Present-day systems, intelligent or otherwise, are limited by the representations of the world given to them by their designers. The primary goal of this research is to automate representationchanges for problem-solving and learning systems. The research is formally grounded in a novel, first-principles approach to automating conceptual shifts for computational efficiency, called the theory of irrelevance. This theory makes precise the intuition that inefficient representations make irrelevant distinctions. The research proposed here is an attempt to extend the theoretical foundations above to coverlarger scale problems in engineering design and robotics. We will test the theory by implementing a reformulation assistant in engineering design and spatial planning.
