Frank Dellaert

Factor Graphs Tutorial

Fri, 10 May 2013 19:47:35 -0400

I will give a tutorial about factor graphs at the ICRA 2013 workshop on Robust and Multimodal Inference
in Factor Graphs, organized by Niko Sünderhauf, Ed Olson, and John Leonard. Michael Kaess will give a tutorial on iSAM right after me.

You can find a PDF version of my slides here. Keynote or ppt available on request. Also check out this Factor Graphs Primer in the domains of logic and CSPs, and a hands-on introduction to factor graphs in GTSAM.

ICRA 2013 Papers

Tue, 7 May 2013 01:47:31 -0400

Congrats to Alex, Hang-Pang, and Kyel for their first-author papers at the International Conference on Robotics and Automation (ICRA):

1. Robust Vision-Aided Navigation Using Sliding-Window Factor Graphs, Han-Pang Chiu, Stephen Williams, Frank Dellaert, Supun Samarasekera, and Rakesh Kumar, IEEE International Conference on Robotics and Automation (ICRA), 2013
1. DDF-SAM 2.0: Consistent Distributed Smoothing and Mapping, Alexander Cunningham, Vadim Indelman, and Frank Dellaert, IEEE International Conference on Robotics and Automation (ICRA), 2013
1. Path Planning with Uncertainty: Voronoi Uncertainty Fields, Kyel Ok, Sameer Ansari, Billy Gallagher, William Sica, Frank Dellaert, and Mike Stilman, IEEE International Conference on Robotics and Automation (ICRA), 2013

CS 3630

Tue, 18 Dec 2012 18:47:34 -0500

This spring I will be teaching CS 3630: Introduction to Robotics and perception.

Update: Pololu is sponsoring the first assignment with a free 3pi Arduino-compatible robot for the winner of a robot competition based on the assignment!!! Details to follow...

Teaching in Fall 2012

Mon, 20 Aug 2012 22:49:09 -0400

This fall I will be coordinating the graduate computer vision class, and teaching a special topics class on 3D reconstruction. Both are graduate classes, so if you are at Tech and a grad student, check them out !

Open Perception Launch

Mon, 18 Jun 2012 15:17:53 -0400

Over the weekend, in overflowing rooms at the Point Cloud Processing workshop and PCL tutorial at CVPR 2012, Willow Garage proudly announced the creation of Open Perception, Inc. (OP), an independent non-profit foundation, focused on advancing the development and adoption of open source software for 2D and 3D processing of sensory data. For more details about the announcement, please see the official press release.

Open Perception is founded by a global community of researchers and engineers for the benefit of the industrial and research 3D perception communities. The Point Cloud Library (PCL) project represents Open Perception's most important work to date consisting of a large scale, BSD-licensed open project for 3D point cloud processing.

The foundation is set up to receive donations and sponsorship from anyone, and will concentrate on paying developers in the community, giving students travel grants and stipends, organizing open source events, and supporting its projects, such as PCL. The more support OP receives, the more it can do and give back to the entire world.

Open Perception is an open organization. We are welcoming support ranging from a pat on a shoulder, to lines of code and monetary donations. Please visit the foundation's Get Involved! page and see how you can contribute today.

GTSAM 2.0

Tue, 15 May 2012 16:44:50 -0400

Today at ICRA 2012 we announced the version 2.0 of the Georgia Tech Smoothing and Mapping toolbox, GTSAM, with a new BSD License.

Notable changes from the previous release include:
• Large runtime speedup
• Simplified and cleaner API
• CMake build system
• Added Dogleg nonlinear optimization (also have Levenberg-Marquardt and Gauss-Newton)
• Added iSAM2 algorithm for incremental nonlinear estimation, including incremental Dogleg
• Suitable for SLAM, SAM, and bundle adjustment

Download GTSAM.
Check out the quickstart guide, or the more detailed documentation. Tutorials will be available soon. A detailed list of changes can be found here.

Factor Graphs Primer

Wed, 1 Feb 2012 02:22:14 -0500

I have long wanted to write a note on how factor graphs are great representations for both logic problems (satisfiability) and constraint satisfaction problems (sudoku puzzles etc.). Today I gave a guest lecture in Merrick Furst’s algorithms class -on this topic- so I thought this would be a great opportunity to pull some figures together and put it out there.

