Manifold Mapping based on Panoramic Localization

Robots are designed to fulfill tasks that are location-based. The robots can only execute their tasks if they know their whereabouts: they need a map and their location on that map. This spatial context enables them to show more elaborate behavior than just reactive behavior. Such map should be acquired by the robot itself, to be shure that all details (including furniture and plants) are present. This requires the combination of the localization and mapping algorithm. In the Virtual Rescue League a simultaneous localization and mapping algorithm was designed based on manifolds.

In this algorithm measurements are combined on a local map until the confidence in the match drops. At that moment a new local map is created, on a short distance from the previous map. In this way a chain of local maps is created. In the Virtual Rescue League one could use a laser scanner for an accurate measurement of both the direction and distance to obstacles. On an Aibo robot such an expensive and heavy sensor is not present, but it is possible to get an accurate measurement of the direction of obstacles with the camera in the head. This research has to find out if the ManifoldSLAM algorithm can be combined with the PanoramicLocalization algorithm, and to evaluate the quality of the maps resulting of this combination.

The progress of the research can be demonstrated at the WorldCup 2007 in Atlanta.

Keywords:
robotics, machine-learning, simultaneous localization and mapping
Study:
Artificial Intelligence
Contact:
Arnoud Visser
Location:
Universiteit van Amsterdam
References:
Virtual Rescue League
4-Legged League