Visual Control of Target-directed Locomotion in a Virtual Environment
Hongjin Sun and Barrie J. Frost Dept. Psychology, Queen's University, Kingston, Ontario (Canada) K7L 3N6 Poster presented at the IRIS/PRECARN Workshop (1998)

Introduction When an observer moves through the environment, various visual information can be utilized to guide his/her movement. As the observer moves towards a target object, various depth information (stereopsis, accommodation, vergence, parallax, etc.) can be used to estimate his/her distance relative to the target. Alternatively, the continuous transformation of the optic array of the environment provides information about the spatial and temporal relationships between the observer and his surroundings. Visual control of braking: Distance cue Timing cue from the target image expansion Through mathematical analysis, Lee (1976) proposed that when an observer is approaching a target object, an optical variable called tau can inform the observer about the "time-to-collision" which refers to the time that will elapse before the observer collides with the object. Lee further argued that tau provides a reliable and rapidly available cue of the impending collision, thus can be used for visual motor control, such as modulating an observer's speed of locomotion to arrive at an intended target. Moreover, neither the object's distance nor observer's movement velocity information is required. Research on the control of some naturalistic visual motor behaviours has provided evidence which is consistent with the tau strategy. However, perhaps due to the difficulty in manipulating environmental information in natural settings, in very few studies of target-directed movement, has the visual information of tau been manipulated independently from other cues, such as distance information. Click here to view the full time-to-collision mathematical equation. VR Setup This study used a virtual environment, in combination with selective manipulation of visual information, to study visual control of braking in human subjects (Ss). Ss wore a Helmet Mounted Display (HMD) and rode a stationary bicycle to move through a virtual world. (A. Virtual i/0 HMD with Ascension 6 DOF head tracker; B. Customized bicycle controller showing the flywheel and electronics that connect to an SGI Indigo 2 computer for turning and speed control; C. Subject riding the bicycle.) The visual scenes viewed through the HMD were updated by an SGI Indigo2 computer through bicycle and head tracking device inputs. Ss were required to ride the bicycle along a straight path towards a virtual target barrier placed across the middle of the path. The Ss' task was to bicycle towards the barrier at a fast (yet comfortable) speed, and apply the brakes to stop in front of the barrier. Ss were asked to take minimal time to approach the target, thus the task required them to maintain peddling motion until it was absolutely necessary to brake to avoid impending collision. Manipulations of Time-to-Collision: On the practice and most of the test trials, Ss rode the bicycle towards a barrier of constant size. The size of the barrier and riding distance was varied from trial to trial. In some of the test trials, the barrier was made to either expand or contract (without changing the barrier's position) during the Ss' movement. These increases or decreases in target size altered the relative rate of retinal image expansion that occurred during the observer's movement. The magnitude of the change was designed to mimic the retinal image expansion of the same object of constant size located either in front of (for expanding trials) or behind (for contracting trials) the actual position of the target as shown below. Scatter plot of the initiation of braking points relative to target (start of maximal deceleration point in braking) for subjects riding the bike at different perceived velocities. Variation of velocity was made through software in the VR setting by adjusting perceived velocity of forward motion (much like changing the gears of the bike, but without change of pedaling force as would occur on a normal bike). The magnitude of the manipulation of the perceived velocity was made proportionally at four levels: 1, 2, 3 and 4 times the standard speed. In other words, if subject's pedalling speed is the same, subject's perceived speed of motion would be 1, 2, 3 and 4 times the standard for four different conditions. The final perceived velocity obtained was also affected by the variation of subject's real pedalling velocity at a particular instant. Data was obtained from 2 subjects. It was found that as the perceived velocity increased, subjects started to stop at a proportionally farther distance from the target. The data is in line with the notion that subject initiate braking at a constant time-to-collision with the target. The top part of both figures for expansion and contraction illustrates how the equation of calculating target size is derived. The bottom part shows that as an observer moved from position 1 to position 2, the size of the target increased (for expansion trial) or decreased (for contraction trials) to the extent that Ss's visual angle substense match that of a constant sized object located in the imaginary postion. Riding velocity as a function of distance for a typical trial. For this trial, Ss was riding towards the target of constant size (normal control trial). The movement velocity increased after the start of the movement (which is 8000 inches away from the target) and was then maintained at a stable level, and eventually decelerated dramatically before collision. For this trial, it appears that there is a bit of overshot in reaching the target. The shape of the velocity profiles are similar in all trials (including trials with targets of constant size and with expanding and contracting size), except that the onset of the deceleration varies. Frequency distribution of initiation of braking points (start of maximal deceleration point in the movement) for normal control trials (n=60), expanding trials (n=30) and contracting trials (n=30). Data was derived from tests of 3 subjects. It was found that during expanding trials Ss braked at a greater distance from the target compared to control trials, and at contracting trials Ss braked at a point closer to the target than in control trials. Paired t tests comparing expansion vs. control, and contraction vs. control were performed and both tests indicated a significant difference (p<0.05) in the predicted direction. Conclusions It was found that expanding the size of the target (decreasing the tau value) caused subjects to decelerate sooner than braking in front of a target of a constant size. Conversely, contracting the target (increasing the tau value) led subjects to decelerate later. The results suggest that subjects used the tau strategy to control locomotion in this task. Not only is the tau strategy used in human visual locomotion control, it could be readily exploited in robotic vision as well, and should be computationally very efficient. Supported by NSERC, TeleLearning and IRIS / PRECARN. visitors since September 1, 1999. Updated December 2004