Training on the Job - Collecting Experience with Hierarchical Hybrid Automata

We propose a novel approach to experience collection for autonomous service robots performing complex activities. This approach enables robots to collect data for many learning problems at a time, abstract it and transform it into information specific to the learning tasks and thereby speeding up the learning process. The approach is based on the concept of hierarchical hybrid automata, which are used as transparent and expressive representational mechanisms that allow for the specification of these experience related capabilities independent of the program itself. The suitability of the approach is demonstrated through experiments in which a robot doing household chore performs experience-based learning.