
The first method focuses on accurate localization by emphasising junction detection and real-time mapping. Utilising the structured nature of pipe networks, the robot identifies and maps junctions, reducing computational load and reliance on continuous video processing. A localization algorithm employing convolutional filters accurately identified junctions in the network. When navigating a simulated pipe network, the robot activated its camera only at junctions, reducing cumulative localization errors significantly. By comparing newly captured images with a pre-existing database, the robot determined whether the junction was known or new, adding it to a topological map.
Building upon this, the second method enhances mapping accuracy by using panoramic images created from junction photos. These images enabled the robot to determine the number and angles of junction exits, with the angle error maintained within 10 degrees. This method generated a topological map that provided a more accurate representation of the physical layout of the pipe network, improving navigation and planning.
Developed in a simulation environment and tested in experiments with physical robots in pipes, the second method demonstrated improved mapping precision and real-time localization, offering potential applications for the mapping, inspection, and maintenance of buried infrastructure.
