This paper explores the pairing between sensors and controllers to allow autonomous navigation in unknown environments. Addressing this problem by directly using all available sensor information in a controller (e.g. a neural network) is tempting, but problematic. If the sensor provides too little useful information, the controller selection and training will be both difficult and unlikely to lead to good system behavior. If on the other hand, the sensor provides too much information, the controller will be overwhelmed and lead to unnecessarily complex and brittle control laws. Ideally, the sensors should provide as much or as little information as is...