Contextual Policy Search for Ball-Throwing on a Real Robot

Contextual policy search allows adapting robotic movement primitives to different situations by learning a versatilely applicable behavior. This work develops an implementation of behavior learning on a real robot with the goal to throw a ball at varying targets on the floor.
Several approaches has been introduced to address the problem of contextual policy search and an applicable policy presentation. In this work I examined these approaches theoretically and in a simulated robotic control task to compare the performance and the adaptability to the real scenario. The results will be presented alongside outcomes from first studies on the real robot.
Based on these results I discuss the most promising approach and the design and initialization of the experiment to evaluate the performance on the real robot.


In der Regel sind die Vorträge Teil von Lehrveranstaltungsreihen der Universität Bremen und nicht frei zugänglich. Bei Interesse wird um Rücksprache mit dem Sekretariat unter sek-ric(at)dfki.de gebeten.

last updated 31.03.2023