Egocentric Direction and the Visual Guidance of Robot Locomotion Background, Theory and Implementation

and Overview. In this paper we describe the motivation, design and implementation of a system to visually guide a locomoting robot towards a target and around obstacles. The work was inspired by a recent suggestion that walking humans rely on perceived egocentric direction rather than optic flow to guide locomotion to a target. We briefly summarise the human experimental work and then illustrate how direction based heuristics can be used in the visual guidance of locomotion. We also identify perceptual variables that could be used in the detection of obstacles and a control law for the regulation of obstacle avoidance. We describe simulations that demonstrate the utility of the approach and the implementation of these control laws on a Nomad mobile robot. We conclude that our simple biologically inspired solution produces robust behaviour and proves a very promising approach. 1 Theoretical Background: Human Locomotion and Egocentric Direction For the past 50 years it has been assumed that humans rely on optic flow for the visual guidance of locomotion. This assumption has underpinned psychophysical studies, neurophysiology, imaging and computational modelling (see [1] for a review). Recently this assumption has been challenged. Rushton et al [2] reported an experimental result seemingly at odds with the use of optic flow. Rushton et al proposed instead a simple heuristic that better described the behaviour they observed. The proposal is that visual guidance of locomotion is achieved by keeping a target at a fixed direction, or eccentricity, relative to the body, rather than regulating behaviour so as to maintain a certain pattern of flow on the retina (the optic flow solution). In short, if the current direction of a target object is known, and the observer walks so as to keep the direction constant then they will reach the target. If the target is kept straight-ahead then a straight-line course to the target will result. If the target is maintained at some other direction then the path will be an equi-angular spiral. The finding of Rushton et al has now been replicated by many others [3-7], and a concise summary of the original study is provided below. In a later section we Egocentric Direction and the Visual Guidance of Robot Locomotion Background 577 illustrate how this simple heuristic can be extended into a general model of the visual guidance of locomotion. We then describe a control law to avoid obstacles. 1.1 The Prism Study, Rushton et al. (1998) The Rushton et al. [2] study involved observers wearing prism glasses. Observers were asked to walk briskly towards a target held out by an experimenter positioned about 10m to 15m away. The glasses contained either paired base-left or base-right wedge prisms. Prisms deflect the image and so shifted the perceived location of objects (relative to the body) approximately 15 to the right or left. Wearing prism glasses had a dramatic effect on the trajectory taken by observers when asked to walk towards the target. Observers veered whilst attempting to walk ‘straight towards’ the target. A typical veering trajectory is shown in the left panel of Figure 1.