Robot localization in pipe networks

Pipe networks transporting clean water and wastewater are critical around the world. This infrastructure needs regular inspection and maintenance to reduce the effect of faults on public health, the environment, and the economic cost of operation.
These networks are made of pipes mostly less than 300 mm in diameter, buried beneath roads in lengths of around 50 to 200 metres.
An autonomous robotic system could pervasively and persistently monitor the infrastructure from within the network.

Robot localization, the ability of a robot to estimate its position in the environment, is required as it facilitates autonomous control and allows the localization of faults.
The buried pipe environment constrains the robot's sensing, locomotion, and computation, so localization is difficult. Pose estimation using GPS and a magnetometer is unavailable, typical vision and rangefinding sensing is less effective in this environment than usual, and the robot's motion is more uncertain.

This thesis develops both the front-end perception and back-end state estimation. In contrast to approaches found in the literature, it is shown that a hybrid continuous-discrete approach to state estimation is well suited for localization in this application. For lower computational cost, this approach shows an average error rate of 0 for values of uncertainty larger than found in the literature, compared to an average error rate of around 0.25 for a typical approach.
It is shown that acoustic echo sensing gives effective perception in this environment, adding to the literature a new means of observing features distant to the robot. Incorporating echoes with a novel localization algorithm gives an average error rate of 0 for larger values of uncertainty than found in the literature.
These perception and estimation aspects are shown to be easily integrated, but also function well independently.

Ongoing research develops other aspects of a robot system for this application.
The results presented here form part of this progress, informing the design of the overall system.
More generally, these results provide some evidence that careful design of the localization system, from front-end to back-end, can provide better performance compared to a typical approach. In particular, the hierarchical approach used here which considers different levels of abstraction, scale, and precision, could be applied more broadly.