Hence, here is the link to a short Factor Graphs Primer in the domains of logic and CSPs. Note the CSP community prefers to call these “hyper-graphs”, just so you know...

Teaching Grad AI in Spring 2012

Fri, 6 Jan 2012 11:26:27 -0500

I will be co-teaching the grad AI with Thad Starner in the spring. The website is up and will be filled with more detailed information soon.

Research Overview: Factor Graphs, Bayes Trees, and Preconditioning for SLAM and SFM

Fri, 2 Dec 2011 08:29:50 -0500

Here is the PDF of the talk.

Abstract: Simultaneous Localization and Mapping (SLAM) and Structure from Motion (SFM) are important and closely related problems in robotics and vision. I review how SLAM and SFM can be posed in terms of factor graphs, and that inference in these domains can be understood as variable elimination. I then present the Bayes tree as a novel data structure for representing the inferred posteriors, and show how the Bayes tree can be updated incrementally, yielding an efficient, just-in-time algorithm (which we call iSAM 2). Finally, I talk about the challenges of using these methods in graphs with dense cliques in them, and show how identifying an efficient sub-problem (subgraph) can yield pre-conditioners for iterative methods to attack truly large-scale problems.

ICCV Paper on Subgraph Preconditioning

Wed, 9 Nov 2011 19:22:16 -0500

Today Yong-Dian Jian will present our ICCV 2011 paper on speeding up 3D reconstruction methods, which can be phrased as an optimization problem on a factor graph, by finding a tractable subgraph to precondition the original problem, as illustrated in the figure.

Generalized Subgraph Preconditioners for Large-Scale Bundle Adjustment,
Yong-Dian Jian, Doru Balcan, and Frank Dellaert, IEEE International Conference on Computer Vision (ICCV), 2011

Abstract
We present a generalized subgraph preconditioning (GSP) technique to solve large-scale bundle adjustment problems efficiently. In contrast with previous work which uses either direct or iterative methods as the linear solver, GSP combines their advantages and is significantly faster on large datasets. Similar to [11], the main idea is to identify a sub-problem (subgraph) that can be solved efficiently by sparse factorization methods and use it to build a preconditioner
for the conjugate gradient method. The difference is that GSP is more general and leads to much more effective preconditioners. We design a greedy algorithm to build subgraphs which have bounded maximum clique size in the
factorization phase, and also result in smaller condition numbers than standard preconditioning techniques. When applying the proposed method to the “bal” datasets [1], GSP displays promising performance.

ICCV 2011 is the premier international conference on computer vision, held this year in Barcelona, Spain.

iSAM 2 Journal Paper !

Sun, 2 Oct 2011 19:14:07 -0400

The journal version on our latest magic, iSAM 2, is now slated for publication. You can download the preprint here:

iSAM2: Incremental Smoothing and Mapping Using the Bayes Tree, Michael Kaess, Hordur Johannsson, Richard Roberts, Viorela Ila, John Leonard, and Frank Dellaert, International Journal of Robotics Research, 2012

Abstract
We present a novel data structure, the Bayes tree, that provides an algorithmic foundation enabling a better understanding of existing graphical model inference algorithms and their connection to sparse matrix factorization methods. Similar to a clique tree, a Bayes tree encodes a factored probability density, but unlike the clique tree it is directed and maps more naturally to the square root information matrix of the simultaneous localization and mapping (SLAM) problem. In this paper, we highlight three insights provided by our new data structure. First, the Bayes tree provides a better understanding of the matrix factorization in terms of probability densities. Second, we show how the fairly abstract updates to a matrix factorization translate to a simple editing of the Bayes tree and its conditional densities. Third, we apply the Bayes tree to obtain a completely novel algorithm for sparse nonlinear incremental optimization, named iSAM2, which achieves improvements in efficiency through incremental variable re-ordering and fluid relinearization, eliminating the need for periodic batch steps. We analyze various properties of iSAM2 in detail, and show on a range of real and simulated datasets that our algorithm compares favorably with other recent mapping algorithms in both quality and efficiency.

Also see our earlier conference papers:
• iSAM2: Incremental Smoothing and Mapping with Fluid Relinearization and Incremental Variable Reordering, Michael Kaess, Hordur Johannsson, Richard Roberts, Viorela Ila, John Leonard, and Frank Dellaert, IEEE International Conference on Robotics and Automation (ICRA), 2011
• The Bayes Tree: An Algorithmic Foundation for Probabilistic Robot Mapping, Michael Kaess, Viorela Ila, Richard Roberts, and Frank Dellaert, Intl. Workshop on the Algorithmic Foundations of Robotics, 2010

CS 4480 Digital Video Special Effects

Tue, 23 Aug 2011 00:24:34 -0400

This fall I am excited to be teaching CS 4480 DVFX again. What a cool job I have!

