I think being able to usefully navigate a noisy, cluttered (real world) environment requires a cognitive architecture with a vast number of edge cases. Finding an architecture like that from first principles takes an evolutionary process operating over a vast area (the biosphere of the earth) and time scale. We can discover useful pieces of it if we study how insects are able to operate so well with such small nervous systems and so little power. I predict we won't make much progress in robotics until we take that able seriously.