A neural model of visually guided steering, obstacle avoidance, and route selection.

Abstract

A neural model is developed to explain how humans can approach a goal object on foot while steering around obstacles to avoid collisions in a cluttered environment. The model uses optic flow from a 3D virtual reality environment to determine the position of objects based on motion discontinuities, and computes heading direction, or the direction of self-motion, from global optic flow. The cortical representation of heading interacts with the representations of a goal and obstacles such that the goal acts as an attractor of heading, while obstacles act as repellers. In addition the model maintains fixation on the goal object by generating smooth pursuit eye movements. Eye rotations can distort the optic flow field, complicating heading perception, and the model uses extraretinal signals to correct for this distortion and accurately represent heading. The model explains how motion processing mechanisms in cortical areas MT, MST, and VIP can be used to guide steering. The model quantitatively simulates human psychophysical data about visually-guided steering, obstacle avoidance, and route selection.Key Words: Heading Perception, Steering, Optic Flow, Obstacle, Goal, Pursuit Eye Movement, Gain Fields, Peak Shift, V2, MT, MST, VIP, LIP 3 IntroductionMany important steering tasks are guided by visual information, including walking through a cluttered environment (Fajen & Warren, 2003), driving (Land & Horwood, 1995;Hildreth, Beusmans, Boer, & Royden, 2000;Wallis, Chatziastros, & Bülthoff, 2002), vehicle braking (Lee, 1976, piloting an aircraft (Gibson, Olum, & Rosenblatt, 1955;Beall & Loomis, 1997), and intercepting a moving target on foot (Fajen & Warren, 2004). Steering through a cluttered environment involves the selection of a path that avoids obstacles and simultaneously approaches the intended goal. Human steering is guided by visual information about the spatial layout of the environment and the direction of self-motion through the environment, as well as proprioceptive feedback and information from other sensory systems. In particular, the visual system provides information about the relative positions of the goal object and obstacles in the environment.Movement through the world creates a full-field pattern of motion on the retina, called optic flow, which contains information about the direction of self-motion, or heading (Gibson, 1950). In principle, optic flow can be used to compute heading (Longuet-Higgins & Prazdny, 1980). During translational movement of the eye, the optic flow field contains a singularity, called the focus of expansion, which specifies the direction of heading in the absence of an eye rotation.Figures 1 & 2 Whereas optic flow relates observer motion to the available visual stimuli, recent research clarifies how visual information governs the dynamics of human navigational behavior. Fajen and Warren (2003) studied the dynamics of human steering behavior in simple goal approach and obstacle avoidance tasks using an immersive virtual reality system. They found that human performa...

A neural model of how the brain computes heading from optic flow in realistic scenes

“…The templates used in these models have been demonstrated to be learnable in both supervised and unsupervised networks (Cameron, Grossberg, & Guenther, 1998;Hatsopoulos & Warren, 1991;Zemel & Sejnowski, 1998). Gain fields have been shown to efficiently remove the effects of eye rotations in template models, allowing them to accurately detect heading while the eye is moving (Beintema & van den Berg, 1998;Elder et al, 2008). Template models are consistent with the known neurophysiology of MT + /MSTd (Perrone & Stone, 1998).…”

“…Our results indicate that a log-polar mapping is not needed to fit the data considered herein. However using a log-polar mapping may allow the model to more completely match human steering data (Browning, Grossberg, & Mingolla, 2008a, 2008bElder et al, 2008;. Data from van den Brenner (1993, 1994) show humans can accurately determine heading from ground plane stimuli with simulated rotation rates up to 5 degrees per second if the fixation point is on the ground plane.…”

“…Flow filters represent the idealized motion pattern, or template, that occurs due to translation in a rigid environment towards the point in space represented by its target MSTd cell. Cameron et al (1998) demonstrated how this type of flow filter can be learned using a self-organizing feature map (SOFM) (Cameron et al, 1998;Elder et al, 2008). As the self-organizing system is exposed to global motion patterns, such as translational motion patterns, it will learn to distinguish between different translational headings.…”

“…ViSTARS produces global motion estimates that determine heading within 1-3 degrees and are highly robust to noise in the input stream. When combined with appropriate refinements to the Steering, Tracking and Route Selection (STARS) model (Elder et al, 2005(Elder et al, , 2008, ViSTARS provides a visual front-end for reactive steering and navigation.…”

“…We present a model of primate neurophysiology that explains a set of human psychophysics data on heading estimation and demonstrates how visual cues drive perception of self-motion. The model is designed to explain how visual processing of real-world scenes can contribute to the steering behaviors demonstrated in humans and in the Steering, Tracking and Route Selection (STARS) model of reactive navigation (Elder, Grossberg, & Mingolla, 2005). Specifically our model determines heading from visual input in order to supply this information to the STARS model, hence the combined model is called Vision STARS or ViSTARS.…”