Abstract
Truly autonomous robot navigation requires an accurate map of the environment, which can for example be generated from lidar measurements or images (either monocular or stereo). Whereas in office-like environments SLAM is considered to be almost solved, the ambiguities and the irregularities that might be present in harsh and unstructured domains make SLAM still an open challenge. While numerous online state estimation approaches have been presented in the past, graph-based SLAM methods are currently undergoing a renaissance due to recent insights into the structure of the SLAM problem resulting in very efficient implementations. Graph-based solutions are typically composed of a front-end and back-end part. The front-end extracts a graph representation from sensor readings, where edges are constraints representing both the estimated trajectory of the robot from pose tracking as well as loop-closures detected by the sensors. Constraints in the graph can be contradictory since observations are always affected by noise. The challenging task of the back-end is the computation of a globally consistent arrangement of the poses given all the constraints.
This tutorial will concentrate on the essential theoretical and practical aspects of graph-based SLAM in harsh environments with emphasis on data sets generated in harsh environments, such as during RobCup Rescue and Disaster City events. The tutorial will be divided into two main sections:
- Front-end solutions to SLAM
This part will discuss algorithms for generating a constraint graph from robotic sensor measurements, such as odometry, vision, and laser scans (either 2D or 3D). On the one hand side, solutions for tracking the pose, and on the other hand, solutions computing data associations for detecting loops, e.g. detecting places that have been visited before, are discussed. - Back-end solutions to SLAM
This part will mainly deal with the problem of computing a globally consistent map given the input data in terms of a graph from the front-end. More specifically, the state-of-the-art least squares formulation based on the information matrix will be discussed.
Date and Format
Full day tutorial on Tuesday, October 4, 2011 at KI 2011 in Berlin.
Registration
Please go to the webpage of the KI'11 http://ki2011.de/ and go to "Partcipation --> Registration". You will find the tutorial under the workshops. Please send also an email to: alexander DOT kleiner AT liu DOT seThe registration / participation is FREE OF CHARGE!
Organizers
 |
Dr. Giorgio Grisetti Dept. of Systems and Computer Science La Sapienza University of Rome Via Ariosto 25 I-00185 Rome, Italy grisetti AT informatik DOT uni-freiburg DOT de |
 |
Prof. Dr. Andreas Nuechter School of Engineering and Science Jacobs University Bremen Campus Ring 12 28759 Bremen, Germany a.nuechter AT jacobs-university DOT de |
 |
Dr. Alexander Kleiner Dept. of Computer and Information Science Linköping University SE-581 83 Linköping, Sweden alexander DOT kleiner AT liu DOT se |
