Domain-specific constraints for localisation and mapping in sewer systems: algorithms and validation in laboratory and field tests

Abstract Precise localisation and mapping are essential for reliable sewer inspection. A robust positioning method is required for the deployment of autonomous robotic systems, the detection, documentation and monitoring of defects and the preparation of maintenance work. Pose estimation based on wheel encoders and inertial measurement units suffers from slip and drift over long distances. Standard lidar- and vision-based simultaneous localisation and mapping algorithms, optimised for indoor or underground environments, struggle in sewers because of sparse geometric features, non-planar walls and repetitive structures. To address these challenges, a lidar-based simultaneous localisation and mapping algorithm designed for sewer systems is proposed. The method incorporates domain-specific constraints: pipe segments are straight elements with known or bounded length, starting and ending at well-defined connections. In addition, an adapted feature detector and loop closure strategy are introduced to deal with smooth surfaces, lack of distinct edges and the presence of repetitive pipe sections. By combining knowledge of the geometric properties of the sewer infrastructure with sensor data, the proposed approach achieves stable localisation in laboratory and operational sewer holdings. In the laboratory setup, the method achieves a mean trajectory error of $$13.4 \pm 4.1\,\textrm{mm}$$ . In the field, endpoint errors below one per cent of the holding length are achieved. This level of accuracy provides a pose reference for associating inspection findings with pipe coordinates and supports autonomous sewer inspection. The algorithm closes a technological gap and contributes towards practically deployable robotic tools for large-scale condition assessment, targeted maintenance and long-term monitoring of critical underground infrastructure.