Robot Simultaneous Localisation and Mapping in Pipe Networks

This thesis investigates the application of visual Simultaneous Localisation and Mapping (vSLAM) for robots navigating sewer pipe networks. SLAM can estimate a robot's trajectory and build a map simultaneously. Given the complexity and vastness of sewer networks, using multi-robot SLAM is a practical solution for efficient exploration. To accurately estimate the robot’s absolute positions, this work also leverages a prior map that details the planar distribution of manholes and pipes with known absolute positions.

vSLAM, however, suffers from accuracy degradation and scale drift within pipes, leading to rapid error accumulation. Thus, the proposed vSLAM algorithm incorporates structural information from the pipe networks to mitigate this problem. The algorithm extracts cylinders and integrates this information into triangulation, local optimisation, and pose graph optimisation. When isodiametric information is available, the algorithm corrects scale drift based on detected cylinders, significantly reducing errors post-correction.

This work introduces a novel, robust, and efficient optimisation-based cylinder extraction method to detect cylinders. It employs a minimal parameter representation, avoiding constraints in optimisation. The thesis also identifies and analyses two numerical instability conditions where point-based cylinder extraction methods can fail. Comparative results with existing methods using both simulated and real data demonstrate the advantages of the proposed algorithm.

Because of unreliable communication, this research also proposed a computationally efficient offline map merging method. The algorithm identifies and matches manholes between two maps based on image features and then merges the maps based on overlapping manholes. This approach reduces computation time to just one-sixth of that of traditional methods.

Finally, the thesis presents a topological fault-tolerant matching method to estimate the absolute pose and scale of the point cloud map using a prior two-dimensional map that details the planar distribution of manholes and pipes. The algorithm accurately determines the absolute pose and scale by aligning the point cloud map with this prior map.

These contributions significantly enhance the accuracy and efficiency of vSLAM in sewer pipe networks, making autonomous exploration and mapping more feasible and reliable.