Using computer graphics techniques, we are able to place subjects in immersive virtual environments displayed only as scintillating random dot cinematograms (SRDC's) with 1-frame lifetimes (Julesz, 1971). Thus, although each eye sees only a scintillating pattern of random dots of uniform density, the subjects experience moving about within room-sized virtual environments. Without training, subjects are immediately able to perform a wide range of complex spatial behaviors, including aiming toward targets, steering along curving paths, and intercepting moving objects, even though there is no optic flow correlated with the environments or actions.
Formal experiments done so far deal with 2 forms of vehicle steering: steering a curving path and steering a straight path in the presence of lateral perturbations. Two viewing conditions have been compared for each task: dioptic stimuli with smooth optic flow produced by high contrast environmental features and SRDC's with the same environmental features raised above the ground plane. The features were approximately matched for visibility in the two conditions. For the straight paths, the rms error of steering performance is 25% greater for SRDC's, and for the curving paths, rms error is about 80% greater.
The ease and accuracy with which subjects can perform complex spatial behaviors with SRDC's signify that optic flow is not necessary for visually controlled locomotion. The suggestion is that optic flow normally acts through perceived flow in the control of spatial behavior. Furthermore, to the extent that subjects are utilizing aspects of optic flow to control behavior (e.g., splay rate), these appear to be aspects of the perceived flow instead.