Prof. Dr. Wolfgang Ertel,
Scientific Projects for Master
ROBOCUP is a soccer competition for robots. We are currently
developing a ROBOCUP-team for the simulation-league and a team for the
middle-size-league. For more info see www.robocup.hs-weingarten.de
Detailled list of themes: www.robocup.hs-weingarten.de/html/Stellen/stellen.html.
some examples for themes are:
If interested, please contact me
by Email oder directly
Learning by demonstration in Robocup.
Programming the "intelligence" of a robocup player by classical means
is impossible. Reasons for this are the immense complexity of the
knowledge and even more the problem of formalizing the knowledge.
Therefore we use learning algorithms to make the agents learn and
improve their playing skills.
In this project, a robocup-agent shall be trained by
demonstration. Demonstration can either be a human guiding a player
via a remote control or the action sequence of an other existing
player (imitation). Classical machine learning techniques from AI such as
decision trees or neural networks shall be tested for this task and
implemented in the simulation league.
Reinforcement Learning in Robocup.
For the fime tuning of a robocup-agent, learning by demonstration is
insufficient. Unsupervised learning with only partial teacher
intervention have to be used.
In a game like soccer, the feedback for action sequences such as
dribbling or passing often comes delayed.
Based on the state of the art in reinforcement learning, in this
project a method for our team in the simulation league shall be
selected and implemented.
Image processing and localization in the middle-size-league
Development of software for self localization and object
recognition. Techniques used here are SLAM, particle filters, SIFT,
clustering. Problems such as variable lighting conditions or
variable ball-color shall be considered.
Prof. W. Ertel, (Room K004) 8.5.08