Visual SLAM - An Emerging Technology

Abstract

Visual SLAM is the process of making maps and staying localized using mainly visual information. VSLAM is challenging because of full 3D motion, the volume of visual information, and the geometry perspective projection. Recent advances have shown the possibility of doing realtime VSLAM over extended regions, indoors and outdoors. This workshop will bring together researchers interested in sharing their latest research, data, and future directions. We will distribute datasets beforehand on which participants can test their systems, to facilitate comparison of techniques.

List of topics

Features for visual tracking
Stability and robustness of VSLAM in small environments
Loop closure in small and large environments
Integration with exogenous sensors (IMUs, motion constraints, etc.)
Place recognition for VSLAM
Large-scale and outdoor mapping

Motivation and objectives

There has been an explosion of research in VSLAM, driven from both the fields of robot mapping with laser sensors, and vision research for mobile robots. There has been a lot of progress, with many new and interesting techniques developed and deployed in different environments. This workshop will be useful in bringing together researchers to talk about the successes and limitations of their tehcniques, and how they will approach the significant problems still remaining. A particular motivation is to see how different systems perform on standard datasets, since it is often difficult to judge comparative performance from separate papers. We will provide several types of datasets, both indoor and outdoor, several months before the workshop. We will ask presenters to exercise their systems on the datasets, and fill out a performance worksheet, which will be analyzed and distributed at the workshop.

Conference Organizers

Motilal Agrawal, SRI International
Kurt Konolige, SRI International

Date and Venue

This workshop will be held on Friday, Nov 2 2007 at San Diego in conjunction with IROS 2007 and will consist of a series of invited talks and a panel discussion by leading researchers in this field

Agenda

List of presenters

J. Sola	SRI International, USA
J. Neira	Universidad de Zaragoza, Spain
K. Konolige	SRI International, USA
P. Elinas	University of British Columbia, Canada
Tim K. Marks	UCSD, USA
M. Agrawal	SRI International, USA
G. Grisetti	University of Freiburg, Germany

Tentative Schedule

Each paper presenter will have 20 minutes to present the paper and 10 minutes to answer any questions

Friday Nov 2, 2007

9:00 - 9:10	Welcome Remarks	M. Agrawal
9:10 - 9:40	Multi-camera VSLAM: from former information losses to self-calibration	J. Sola
9:40 - 10:10	6DOF SLAM with Stereo-in-hand	J. Neira
10:10 - 10:40	Large Scale Visual Odometry for Rough Terrain	K. Konolige
10:40-11:10	Stereo vision SLAM: Near real-time learning of 3D point landmark and 2D occupancy-grid maps using particle filters	P. Elinas
11:10 -11:40	Gamma-SLAM: Stereo Visual SLAM in Unstructured Environments Using Variance Grid Maps	Tim K. Marks
11:40 - 12:10	Efficient Estimation of Accurate Maximum Likelihood Maps in 3D	G. Grisetti
12:10 - 12:40	Panel Discussion	Moderated by M. Agrawal and K. Konolige

Paper Submission

Deadlines:

Abstract and Title - 20 Aug
Final Paper 9 Sep