Success in integrating aspects of AI, control theory, and fuzzy reasoning (or whatever technologies you want to integrate) demands choosing a viable physical configuration for the integration platform. This configuration should be chosen as much for the software environment it is to support as for the physical environment through which it is to negotiate. Exploring physical agent learning on a robot without redundant sensing sources makes no more sense than using wheel to navigate across sand or training a bear to dance. The physical configuration will keep getting in your way. You might have some limited sucess with all of these, but you will be expending effort in the wrong places, and in the end all you will have created is a curiosity that hasn’t taught you much.