Abstract
Judgments of heading direction from optic flow often show a bias toward the center of the display, but this has not been systematically investigated. We measured center bias in conditions with varied optic flow information, and tested the coordinate frame of the bias. Observers viewed 3.5s displays of simulated self-motion presented on a head-mounted-display, and indicated their perceived heading direction with a cursor. As a measure of center bias, we estimated the slope of the linear relationship between simulated and judged heading direction. Four simulated environments were tested: a textured ground plane with scattered posts, a textured ground plane alone, a cloud of fixed dots, and a cloud of randomly drifting dots. We also varied field of view (100°x64° vs. 66°x40°) and simulated observer speed (1 m/s vs. 4 m/s). We found that center bias varied systematically with the quality of optic flow information. Biases were larger for a ground plane alone than with scattered posts, and for a noisy dot cloud than a rigid dot cloud. Biases decreased with larger field of view and faster speed. To rule out response bias, we varied the distribution of simulated heading directions. Similar results were observed when heading directions were sampled from an M-shaped distribution, suggesting that biases are perceptual. To test the coordinate frame of the bias, we varied the orientation of the head relative to the body (±10°), and found that biases followed the direction of the head and display. Our results are consistent with the influence of a Bayesian prior that prefers heading directions aligned with the head or gaze. The effect of such a prior would depend on the reliability of optic flow information, as observed here. This model could be implemented with center-weighted templates, which have been proposed to account for other findings.
Meeting abstract presented at VSS 2016