This paper argues that for those neuronal systems which control behaviour,
reliable responses are more appropriate than precise responses. We illustra
te this argument using a mobile robot controlled by the responses of a neur
onal model of the locust LGMD system, a visual system which responds to loo
ming objects. Our experiments show that although the responses of the model
LGMD Vary widely as the robot approaches obstacles, they still trigger avo
idance responses. (C) 2001 Elsevier Science B.V. All rights reserved.