Topological robot localization in a large-scale water pipe network

Topological localization is well suited to robots operating in water pipe networks because the environment is well defined as a set of discrete connected places like junctions, customer connections, and access points. Topological methods are more computationally efficient than metric methods, which is important for robots operating in pipes as they will be small with limited computational power. A Hidden Markov Model (HMM) based localization method is presented here, with novel incorporation of measured distance travelled. Improvements to the method are presented which use a reduced definition of the robot state to improve computational efficiency and an alternative motion model where the probability of transitioning to each other state is uniform. Simulation in a large realistic map shows that the use of measured distance travelled improves the localization accuracy by around 70%, that the reduction of the state definition gives an reduction in computational requirement by 75% with only a small loss to accuracy dependant on the robot parameters, and that the alternative motion model gives a further improvement to accuracy.