Deep Reinforcement Learning (DRL) has shown a lot of success over the last years in areas ranging from playing computer games to control of robots. However, when facing real world tasks, Deep Reinforcement Learning approaches have difficulties with already slight variations of the task or dealing with noise. Therefore, there is a growing interest to make this systems more robust and a lot of inspiration can be taken from animals and their adaptive control strategies.
This project aims at realizing adaptive control of walking for a simple four-legged walking robot. It is based on a DRL standard environment (OpenAI’s Ant walker) which will be transferred from simulation to a real robot. The main tasks for the group will be:
- construction of the robot (3D printed, following https://github.com/OteRobotics/realant; plus using standard servos and a ROS platform)
- application of deep reinforcement learning on the robot (probably use a pretrained model from simulation)
and at a possible later stage apply/evaluate our Decentralized DRL approach
- realization of a global camera system that is able to track the robot (in order to provide position/reward information for learning)
Interest in robotics - and it would be good to have some students with experience in ROS.
Introduction to Neural Networks and python programming - basic knowledge in DRL is a plus
In case this would not find enough interest for a team project, this project proposal would be also offered (in reduced/modified form)
Rhythmus | Tag | Uhrzeit | Format / Ort | Zeitraum |
---|
Modul | Veranstaltung | Leistungen | |
---|---|---|---|
39-M-Inf-GP Grundlagenprojekt Intelligente Systeme | Gruppenprojekt | unbenotete Prüfungsleistung
|
Studieninformation |
Die verbindlichen Modulbeschreibungen enthalten weitere Informationen, auch zu den "Leistungen" und ihren Anforderungen. Sind mehrere "Leistungsformen" möglich, entscheiden die jeweiligen Lehrenden darüber.