Abstract
First demo: A circular untextured 2D ring translated horizontally on top of a straight line is perceived predominantly as rolling like a tire on a road. The perception of rolling would be supported by motion signals tangential to the contour, but the local velocities extracted from the stimulus by a bank of motion-energy units are predominantly orthogonal to the contour as expected from the aperture effect, while feature tracking extracts motion in the direction of the translation. Hence the rolling illusion goes against the sensory information. Second demo: If a slanted 3D ring is attached rigidly to a similar horizontal ring, and the horizontal ring rotated around its center, the slanted ring appears to be rolling non-rigidly around the circumference of the horizontal ring without any evidence of a force propelling it or keeping it in orbit, thus violating the laws of physics. Experiment: This illusion demonstrates the power of prior probabilities. To identify factors that enhance or attenuate the illusion, we used two slanted rings connected at an angle and rotated them at three speeds (0.6, 6.0, 60.0 deg/sec) so that either they stayed rigidly connected, or wobbled against each other, or rolled against each other. On each trial, observers said yes or no to whether rings were rotating, wobbling, or rolling (5 repetitions/trial, 10 observers). The frequency of trials seen as rolling increased with speed and decreased when gaps, paint, or corners were added to the rings. The prior for rolling could reflect the rotational symmetry of the shape, with aliasing increasing symmetry at high speeds. In addition, features seen as not rolling could attenuate the illusion at slow speeds but be ineffective at speeds too fast for feature tracking. The two factors together predict rolling frequency with R2=0.53, 0.80 & 0.74 for slow, medium, and fast speeds.