Active steering along corrugated surfaces

To date, research on the visual control of locomotion, or optic flow, has been treated largely as a computational problem wherein formal descriptions are sought to character-ise the change in the image pattern on the retina that results from observer translation with concomitant eye rotation (se Lappe et al 1999, for details). Various models have been proposed (see Hildreth and Royden 1998, for a review) and a number of psycho-physical experiments have been conducted to evaluate these models (see Warren 1998a, for a review). Despite these efforts, the question how the visual system could extract, computationally or neurobiologically, the relevant parameters that would be used for the guidance of locomotion remains a formidable task. An equally important, but largely neglected, aspect of the visual control of locomo-tion is the effect of occlusion. When you walk down a busy street, drive on a congested motorway during rush hour, or walk on a woodland trail, the surfaces (eg of buildings, pedestrians, trees, and cars) are at times occluded and at times visible. Because light can only travel in a straight line, not all surface elements are projected to the point of observation. Only in an open environment in which a level ground recedes to the horizon with cloudless sky (a rare occurrence indeed) are all the surface elements projected to the point of observation. Under this condition, as the observer moves, every unit of an optic array is mapped into a corresponding unit in a subsequent array. The resultant optical change can therefore be represented as a flow or a transformation , ie a smooth, continuous field with no breaks or discontinuities in the directions of neighbouring velocity vectors (figure 1). By contrast, optical change resulting from locomotion in a cluttered environment is characterised by continuous deletion of the optical structure corresponding to the background texture along the leading edge of the image of the object and continuous accretion along the trailing edge. More importantly, the effect destroys the one-to-one correspondence between environmental elements and their optical counterparts. The resultant optical disturbance, therefore, is not a transformation, but a breaking of adjacent order. This fact poses immense difficulties for any computational model that extracts certain properties of the flow field by performing computations defined over