Abstract:
Pipelines are critical infrastructure for transporting gas, stormwater, and electricity, yet many are aging, poorly documented, and difficult to inspect—particularly those with small diameters or GPS-denied underground environments. This thesis investigates how accurate localization and mapping can be achieved inside small, constrained pipelines through multi-sensor fusion, advancing both algorithmic methods and practical deployment strategies. It introduces two complementary systems: VILL-SLAM, which combines a monocular camera, inertial measurement unit (IMU), ring-shaped laser profilometer, and LiDAR to enable real-time localization and dense RGB-D mapping with sub-millimeter accuracy and less than 1% drift in 12-inch and larger pipes; and PiProbe, a method designed for small and visually degraded environments that performs localization through encoder and IMU fusion with Extended Kalman Filtering, and refines global consistency using pose graph optimization with sparse above-ground GPS anchors, achieving under 2% localization error in pipes as small as two inches. Recognizing the importance of real-world readiness, the thesis also proposes a systems engineering methodology, Design for Deployment (D4D), which emphasizes formal operational modeling, early user engagement, and iterative development through field trials. Together, these contributions offer a comprehensive, field-validated framework for in-pipe robotic inspection, advancing confined-space localization and mapping and providing a blueprint for translating robotics research into deployable infrastructure solutions.

Committee:
Prof. Howie Choset (advisor)
Prof. George Kantor
Chao Cao