This tutorial addresses Visual SLAM, the problem of building a sparse or dense 3D model of the scene while traveling through it, and simultaneously recovering the trajectory of the platform/camera. Visual SLAM has received much attention in the computer vision community in the last few years, as more challenging data sets become available, and visual SLAM is starting to be implemented on mobile cameras and used in AR and other applications. We will provide an introduction to the core concepts underlying current sparse, dense and semantic visual SLAM systems.


Frank Dellaert, Georgia Institute of Technology
Michael Kaess, Carnegie Mellon University

Invited Lecturers

Stephan Weiss, NASA Jet Propulsion Laboratory
Richard Newcombe, University of Washington
Chris Beall, Georgia Institute of Technology


7:30 – 8:30 Breakfast
8:30 – 10.15 AM Session 1: Visual Odometry
10:15-10:45 Coffee Break
10:45-12:30 AM Session 2: Visual SLAM
  • Scale ambiguity, inertial (Stephan Weiss) [slides]
  • Efficient inference (Michael Kaess) [slides]
  • VSLAM on phones with demo and loop closing (Frank Dellaert) [slides]
12:30-13:30 Lunch Buffet
13:30-15:25 PM Session 1: Advanced Topics
15:25-15:55 Coffee Break
15:55-17:00 PM Session 2: Dense SLAM
  • Kintinuous (Michael Kaess) [slides]
  • SLAM++ (Richard Newcombe) [slides]
  • Discussion