Learning of temporal sequences for motor control of a robot system in complex manipulation tasks
Karlheinz Hohm, Y. Liu, Hans Boetselaars · 1998
Learning of temporal sequences is a topic of research in such different areas as speech and other temporal pattern recognition as well as motor control. On the other hand there is very much research effort in the area of rule based control, mainly fuzzy control, to achieve flexible systems with high autonomy. This paper concentrates on the question how a transfer of information from rule based behavior towards fast stepwise feed forward controlled motion trajectories can be achieved, so that an artificial system is able to learn from itself by generating motion sequences out of rule based behavior on its own. This allows to learn suitable trajectories, which lead to solutions of problems, where they were originally unknown, but where rules exist, e.g. stored in fuzzy controllers, which specify a desired behavior of the system in that context.