| Home | Members | RoboCup | Videos | Documents | Blog | Sponsor | Impressum |
Research interestsOur main research interest is the automated creation of dynamic movements of simulated or real humanoid robots with evolutionary algorithms.Our current algorithm allows a robot to autonomously train movements utilizing its complete potential given by its architecture and sensoric equipment. A previous knowledge of the robot's architecture is not needed for this algorithm to work. Sensoric equipment (e.g. accelerometers, gyroscopes) is used dynamically to support and stabilize the trained movements. Because of the algorithm's independence of the robot's architecture, it can be easily adapted to a new robot. The algorithm is currently successfully used in a simulation environment (Webots) to train efficient omni-directional movements for different robots (i.e. Nao, Qrio, Aibo and KHR-2). In contrast to an application in physics simulation software, the practical usage of such an algorithm demands a drastical reduction in the required training steps. Through major algorithmic optimizations, we were able to greatly reduce the needed individuums an generations of our evolutionary algorithm. At the moment, we are able to completely autonomously train a single movement (e.g. straight walking, bended walking, turning, shooting) within a few days. (real time) Upcoming events
Past events
 Nao-Team HTWK @ RoboCup 2010, Singapore  Nao-Team HTWK @ RoboCup 2009, Graz |
|
Faculty of Computer Science, Mathematics and Natural Sciences Leipzig University of Applied Science |