Combining Kalman Filtering and Markov Localization in Network-Like Environments

This paper presents a hybrid localization method designed for environments having the structure of a network (road networks, sewerage networks, underground mines, etc.. .). The method, which views localization as a problem of state estimation in a switching environment, combines the exibility and robustness of Markov localization with the accuracy and eeciency of Kalman ltering. This is achieved by letting Markov localization handle the topological aspects of the problem, and Kalman ltering the metric aspects. The two techniques are closely coupled: the Markov model determines the Kalman lters to be initiated, and statistics computed by the Kalman lters are used to deene the transition and observation probabilities in the Markov model. This approach has been applied to the problem of localizing a motor vehicle traveling on an urban road network, providing robust and accurate localization at low cost.