Abstract
Fajen and Warren (2003) developed a dynamical systems model of steering and obstacle avoidance based on data from human subjects, in which locomotor paths emerge on-line. By linearly combining goal and obstacle components, the model can be used to predict route selection behavior in complex scenes containing multiple obstacles. In this study, we compare the predictions of the steering dynamics (SD) model with models that minimize path length (MPL) and minimize total lateral impulse (MLI), where I = ∫ F dt. The experiment was conducted in a 12 m x 12 m virtual environment viewed through a head-mounted display (FOV 63° H x 53° V). Subjects (N = 11) walked from a home location to a goal 8 m away while avoiding an array of 12 randomly positioned obstacles (2 m posts). There were eight different obstacle arrays, and each array was presented in both the forward and backward directions six times, yielding 16 configurations and a total of 96 trials. The MLI model was the worst predictor of human routes, for the mean total lateral impulse on all observed routes exceeded that of the MLI route by 67%. The MPL model was comparatively better, for the mean length of all observed paths exceeded the minimum path length by just 8%. The SD model generated paths that were nearly identical in length to the human paths, and predicted human routes as well as the MPL model. We conclude that people select routes that nearly minimize path length but not total impulse, and that the SD model captures an on-line control strategy from which human-like, nearly minimum length paths emerge.