ManifoldSLAM: a Multi-Agent Simultaneous Localization and Mapping System for the RoboCup Rescue Virtual Robots Competition

M. Pfingsthorn and B. Slamet.
M.Sc. Thesis, Universiteit van Amsterdam, December 2006.

Abstract

This thesis presents ManifoldSLAM, an award-winning multi-robot system that provides simultaneous localization, mapping and exploration functionality. It enables a team of robots to be placed in an unknown environment, to explore it autonomously, and afterwards to produce a detailed map of the explored areas.

The design of our system focuses on the sophisticated Manifold data structure that was published by Howard et al. The Manifold is a layered data structure that employs a graph organization which decomposes the global map into small-scale local metric maps. This classifies ManifoldSLAM as a hybrid SLAM approach that attempts to merge the individual strengths of metrical and topological representations.

While Howard et al. refer to the IDC scan matcher by Lu and Milios, in ManifoldSLAM we base the SLAM-related functionality on the Weighted Scan Matcher (WSM) published by Pfister et al. The superior performance of WSM in our domain is demonstrated in an extensive set of experiments specific to our setting that also included MbICP by Minguez et al. and the Normal Distribution Transform by Biber and Straßer. The high speed and accuracy of WSM in our domain enables a light-weight implementation of the parts of the loop-closing and island-merging processes that are executed online. This significantly improves our system’s online performance, which finally allowed ManifoldSLAM to demonstrate a scalability up to at least 8 robots at the RoboCup World Championships of 2006.

Using ManifoldSLAM we have successfully competed in the Virtual Robots league of RoboCup Rescue. During the 2006 RoboCup World Championships we have acquired third place. The accuracy of our maps, the good exploration exposed by our robot team, and the fully autonomous and robust behavior control were key to our achievements. Moreover, the maps that we produced preserved an amount of detail that was unmatched by other competitors in the league. Therefore, we also won the Best Mapping Award.

In additional experiments we also illustrate that ManifoldSLAM can be applied on real-world data as our system has been demonstrated to deliver equally accurate and detailed maps from raw laser range data that suffers from real-world odometric error and sensor noise. This thesis contributes a hybrid SLAM approach that advances the current state of the art. We show that by combining the Manifold concept with WSM an efficient multi-robot SLAM system can be implemented, which has been proven successful at RoboCup Rescue.

Bibtex:

@mastersthesis{Pfingsthorn2006msc,
  author  = {Bayu Slamet and Max Pfingsthorn},
  title   = {{ManifoldSLAM: a Multi-Agent Simultaneous Localization and Mapping System for the RoboCup Rescue Virtual Robots Competition}},
  school  = {{Universiteit van Amsterdam}},
  address = {{Amsterdam}}
  year    = {2006},
  month   = {December}
}
© 2015-2019 Max Pfingsthorn. Made with Hugo, Bootstrap, and FontAwesome.