The stimulus simulated walking over a ground plane with a static sky above. The stimulus showed the central view from an artificial camera \(1.8\, \text{m}\) above the plane. Sky and plane were represented by white pixel-sized dots on a black background. Those dots were placed randomly with a density of \(1.4694\, \text{dots/deg$^2$}\), so that any single frame showed no spatial feature of the plane or sky and appeared as white noise. Motion of the observer was simulated by updating the position of the dots on the plane according to the optic flow calculated for that observer motion. Dots in the sky remained static. Dots had a limited lifetime such that every dot had a probability of 0.15 per frame to be replaced with a dot at a random position. All dots that left the screen during simulated observer movement were spawned back at a random position. This ensured a steady dot density over time and a white noise distribution in every single frame. The observer traveled with \(1.5\, \text{m/s}\) on a straight line over the plane, either toward the center of the screen or 5° or 10° to the left or right, resulting in five possible heading directions. The “no-walker” condition simply presented these stimuli. The stimulus duration was \(1.7\, \text{s}\).
In the other conditions, a walker was placed on the ground plane. The walker was constructed from a motion-capture recording of a walking human actor (
de Lussanet et al., 2008). The recording consisted of the motion of 18 points representing the head, shoulders, elbows, wrists, hands, hips, knees, ankles, and tips of the feet. To create the body shape of the walker, each of those points was assigned a circular area of radius
\(\sqrt{0.02}\, \text{m}\) in which dots could be presented. Given the dot density of
\(1.4694\, \text{dots/deg$^2$}\) and a walker distance of
\(1.5\, \text{m}\), 135 dots were inside each of these circular areas on average. The collective of these areas, which changed in every frame due to the articulation of the limbs, formed the body of the walker. The walker walked with a speed of
\(1.2\, \text{m/s}\) and took
\(1.1333 \, \text{s}\) or 68 frames to complete one walking cycle. The stimulus duration was
\(1.7\, \text{s}\), resulting in 1.5 walking cycles.
We tested two different walker paths, following earlier studies on independent object motion (
Warren & Saunders, 1995;
Royden & Hildreth, 1996;
Li et al., 2018). In the “equidistant” condition, the walker started
\(1.02\, \text{m}\) left or right from the center of the display, then crossed the observer's path, and finished
\(1.02\, \text{m}\) on the other side of the center of the display. The walker's facing was
\(\pm 90 ^\circ\) and the articulation matched the lateral translation. During this movement, the walker was shifted in depth with the observer movement such that it kept a fixed distance of
\(1.5\, \text{m}\) from the observer while walking. This is similar to the paradigm of
Royden and Hildreth (1996) and to the lateral motion condition of
Li et al. (2018). Based on the results of those studies, we expect a bias in the direction of the local optic flow within the equidistant walker shape. In the “approaching” condition, the walker started in the center of the display,
\(5\, \text{m}\) in front of the observer, and walked along a
\(45 ^\circ\) tilted path to the left or right, facing in walking direction and articulating according to the translation. In this condition, the walker moved in depth toward the observer, similar to the paradigm of
Warren and Saunders (1995) and the motion-in-depth condition of
Li et al. (2018). Based on the results of those studies, we expect a bias against the direction of the local optic flow within the approaching walker shape.
We tested three different motion patterns of the dots within the walker shape. For the “congruent-with-walker” condition, the circular areas that formed the body of the walker were filled with white pixel-sized dots with the same density as the background flow. These dots were locked in place relative to the center of each respective circular area and thus moved with the walker. For the “incongruent-with-walker” condition, the body of the walker was superimposed on the scene and any dot of the plane or the sky that fell within the body area was moved with a local motion vector inverse to the lateral motion of the walker (i.e., same speed but opposite lateral component). Thus, in the equidistant condition in which the walker moved rightward but kept a fixed distance from the observer, the dots simply moved leftward. In the condition in which the walker approached the observer while walking from the center to the right, the dots moved as if the walker was still approaching but moving from the right to the center. For the “congruent-with-flow” condition, any dot of the plane or the sky that fell within the body area was treated as if it corresponded to a static object at the depth position of the walker. Thus, those dots moved as if they belonged to the background flow but with increased speed because the walker was closer to the observer than the background.
For the incongruent-with-walker and congruent-with-flow conditions, dots often crossed the walker outline as the dot motion and the walker shape motion were not aligned. When dots crossed the outline, they were captured such that the speed and direction of their motion matched the intended optic flow at their current position.
In the facing and articulation tasks of Experiments 2 and 3, observer motion was straight ahead, walkers faced left or right, and walker motion presented four different cases. Articulation was either forward or backward, the latter generated by inverting the order of the frames in the walking cycle. Moreover, translation of the walker could either match the articulation or be in opposite direction, resulting in either normal or “moonwalking” stimuli. The presentation duration of each stimulus was \(1.7\, \text{s}\), corresponding to 1.5 walking cycles and ensuring that the walker’s posture in the final frame was the same in all conditions.