CS 7495 Grad Computer Vision

Thu, 11 Aug 2011 11:26:37 -0400

This fall I will be coordinating the graduate computer vision class, which will be taught as a separate course, unlike before when undergrad and grad were cross-listed. In another innovation, the graduate class will be co-taught by all faculty in the Computational Perception Lab at Georgia Tech. For more information, please consult the CS 7495 website, including the syllabus and the schedule.

Barcelona Comp. Photography Portfolio

Mon, 8 Aug 2011 14:06:12 -0400

I had a blast teaching computational photography in Barcelona this summer. To get a flavor for what we did, visit the just created portfolio page, which shows off some of the students’ work, with their own remarks on the program, the class, and their chosen work. The above image, BTW, is an average of the faces of all students in the class :-)

Dr. Kai Ni!!!

Fri, 5 Aug 2011 19:40:22 -0400

Congratulations to my (ex-) student Kai Ni who officially graduated on this date from Georgia Tech with a Ph.D. in computer science. Above a snapshot of him (and me) in full uniform. Kai is now working at Microsoft on super-secret stuff: he would not even tell me what it is :-)

ICRA 2011

Fri, 11 Feb 2011 23:48:10 -0500

The two papers below will be presented at the 2011 ICRA conference in Shanghai, China.

The first one is about learning to predict which 3D features from a large environment model will be detected in a camera image, which can cut down tremendously on computational cost in localizing the camera, e.g., in robot or phone localization:

• Visibility Learning for Large-Scale Urban Environment, Pablo Fernández Alcantarilla, Kai Ni, Luis Miguel Bergasa, and Frank Dellaert, IEEE International Conference on Robotics and Automation (ICRA), 2011

The second one is a very exciting new incremental optimization technique, iSAM 2, which we apply to mapping and localization in robotics. We represent what we know about the robot’s environment as a large tree data structure called the Bayes tree, which we iteratively update as more measurements come in via the robot’s sensors:

• iSAM2: Incremental Smoothing and Mapping with Fluid Relinearization and Incremental Variable Reordering, Michael Kaess, Hordur Johannsson, Richard Roberts, Viorela Ila, John Leonard, and Frank Dellaert, IEEE International Conference on Robotics and Automation (ICRA), 2011

You can see more pictures and a movie about iSAM 2 here.

Photometric Stereo on the iPhone

Sun, 9 Jan 2011 19:09:59 -0500

My ex-student Grant Schindler made a very cool iPhone app that uses photometric stereo to create a 2.5D rendering of your face, or anything else. Check out the Trimensional webpage or the app on the iTunes store.

IJRR Paper on Topological Mapping

Mon, 15 Nov 2010 23:07:54 -0500

Ananth Ranganathan and myself just got a paper accepted to the International Journal of Robotics Research, on on-line probabilistic topological mapping. The cool aspect of this work is that we never make hard choices about what topology is correct. Rather, we maintain a set of samples over time over the space of all possible topologies. Ananth (now at the Honda Research Institute) came up with an efficient Rao-Blackwellized particle filter approach to do this on-line.

You can find a preprint here (PDF, 2.6MB). For a full list of my publications, including other papers on topological mapping, click here.

The Bayes Tree and GTSAM

Mon, 25 Oct 2010 09:14:28 -0400

I gave talk on the Bayes tree (link is pdf, 2MB) and our new GTSAM toolbox at the GraphBot Workshop on Probabilistic Graphical Models in Robotics at at IROS 2010. Please also take a look at our upcoming WAFR paper:

The Bayes Tree: An Algorithmic Foundation for Probabilistic Robot Mapping, Michael Kaess, Viorela Ila, Richard Roberts, and Frank Dellaert, Intl. Workshop on the Algorithmic Foundations of Robotics, 2010

IROS

Thu, 15 Jul 2010 02:54:14 -0400

We just put the finishing touches to four papers that will be presented at IROS 2010 this coming October, on underwater mapping, distributed SLAM, hierarchical SLAM, and a new supercool optimization method. Read more for details.