SLAM Tutorial 2007

Introduction

Robotic mapping addresses the problem of acquiring spatial models of physical environments through mobile robots. Maps are commonly used for robot navigation (e.g., localization) and in a wide variety application areas. The robotic mapping problem is commonly referred to as SLAM (simultaneous localization and mapping) or CML (concurrent mapping and localization).

If the robot poses were known, the local sensor inputs of the robot, i.e., local maps, could be registered into a common coordinate system to create a map. Unfortunately, any mobile robot's self localization suffers from imprecision. Therefore, the structure of the local elements, e.g., of single scans, needs to be used to create a precise global map. On the other hand, it is relatively easy to localize a robot if the environment is know, i.e., a map is provided.
The nature of SLAM is the one of a chicken-egg problem, namely the coordinates of the elements of the map depend on the robots postion, and the robots position is estimates with the usage of the map, you navigate.

Scope

This tutorial provides a state of the art introduction to the SLAM problem. The tutorial targets on the users side, i.e., people who's research field is robotics, but not SLAM. Since robotic maps are needed in so many applications, our goal is to enable non-SLAM people to use SLAM softwares and to gain hands-on experiences on the problems occurring, when trying to solve SLAM.

The tutorial consists of presentations, software demonstrations and hands-on experience on our software!

Participation

To participate at the full-day tutorial please register for the ECMR 2007 conference.

Please bring our own Linux-Laptop (!!!) to be able to run our software!

Note Windows won't be supported in this tutorial. The following packages should be installed in order to run our software:

• For GMapping gsl (incl. gslcblas) and qt (we have 3.3.8 running here)
• For Treemap you need lapack, blas, f2c, qt3 and cmake.
  NEW: Please download, install and TRY OUT the treemap code from the following tar-ball. Detailed information can be found in the readme file contained. Treemap
• For 6DSLAM you only need a running X11 environment. All libs are linked statically to the executables.

Programm and Material

1. GMapping
   In this first part of the tutorial, you will learn how to use particle filter applications for mobile robot navigation. These tools include the Carnegie Mellon Robot Navigation Toolkit (CARMEN), and a Rao-Blackwellized particle filter for learning accurate grid maps GMapping.
   After providing the general concepts of particle filters (see also introduction to particle filters). We will explain the usage on how to use the mapper system for creating maps based on real robot data. These maps will then be used for navigating a robot using CARMEN.
   Download: Slides

2. Treemap
   This second part of the tutorial focuses on how to view different variants of SLAM as a least square problem represented by a graph. This view is very generic being applicable to 2D and 3D SLAM based on laserscan-matching or (visual) feature extraction. You will first learn, how to define your SLAM problem as a so called information matrix and how the sparsity pattern (which entries are 0) of that matrix encodes its the structure as a graph. Then you will learn, how this structure can be used to tremendously speed up computation. Finally, in the practical lab session, you will use the treemap algorithm provided as an open-source library to efficiently solve an extremely large SLAM problem with this approach.
   Download: Slides

3. 6D SLAM
   In this third part of the tutorial you will get hands-on experience to 3D scan matching and the performance issues that come along with matching large point clouds. In addition, you will learn about 6D SLAM, i.e., global consistent 3D scan matching using the six degrees of freedom in the robot pose.
   Given 3D scans of the environment, for instance aquired with a tilting or rotating laser range finder, the 6D SLAM software merges the recorded scans from its local into a common coordinate system. A variant of the widely used Iterative Closest Points Algorithm (ICP) is used for that purpose. Several algorithmic stategies for speeding up the computational requirements can be explored. Based on the matched scans a graph containing the relations is built and a global relaxation scheme creates a overall consistent map. As result the software presents precise 3D maps of the environment.
   Download: Slides

Author Biographies

• Giorgio Grisetti
  Giorgio Grisetti is working as a Post-doc at the research lab for Autonomous Intelligent Systems headed by Wolfram Burgard at the University of Freiburg. His research interests lie in the areas of, SLAM, mobile robot localization, and probabilistic state estimation. He was a PhD student at University of Rome "La Sapienza" in the Intelligent Systems Lab. His advisor was Daniele Nardi and he received his PhD degree in April 2006. His PhD thesis focused on SLAM using Rao-Blackwellized particle filters. In 2001, he received his M.Sc. degree in computer engineering, at the University of Rome.
  Together with Dr. Cyrill Stachniss he developed the open source "GMapping" package for 2D mapping.
  

• Cyrill Stachniss
  Cyrill Stachniss is currently working as a Postdoc in the Lab for Autonomous Intelligent Systems at the University of Freiburg. Up to November 2006, he was a senior researcher at the Swiss Federal Institute of Technology (ETH) in Zurich. From 2002 to April 2006, he joined Wolfram Burgards lab in Freiburg. In April 2006 he received his PhD at the Department of Computer Science at the University of Freiburg. The title of the thesis is "Exploration and Mapping with Mobile Robots" . Currently, he is working on exploration techniques for mobile robots in combination with SLAM. He is furthermore interested in computer controlled cars and scene analysis.
  

• Udo Frese
  Udo Frese is a senior research at the German Center for Artificial Intelligence (DFKI) in Bremen, He is also member of the Sonderforschungsbereich SFB/TR 8 Spatial Cognition located at University of Bremen and Freiburg. Past affiliations include the German Aerospace Center in Oberpfaffenhofen. He received his PhD degree in May 2004 at University of Erlangen-Nürnberg with his thesis titled "An O(log n) Algorithm for Simultaneous Localization and Mapping (SLAM) of Mobile Robots in Indoor Environments". Currently he is working in visual SLAM and in real-time computer vision for sport robotics.
  

• Andreas Nüchter
  Andreas Nüchter is a research associate at University of Osnabrück. Past affiliations were with the Fraunhofer Institute for Autonomous Intelligent Systems (AIS, Sankt Augustin) and University of Bonn, from which he received the diploma degree in computer science from University of Bonn in 2002 and was awarded with the best paper award by the German society of informatics (GI) for his thesis. He holds a doctorate degree (Dr. rer. nat) from University of Bonn. His research interests include reliable robot control, 3D environment mapping, 3D vision, and laser scanning technologies. He is a member of the GI and the IEEE.
  Andreas has devoloped fast 3D scan matching algorithms, enabling robots to map their environment in 3D using 6 degrees of freedom for pose estimates. He showed the capabilities of his robotic SLAM approach at RoboCup Rescue competitions, ELROB and several other events.

Schedule

Building 101 (same as the conference)
2. floor
room 016 / 018

Further Information about SLAM

General Contact