Data-Driven Robot Control

University of Southern Denmark

Title: Data-Driven Robot Control
The Maersk Mc Kinney Moller Institute, SDU Robotics
Teaching language: English
Teachers: Christoffer Sloth chsl@mmmi.sdu.dk / Inigo Iturrate inju@mmmi.sdu.dk
ECTS: 2.5 ECTS
Period: May 2024
Offered in: Odense

Prerequisites
It is recommended that students participating in the course have:

a.                         basic knowledge in control of robots

b.                         basic knowledge in optimization

c.                          basic knowledge in machine learning

Content

Data-driven methods, such as Gaussian processes, make it possible to obtain models of unknown processes with uncertainty quantifications, and have found widespread applications in recent years. This course gives an introduction to data-driven methods for robot control.

The course will start with a general introduction on the theory of Gaussian Process Regression [1], which will serve as a backbone for the remaining topics.

The theory will subsequently be exemplified through three use-cases: identification of inverse dynamics models of robotic manipulators [2], safety guarantees for uncertain dynamical systems [4], and learning of robot trajectories based on a given set of demonstrations [4].

[1] Wang, Jie. "An intuitive tutorial to Gaussian processes regression." Computing in Science & Engineering, 2023.

[2] J. S. de la Cruz, W. Owen and D. Kulíc, "Online learning of inverse dynamics via Gaussian Process Regression," 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems, Vilamoura-Algarve, Portugal, 2012, pp. 3583-3590, doi: 10.1109/IROS.2012.6385817.

[3] Y. Kim, I. Iturrate, J. Langaa and C. Sloth, “Safe Robust Adaptive Control under Both Parametric and

Non-Parametric Uncertainty”, Advanced Robotics, 2024.

[2] M. Arduengo, A. Colomé, J. Lobo-Prat, L. Sentis and Carme Torras, “Gaussian-process-based robot learning from demonstration,” J Ambient Intell Human Comput. 2023. https://doi-org.proxy1-bib.sdu.dk/10.1007/s12652-023-04551-7

Learning outcomes

The aim of the course is, to give the student knowledge about:

• Gaussian Process Regression and its application to robotics

Be able to work with the following skills:

• Quantify the uncertainties of a dynamical system trajectories based on data

And have the competences to:

•       Design controllers for uncertain dynamical systems using data-driven methods
   

Time of classes

The course will start in May 2024 and will have five sessions.

The course will last 5 days, i.e., 40 hours.

2.5 ECTS = 67.5 h (40 h teaching, 15 h preparation, 12.5 h hand-in)
 

More information and registration:
Via email to Christoffer Sloth (chsl@mmmi.sdu.dk) or Pia Mønster (pmkr@mmmi.sdu.dk).
Deadline: 1 week before the classes start.
 

Form of instruction
The teaching is a mixture of lectures and exercises, where the students can apply the theory in practice.
 

Examination conditions
Following is a prerequisite to attend the final project exam:

• Participation in 80 % of the classes.

Evaluation

Internal examination with no co-examiner based on a submitted report with solutions to problems addressed during the classes. The assessment will be pass/fail.
 

Price
No charge

Back