Training for system codesign in robotics

A dichotomy exits within current research into robotics and intelligent systems. Conventional research centres on a traditional instructionalist model. In this paradigm the designer provides the robot system with relevant internal models of the real world and a set of instructions for carrying out clearly specifiable tasks. This approach often requires enormous computing power and can succeed in only the most constrained of circumstances. The limitations derive from the fact that the real world is not a neatly labelled place but presents an infinite variety of unquantifiably noisy input data and unforeseen circumstances. In contrast a new trend in robotics research supports the use of architectures based not on functional but on behavioural decomposition. These, so called, subsumption-based robots are equipped with a repertoire of behaviours and sets of instinct rules. Task competence is an emergent phenomenon achieved through experiential interaction with the real world. The paper details the results of an attempt to integrate these new philosophies into teaching at postgraduate level