5/24 Robotics Special Seminar: Toward Life-long Visual Localization and Mapping 3:30 pm Hill Conference Room

[M. Ani Hsieh]

All,
This is a special seminar by Prof. John Leonard from MIT. Prof. Leonard has done some of the most interest work in SLAM and 3D mapping/vision in recent years.  Some of his students include Ryan Eustice (UMich), Edwin Olson (UMich), and Gabe Sibley (GMU) who have all gone one to make significant contributions in the area of 3D robot perception.

This is definitely a talk worth showing up for.

Best,
ani

Talk Title: Toward Life-long Visual Localization and Mapping
By:
John Leonard
MIT Computer Science and Artificial Intelligence Laboratory
jleo...@mit.edu

Time: 5/24 3:30 pm
Location: Hill Conference Room

This talk will describe recent work on localization and mapping for
autonomous mobile robots.  The first part of the talk will describe a
system for temporally scalable visual SLAM using a reduced pose graph
representation.  Unlike previous visual SLAM approaches that maintain
static keyframes, our approach uses new measurements to continually
improve the map, yet achieves efficiency by avoiding adding redundant
frames and not using marginalization to reduce the graph. To evaluate
our approach, we present results using an online binocular visual SLAM
system that uses place recognition for both robustness and
multi-session operation.  We demonstrate long-term mapping in a large
multi-floor building, the MIT Stata Center, using approximately nine
hours of data collected over the course of six months.  In the second
part of the talk, we will describe Kintinuous, a spatially extended
KinectFusion algorithm that permits dense mesh-based mapping of
extended scale environments in real-time. This is achieved by allowing
the region of space being mapped to vary dynamically, extracting a
dense point cloud from the regions that leave the KinectFusion volume
due to this variation, and incrementally adding the resulting points
to a triangular mesh representation of the environment.  We also
describe the integration of dense GPU-based RGB-D visual odometry and
fused real-time surface coloring.  Extensive experimental results are
provided for a variety of environments.

Joint work with Hordur Johannsson, Tom Whelan, Michael Kaess, Maurice
Fallon and John McDonald.  For more information see:

 http://groups.csail.mit.edu/marine/pub/Johannsson13icra.pdf
 http://www.cs.nuim.ie/research/vision/data/icra2013/
 http://www.youtube.com/watch?v=MsLnCaM6qrc
 http://www.youtube.com/watch?v=MEugh12dcYA
 http://www.youtube.com/watch?v=gxcHQPyPGSA

--
M. Ani Hsieh, Ph.D.
Assistant Professor
Mechanical Engineering & Mechanics
Drexel University
Email: mhs...@drexel.edu
Web: http://drexelsaslab.appspot.com/
Phone: 215-895-5870/5803