Reinforcement Learning is a popular machine learning method where an artificial agent has to learn how to act optimally in an unknown environment by trial and error. In this seminar, we will focus on recent developments in RL for robotics, i.e., RL for continuous state and action spaces.The students can choose from different topics from the area of reinforcement learning (RL) for robotics, including deep reinforcement learning, model-free RL, actor-critic methods, model-based RL, meta learning, hierarchical reinforcement learning and robot applications of RL. Each topic consists of several research papers for which the students have to prepare a presentation as well as a report in form of a scientific research paper.

Qualifikationsziel:Students are able to independently understand a complex research topic, present the content in a concise and understandable way and prepare a scientific report summarizing the topic.Lernziele: Students are able to independently understand a complex research topic, present the content in a concise and understandable way and prepare a scientific report summarizing the topic. Students get a deeper understanding of state-of-the art RL algorithms and get to know current research challenges.

Empfehlungen:

Der Besuch der Vorlesung „Maschinelles Lernen 1 – Grundverfahren“ ist empfehlenswert.

ArbeitsaufwandArbeitsaufwand = 90 h = 3 ECTS

Erfolgskontrolle(n)Die Erfolgskontrolle erfolgt benotet durch Ausarbeiten einer schriftlichen Seminararbeit sowie der Präsentation derselbigen in Form einer Prüfungsleistung anderer Art nach § 4 Abs. 2 Nr. 3 SPO