Least Squares Optimization on Factor Graphs
an overview
Abstract
Factor Graphs are graphical models that can represent a wide range of problems in Spatial AI, including - but not limited to SLAM, Visual and Lidar Odometry, Calibration, Structure from Motion, Model Predictive Control, and Image Classification. Effective methods to solve factor graphs are nowadays available.
This course aims to provide the students with a basic background on formalism and the techniques that can be used to solve these models.
We plan to present several practicals with small projects covering some of the topics included in this course.
The course is done within 20 hours of teaching and practical tests, is specifically thought for PhD programs but can be followed by anyone enrolled.
Teachers
Barbara Bazzana
Giorgio Grisetti
Luca Di Giammarino
Syllabus (tentative)
The syllabus might be adjusted based on the student's feedback.
Basics (27/05)
Belief
Gaussian Distribution
Least Squares
Manifolds
Robust Cost Functions
Multivariate Problems (28/05)
Factor Graphs
Pose-Graphs
Bundle Adjustment
Applications with Real Data (29/05)
Use multiple sensors within Factor Graphs
SLAM and Bundle Adjustment Applications
Constrained Optimization (30/05)
Alternate Lagrangian
Synchronization
MPC
GBP on Factor Graphs (31/05)
Belief Propagation and Message Passing on Factor Graphs
Spatial AI Applications
Logistics
Where:
in presence: DIAG Sapienza, Via Ariosto 25, 00185, Rome, Italy
remotely: Via Zoom Link, shared with the people upon enrollment
When:
May 27-31, 2024 @ 9:00-13:00
Evaluation:
small quiz in class submitted through Google Forms on the last day
complete two small homeworks started in class, and finish by Sunday June 2nd @ 23:59:59
start from Multi-view Registration 3D downgrade to 2D (x, y, theta)
start from Multi-view Registration 3D upgrade to Bundle Adjustment (exercise slides)
Send completed exercises by mail in a zip folder to digiammarino[at]diag.uniroma1.it, tag [FG24_EX] and put the correct dependencies in the folder (like tools, etc).
Enrollment Form
Please enroll using the following form. Info about the classroom and links to follow remotely will be sent to your email before the course.