By: Christoph Naue
Adding new gestures for human-robot interaction using EEG-based reinforcement learning
Reinforcement learning applications have evolved rapidly throughout the last decade. Recently the DFKI successfully implemented a reinforcement learning (RL) algorithm that uses error potentials evoked by error recognition in the human brain for human-robot interaction. The algorithm makes it possible to map gesture inputs to robotic actions using error potentials as feedback (reward in RL).
New situations in real world applications could make it necessary to adapt the approach. This master thesis focused on investigating approaches that make it possible to add gestures and actions, while retaining already learned knowledge. The thesis also focused on reduction of knowledge that is obtained by training an algorithm before online learning (pretraining), which can be especially relevant for initial phase of online learning process. To this end, the algorithm from the DFKI is modified and modifications have been evaluated on data that is recorded from 6 subjects. The results will be presented.
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.