Towards a General Measure of Skill forLearning

Within this paper, we present an approach to evaluate the skillfulness of robot behaviour and the progress made by robots when learning new and adapting existing skills. The principle measure that we use is the amount of instruction (e.g., by means of examples) that the robot needs to execute some task with a certain level of performance. Since we require only a simple indication of performance and a mean to check syntactic equivalence, this measure is independent on the level of abstraction, the kind of tasks, and the learning technique in use.