Robot Mapping With Real-Time Incremental Localization Using Expectation Maximization
Abstract
This research effort explores and develops a real-time sonar-based robot mapping and localization algorithm that provides pose correction within the context of a singe room, to be combined with pre-existing global localization techniques, and thus produce a single, well-formed map of an unknown environment. Our algorithm implements an expectation maximization algorithm that is based on the notion of the alpha-beta functions of a Hidden Markov Model. It performs a forward alpha calculation as an integral component of the occupancy grid mapping procedure using local maps in place of a single global map, and a backward beta calculation that considers the prior local map, a limited step that enables real-time processing. Real-time localization is an extremely difficult task that continues to be the focus of much research in the field, and most advances in localization have been achieved in an off-line context. The results of our research into and implementation of realtime localization showed limited success, generating improved maps in a number of cases, but not all-a trade-off between real-time and off-line processing. However, we believe there is ample room for extension to our approach that promises a more consistently successful real-time localization algorithm.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 2005
- Accession Number
- ADA431560
Entities
People
- Kevin L. Owens
Organizations
- Air Force Institute of Technology