Obstacle avoidance during walking in real and virtual environments

Immersive virtual environments are a promising research tool for the study of perception and action, on the assumption that visual--motor behavior in virtual and real environments is essentially similar. We investigated this issue for locomotor behavior and tested the generality of Fajen and Warren's [2003] steering dynamics model. Participants walked to a stationary goal while avoiding a stationary obstacle in matched physical and virtual environments. There were small, but reliable, differences in locomotor paths, with a larger maximum deviation (Δ = 0.16 m), larger obstacle clearance (Δ = 0.16 m), and slower walking speed (Δ = 0.13 m/s) in the virtual environment. Separate model fits closely captured the mean virtual and physical paths (R2 > 0.98). Simulations implied that the path differences are not because of walking speed or a 50% distance compression in virtual environments, but might be a result of greater uncertainty about the egocentric location of virtual obstacles. On the other hand, paths had similar shapes in the two environments with no difference in median curvature and could be modeled with a single set of parameter values (R2 > 0.95). Fajen and Warren's original parameters successfully generalized to new virtual and physical object configurations (R2 > 0.95). These results justify the use of virtual environments to study locomotor behavior.