Automatic Differentiation Techniques for Real Time Kalman Filtering

From AIRWiki
Jump to: navigation, search
Title: Evaluation of Automatic Differentiation Techniques for Gauss-Newton based Simultaneous Localization and Mapping


Description: In Gauss-Newton non linear optimization one of the most tedious part is computing Jacobians. At the AIRLab we have developed a framework for non linear Simultaneous Localization and Mapping suitable for different motion models and measurement equations, but any time you need to change something you need to recompute the required Jacobian. Automatic differentiation is a tool for the automatic differentiation of source code either at compiling time or at runtime; we are interested in testing these techniques in the software we have developed and compare their performance with respect to (cumbersome) optimized computation.


Expected outcome: New modules implementations based on automatic differentiation A comparison between the old stuff and new approach

Required skills or skills to be acquired:

  • C++ programming under Linux
Tutor: MatteoMatteucci (
Start: 2015/01/01
Students: 1 - 2
CFU: 10 - 20
Research Area: Robotics
Research Topic: none
Level: Ms
Type: Thesis
Status: Active