Behavioural Categorisation: Behaviour makes up for bad vision

The performance of a mobile robot with a vision system is assessed in an everyday object categorisation task. The ability of the robot to arrive at a specified object, behavioural categorisation, is compared to the moment to moment results from the computations of its vision system, here called perceptual classification. It is found that the mobile robot using the vision system is significantly more accurate at behavioural categorisation than the underlying performance of the visual system's perceptual categorisation. This result is discussed as supporting the hypothesis that embodied systems using real time algorithms find that ‘fast, cheap’ visual systems are sufficient for their needs.