The robot butler is not yet for tomorrow. Today's robots are programmed to perform certain actions in advance, which means they cannot cope with changing environments. For example, the robots used in the production of cars.  They can only perform pre-programmed actions in an always identical situation. These robots are not able to adapt their actions if the situation should suddenly change.

Fortunately, there are learning algorithms that take this variability into account. However, these methods require a lot of data and robots are too expensive and too slow to generate the large amount of data.

So is there no way to speed up the learning process of robots?

"In my research, I studied how robots can autonomously learn to fold clothes more quickly. For this purpose, I developed a smart digital towel that shows the robot how it is folded. This way, the robot does not have to figure out how the towel is folded and speeds up the learning process considerably," Andreas explains.

"In addition, I involved the general public in my research. During an exhibition in the library De Krook in Ghent, people could participate in this robot research. They folded a thousand garments that helped to create a system that helps robots understand the purpose of a task."

"Using these examples, I developed a system that allows robots to accurately assess task quality. It provides future robots with the task knowledge they need to interact with their environment. They learn to understand the consequences of their actions and eventually perform complex tasks, such as folding laundry," Andreas concludes.

Read a more detailed summary or the entire PhD

-

PhD Title: Learning Robotic Cloth Manipulation

-

Andreas Verleysen obtained a Master of Science degree in Business Engineering from Ghent University in 2014. After his studies, he worked as a software developer around route optimization. He joined the AIRO IDLab in 2017 and started his PhD a year later, with financial support from the Scientific Research Fund.

During his PhD, Andreas also supervised 25 thesis students and provided support for three course units. His work has been published in four academic journals. Today, Andreas is working to apply the robotics knowledge from the AIRO research group to industrial applications. Reach out at andreas.verleysen@ugent.be to collaborate on smart robotic applications!

Contact: Andreas Verleysen, Francis wyffels, Joni Dambre

-

Editor: Jeroen Ongenae - Illustrator: Roger Van Hecke