Abstract
The steering dynamics model of human walking (Fajen & Warren, 2003) was developed to account for paths of locomotion with a single goal and a single obstacle (either stationary or moving). If locomotor behavior scales with the complexity of the environment, then a linear combination of the model's (nonlinear) terms should generalize to more complicated scenes. However, Cohen, Bruggeman, & Warren (VSS 2006) reported that human routes with two moving obstacles were surprisingly inconsistent. One explanation is that inconsistent routes are due to sensitivity to initial conditions, a hallmark of complex nonlinear systems. Another is that they result from differences in the allocation of attention during locomotion. Here we manipulated the attended object during walking in a virtual environment (the VENLab). Participants avoided two moving obstacles en route to a stationary goal. In the fixation condition (N = 16), the participant started walking to the goal, after 1 m the obstacles appeared, then one object flashed to indicate the fixation/attention target: (i) near obstacle, (ii) far obstacle, or (iii) goal. In the free fixation condition (N = 16), participants received no fixation instructions and no objects flashed. Observed routes were significantly more consistent in the fixation condition (86% > 78% matching the preferred route). However, this is attributable to two fast participants in the fixation group. Model simulations that incorporated the initial conditions of each trial (speed, position, heading when obstacles appeared) exhibited a similar improvement in consistency (84% > 73%) and predicted the observed route on individual trials with equal success (65%). The results demonstrate that inconsistent routes are primarily due to sensitivity to initial conditions in human locomotion, not to differences in the allocation of attention. The steering dynamics model thus generalizes to more complicated scenes, although its predictiveness on individual trials is limited by biological noise near bifurcation points.
Supported by NIH R01 EY010923.