Probabilistic Localization Methods for a Mobile Robot Navigation

Localization is a process by which a robot finds its position with respect to its own representation of the world. Mobile robot localization is one of the most important tasks in mobile robot research. Various probability-based localization approaches are introduced and compared. All of these researches employ the Bayesian rule as a fundamental theory. First, we introduce a Markov localization, which represents a posterior density – robot’s belief– by arbitrary density over the whole pose space (X, Y, θ). As a result of probability representation, the Markov localization produced many successes. Nonetheless, the Markov localization remains computationally complex and requires large storage space. Recently, computationally more efficient approaches have been introduced using a particle filter technique – also known as Sequential Monte Carlo method. Because the particle filter localization method represents the belief by a set of samples(particles), it is computationally efficient and can be accommodated to arbitrary sensor characteristics. Most recently, a couple of adaptive sampling researches have focused on reducing computational complexity by reducing the size of the samples on-the-fly, and they have demonstrated much better results than simple particle filter approaches. keywords: Mobile robot localization, Bayesian estimation, Particle filter, Sequential Monte Carlo method, Markov localization, Mobile robot. SD675: Pattern Recognition, Winter 2002 by Dr. Paul Fieguth. ID: 01800729, e-mail: [email protected] 1 SD675: Pattern Recognition, Winter 2002 by Dr. Paul Fieguth 2