Robust efficient localization of robots in pipe networks using a particle filter for hybrid metric-topological space

Water distribution and drainage pipe inspection and maintenance is costly, and could be improved by using robots to locate faults from within the pipes. Robot localization is critical in this operation, but is challenging due to the constraints of the pipe environment. An efficient, robust algorithm is needed for localization using limited sensors. A novel particle filter algorithm is proposed for localization, which estimates the robot’s position in a hybrid metric-topological state space, allowing efficient computation and relocalization. The algorithm is demonstrated in simulation at a large scale, considering substantial uncertainty in motion, measurements, and the map of the environment, showing an improvement over a benchmark algorithm developed for this application.