Probabilistic Topological Mapping for Mobile Robots using Urn Models

Abstract

We present an application of Bayesian modeling and inference to topological mapping in robotics. This is a potentially difficult problem due to (a) the combinatorial nature of the state space, and (b) perceptual aliasing by which two different landmarks in the environment can appear similar to the robot's sensors. Hence, this presents a challenging approximate inference problem, complicated by the fact that the form of the prior on topologies is far from obvious. We deal with the latter problem by introducing the use of urn models, which very naturally encode prior assumptions in the domain of topological mapping. Secondly, we advance simulated tempering as the basis of two rapidly mixing approximate inference algorithms, based on Markov chain Monte Carlo (MCMC) and Sequential Importance Sampling (SIS), respectively. These algorithms converge quickly even though the posterior being estimated is highly peaked and multimodal. Experiments on real robots and in simulation demonstrate the efficiency and robustness of our technique.