At the 6th ISRR, Brian Carlisle, representing the American industrial robot manufacturer ADEPT, tried to formulate what he thought to be the most important tasks to be solved by robot researchers. “To teach a robot a task within typically 10 minutes so that it is able to repeat it in an at least slightly changing environment” was one of his major issues. Several years ago AI techniques seemed to be able to solve the problems of robot programming but moderate success led to a certain disappointment; apparently sensory feedback control and the adaptivity it may generate had been completely underestimated. And describing tasks by linguistic means soon had to realize its limits. In the last ten years learning by neural nets (partly based on Rumelhart’s backpropagation technique, partly based on self-organizing nets like KOHONENs feature maps) celebrated quite a revival not only in robotics; and thus it does not astonish that “robot learning by showing” (be it with or without nets) has gained increasing interest in the last years.