Gaze and the Control of Foot Placement When Walking in Natural Terrain

Abstract

Human locomotion through natural environments requires precise coordination between the biomechanics of the bipedal gait cycle and the eye movements that gather the information needed to guide foot placement. However, little is known about how the visual and locomotor systems work together to support movement through the world. We developed a system to simultaneously record gaze and full-body kinematics during locomotion over different outdoor terrains. We found that not only do walkers tune their gaze behavior to the specific information needed to traverse paths of varying complexity but that they do so while maintaining a constant temporal look-ahead window across all terrains. This strategy allows walkers to use gaze to tailor their energetically optimal preferred gait cycle to the upcoming path in order to balance between the drive to move efficiently and the need to place the feet in stable locations. Eye movements and locomotion are intimately linked in a way that reflects the integration of energetic costs, environmental uncertainty, and momentary informational demands of the locomotor task. Thus, the relationship between gaze and gait reveals the structure of the sensorimotor decisions that support successful performance in the face of the varying demands of the natural world. VIDEO ABSTRACT.

Keywords: eye movements; foot placement; gaze; locomotion; real-world; rough terrain; walking.

Figure 1. Terrain Types and Data Reconstruction

(A) Subjects walked over three different types of natural terrain: Flat, Medium, and Rough. (B–E) A sample frame from Video S1, which depicts the full-body kinematic and 3D gaze data used in this study. (B) The reconstructed skeleton and 3D gaze vector (pink line) of a subject walking in the Rough terrain condition. The red and cyan dots show upcoming footholds. Intersections between the subject’s 3D gaze and the groundplane are shown as small black dots, and the Gaussian heatmap represents the kernel density estimate of these gaze-groundplane intersection points—that is, regions where the walker has fixated appear as hotspots. (C) A frame from the Positive Science eye tracker, with the subject’s 2D point of regard shown as the blue crosshairs. These data were used to calculate subject’s 3D gaze (pink line in [B]; see STAR Methods). (D) A plot of subjects’ lookahead versus time. The black trace represents subjects gaze on the ground along the primary direction of locomotor versus time. The crossed circle shows the position of the subjects’ center of mass (COM), and the pink star shows their current gaze-on-ground position, so the vertical distance between the COM and the star represents the subject’s current look-ahead distance. Horizontal red and cyan bars show the location and duration of upcoming right and left footholds, respectively. In the frame depicted here, the subject is supported over on their right foot (red horizontal bar) while looking ahead near the location of their next upcoming right foothold. (E) A 2D top-down view of subject’s gaze and footholds. Crossed circle and black line show the subject’s COM position and trajectory, and the pink star shows the subject’s current gaze-on-ground position. Black dots show upcoming gaze data, and red and cyan dots show upcoming right and left footholds. This subplot is essentially a simplified, top-down view of the groundplane represented in (B). (F) Traces of the subject’s eye-in-head position in the vertical (blue line, top) and horizontal (green line, bottom) dimensions. Each dot corresponds to one frame of eye-tracking data (30 fps [frames per second]). Note that fixations on the ground during locomotion require the subject to move their eyes downward at a constant velocity, resulting in the sawtooth-like pattern of the gaze trace in the vertical axis (as opposed to the square-wave-like shape of fixations when stationary). Combining these slow, downward eye movements with the movements of the head and body during locomotion results in the stable fixations on the ground shown in (D) and Video S1. An external video of a subject walking in the Rough terrain condition is presented in Video S2.

Figure 2. Step Length and Duration in the Three Terrains

(A and B) Step length (A) and duration (B) in the different terrains. Thick lines are means across subjects; shaded regions show ± SEM; bin sizes are 0.1 leg lengths and 50 ms, respectively. (C) Density map of the relationship between step length and duration for individual steps in the Flat, Medium, and Rough conditions. White crosshairs show the peak step length (.825 leg lengths) and duration (.525 s) in the Flat terrain, which represents subjects’ preferred gait parameters. Although the distributions in the Medium and Rough are more diffuse, the peaks remain close to that shown for the Flat terrain.

Figure 3. Gaze Density in Different Foothold-Centered Reference Frames in the Three Terrains

Gaze density in different foothold-centered reference frames for the different terrain. The plots are cross-sections of the 2D distributions along the locomotor axis. The top row shows the distribution of gaze-groundplane intersections when plotted relative to the foot that was planted at the time that the intersection was recorded (Foothold N [planted foot], red triangle). Subsequent rows show gaze density when plotted relative to steps that were upcoming at the time the gaze-ground intersection was recorded (Footholds N+1 – N+4, orange, yellow, green, and cyan triangles). Vertical dotted lines show ± 0.3 leg lengths, which was the area used in the analysis shown in Figure 4B. Horizontal red line shows the half-maximum height. This figure shows mean gaze distribution for all six subjects, but all later analyses were based on individual subject’s gaze distribution (see also Figure S1A for a 2D heatmap version of this figure and Figures S2–S7 for individual subject data).

Figure 4. Change in Maximum Gaze Probability and Sum Gaze Density near the Origin

(A) The change in the maximum value of the gaze probability distributions in the different foothold-centered reference frames shown in the rows of Figure 3. Each value was normalized by the maximum value of the distribution in the Foothold N (planted Foot) reference frame. (B) The sum of the probability distribution within ± 0.3 leg lengths of the origin (dotted vertical lines in Figure 3) for each foothold-centered reference frame. Dots show subject means; error bars are ± 1 SEM. See also Figure S1B for a measure of spread in the gaze probability distributions. See also Figures S2–S7.

Figure 5. Changes in Gaze/Foothold Relationship in the Early and Latter Halves of Steps

(A and B) Changes in subjects’ gaze probability distribution during the first and second half of each step in the different terrains ([A] and [B], respectively). Steps were defined as the period between heelstrike on a given foothold and the frame prior to subsequent heelstrike. Other than the separation of data according to the gait cycle, the max-probability and sum-probability analyses are the same as those shown in Figure 4.

Figure 6. Look-Ahead Distance and LookAhead Time in the Three Terrains

(A) Distribution of subject look-ahead distances for each of the terrain conditions. Look-ahead distance was calculated as the distance between subjects’ center of mass and gaze-groundplane intersections on frames where gaze was within 0.5 leg lengths of the walking path. The data presented in this figure are similar to data shown in the top row of Figure 3. (B) Distribution of look-ahead times in each terrain condition. On every frame where gaze was within 0.5 leg lengths of the walking path, look-ahead time was defined as the time that would elapse before the subject would reach the point on their walking trajectory nearest to their point of gaze. Solid lines show the mean across subjects; shaded region is ± 1 SEM; bin sizes 0.1 leg lengths for (A) and 100 ms for (B). Note that the distributions were normalized by trial duration and do not include frames where the subject was not looking within 0.5 leg lengths of the path. As such, each distribution sums to the proportion of the time that the subject’s gaze was near their walking path (.41 ± .16, 0.69 ± .06, 0.66 ± .08 for Flat, Medium, and Rough, respectively).