Office: HB 12.090
Address: Mekelweg 4
2628 CD Delft
On-line Learning of Affordances in Robotic Tasks
Affordances such as, for example, movability, rollability, liftability, are important for a robot to obtain. An affordance is defined as a relation between an entity, a robot behavior, and the effect of the behavior on the entity. One key benefit of the concept of affordance is that it is task independent and can be reused in a range of tasks that a robot needs to learn to perform. By providing a robot with conceptual knowledge of common affordances, we show that a robot is able to learn actual affordances fast. We present an approach that is able to use these obtained affordances to speed up task learning. We demonstrate the effectiveness of this approach by integrating it into an Extended Classifier System (XCS) for learning general rules in a reinforcement learning context. Our experimental results on a NAO are promising and show significant speedups in learning how to behave in a given task setting.
Please see this video :