Adaptive intelligent behavior and control
The field of work of an automous robot includes tasks which are too complex or comparatively extensive if programmed directly. Therefore, methods oriented on natural learning behavior are more effective. Robots acting by the principle of trial and reward/punishment can optimize their behavior by iteratively repeating the learning process. With the help of simulations, even complex behaviors can be learned. For a mobile service robot, for example, it is possible to learn to avoid an obstacle or perform a pick-and-place task.
One branch of artificial intelligence we use is called reinforcement learning. Tasks of our project group include implementing the theory-loaded algorithms into practical applications in order to extend or improve them or to develop new methods. Finally, they will be joined in a universally applicable learning framework, the so-called Teachingbox.
Conceived as a stand-alone module, the Teachingbox can be used in other projects and will be accessible to developers from
any discipline.
|