20th February – 13th March 2009 Lectures on Probabilistic Robotics

Eng. Paolo Villella held a series of lectures in Probabilistic Robotics.

All events has taken place in Gustavo Stefanini Center Conference Room, h. 10:00 – 13.00.

Lectures calendar and Abstracts.

20/02/09, Part 1: Introduction to Probabilistic Robotics
.Probabilities
.Bayes rule
.Bayes filter
A brief overview of the main problems faced in robotic navigation is provided, pointing out characteristics of the probabilistic approach. In the first part of the lesson, basics
in probability theory are given with a focus on bayesian interpretation of probability. The second part of the lesson is focused on Bayes rule and its application to some case
studies in robotics.

27/02/09, Part 2: Bayes Filter Implementations
.Gaussian filters
In the first part of the lesson, the Kalman filter is described, as an implementation of the Bayes filter using moments parameterization for linear systems. Next, the Kalman filter is extended to nonlinear problems (Extended Kalman Filter, EKF). The second part of the lesson is focused on the unscented Kalman filter and the information filter. Finally, real robotic applications of these filters are presented.

06/03/2009, Part 3: Bayes Filter Implementations
.Discrete filters
.Particle filters
This lesson discusses two nonparametric approaches for approximating posteriors over over continuous spaces with finitely many values. The first (histogram filter) assigns to each region a single cumulative probability. The second (particle filter) represents posteriors by finitely many samples. Finally, real robotic applications of these filters are presented.

13/03/09, Part 4: Probabilistic Motion Models and Probabilistic Sensor Models
I) PROBABILISTIC MOTION MODELS
II) PROBABILISTIC SENSOR MODELS
.Beam based
.Scan based
.Landmarks
The first part of the lesson focuses on mobile robot kinematics for robots operating in planar environments. Probabilistic motion models for skid-steering robots are derived, pointing out its robustness with respect to deterministic models. Finally, implementation of the resulting algorithms on real robots are discussed.
The second part of the lesson focuses on probabilistic perception, providing measurement models for range finders. In particular the use of scanning laser range finders is discussed. Finally, the use of grid based or feature based map is introduced, with some case studies.