Impact of task constraints on a 3D visuomotor tracking task in virtual reality

The study investigates how task constraints in a 3D virtual reality (VR) environment affect visuomotor tracking performance and arm kinematics, with a focus on the depth (z) dimension. Motivated by the need to better characterize motor actions in virtual environments and to control informational and task constraints, the authors developed a 3D tracking task to assess adaptation and control laws underlying continuous perceptual-motor regulation. The main research question was: What is the impact of various task constraints—specifically speed of the target, size of the interaction space, and gain transforming real into virtual movement—on target tracking in 3D VR? The authors hypothesized: (H1) limited practice would enable adaptation for accurate tracking; (H2) increasing each constraint (higher gain, larger space, faster speed) would increase task difficulty and reduce accuracy; (H3) the depth dimension would be the principal source of difficulty; and (H4) depth displacements would specifically involve elbow flexion/extension coupling.

Prior work has demonstrated VR’s utility for studying perception-action coupling and controlling environmental/task constraints. Traditional visuomotor tracking has often been studied in 2D, with evidence that increasing target speed reduces tracking accuracy and shifts control from feedback to feedforward (Ao et al., 2015). Tracking tasks have been used to assess motor control deficits (e.g., cerebellar ataxia; Beppu et al., 1984, 1987) and to compare dominant vs. non-dominant hand control in 3D VR showing similar tracking skills across hands and speed-dependent kinematic changes (Choi et al., 2018, 2021; Jo et al., 2020; Park et al., 2020). Depth perception in VR depends on binocular and monocular cues (e.g., stereopsis, motion parallax, occlusion), and performance in reaching/tracking can be affected by the availability and quality of these cues (Naceri et al., 2011; Gerig et al., 2018; Vienne et al., 2020). Sensorimotor control deteriorates as trajectory dimensionality increases from 1D to 3D (Fan et al., 2019). Manipulating gain transformations in VR can degrade accuracy and comfort at high levels (Wilson et al., 2018). These findings motivate examining how gain, interaction space size, and target speed jointly shape 3D visuomotor tracking and depth-related control.

Participants: Twenty-three healthy adults (12 males, 11 females; age 19–60 years, mean 32.8 ± 11.7), novice VR users, without physical/balance disorders, provided informed consent (ethics: University of Caen Normandy). Preferred hand used (22 right, 1 left). Apparatus: HTC Vive Pro Eye HMD (dual AMOLED, 110° FOV, 2880×1600, 90 Hz) with Vive trackers on shoulder and elbow and a Vive controller in the preferred hand. Unity engine implemented a cubic 3D workspace; cube front plane 1.75 m from participant, center at 1.60 m height. World-fixed x (horizontal), y (vertical), z (depth) axes mapped to virtual space. Depth cues included stereopsis, shadows, motion parallax, relative size, and overlap. Task: Track a moving virtual target with an effector controlled by the hand-held controller. Target and effector were translucent spheres (20 cm diameter) that could interpenetrate. Each trial began by contacting a stationary target; after 0–2 s, the target moved linearly at constant speed in a random direction, reflecting off cube walls with specular symmetry. Target trajectories were randomized per trial/participant. Trial duration: 10 s. Experimental design: 120 trials total across three manipulated variables, each with four conditions, randomized; each condition repeated 10 times (3×4×10). Values chosen after pre-testing to avoid floor/ceiling effects and to counterbalance training/order. - Gain (transformation from real to virtual movement): 4.583, 6.111, 9.167, 18.333; during gain tests, target speed = 1.7 m/s and cube size = 20.8 m³ (side 2.75 m). - Size of cube (interaction volume): 1.33 m³ (side 1.1 m), 10.65 m³ (side 2.2 m), 35.94 m³ (side 3.3 m), 85.18 m³ (side 4.4 m); during size tests, speed = 1.7 m/s and gain = 7.333. - Target speed: 1.4, 1.6, 1.8, 2.0 m/s; during speed tests, cube size = 20.8 m³ (2.75 m side) and gain = 7.333. After each trial, mean target-effector distance and percent time in contact were displayed. Participants’ reach coverage within the cube was verified without trunk movement; arm movement was calibrated so the effector started at cube center. Data processing: Vive tracking provided 3D positions of shoulder (c), elbow (a), and hand/controller (b) to compute elbow angle and maximal elbow range of motion (ROM; max–min angle per trial). Variables: (1) percent of target-effector contact; (2) absolute target-effector distance (Euclidean); (3) target-effector distance per axis (x, y, z); (4) maximal elbow ROM (degrees); (5) coupling via regression R² between target position in each dimension (x, y, z) and elbow angle. Statistical analyses: Stage 1 assessed adaptation over practice using repeated-measures ANOVAs across 12 Blocks × 10 Trials without considering gain/size/speed: ANOVAs on absolute distance and elbow ROM (12×10); ANOVA on per-axis distance and R² (Dimension [3] × Block [12] × Trial [10]); descriptive analysis for percent contact. Stage 2 assessed each independent variable separately with one-way repeated-measures ANOVAs across their four conditions for absolute distance and elbow ROM; and a 3 (Dimension) × 4 (Condition) ANOVA for R². Alpha = 0.05; sphericity tested (Mauchly) with Greenhouse-Geisser or Huynh-Feldt corrections as appropriate; Bonferroni post hocs used.

Adaptation (Stage 1): - Percent contact increased from about 30% at the beginning to about 45% by the end of 120 trials, indicating rapid adaptation. - Absolute distance showed significant main effects of Block, F(11,242)=6.22, p<.001, η²=0.22, Trial, F(9,198)=7.44, p<.001, η²=0.25, and Block×Trial interaction, F(99,2178)=1.68, p<.001, η²=0.07; early trials in the first three blocks had higher distances. - Per-dimension distance showed a strong Dimension effect, F(2,44)=779.13, p<.001, η²=0.97, with consistently greater distances in the z (depth) dimension across blocks and trials (significant Dimension×Block and Dimension×Block×Trial interactions). - Maximal elbow ROM showed no significant effects of Block or Trial. - Coupling (R² between elbow angle and target position) showed main effects of Dimension, F(2,44)=706.51, p<.001, η²=0.97, and Block, F(11,242)=2.74, p<.05, η²=0.11, with interactions; R² was greatest in the z dimension, with differences between blocks/trials predominantly in z. Task constraints (Stage 2): Absolute target-effector distance: - Gain: F(3,66)=62.93, p<.001, η²=0.74; highest distance at the largest gain (18.333) compared to the other three gains. - Size: F(3,66)=51.55, p<.001, η²=0.70; higher distance in the largest cube (side 4.4 m) vs the other sizes. - Speed: F(3,66)=66.71, p<.001, η²=0.75; significant increases in distance across all speed levels (1.4 < 1.6 < 1.8 < 2.0 m/s). Maximal elbow range of motion (ROM): - Gain: F(3,66)=488.04, p<.001, η²=0.96; ROM decreased as gain increased (shorter movements required at high gain). - Size: F(3,66)=245.51, p<.001, η²=0.92; ROM increased with cube size. - Speed: F(3,66)=6.57, p<.001, η²=0.23; ROM at the two fastest speeds was significantly larger than at the slowest speed. Coupling (R² elbow-target per dimension): - Gain: Dimension F(2,44)=771.24, p<.001, η²=0.97; Gain F(3,66)=10.93, p<.001, η²=0.33; Dimension×Condition F(6,132)=3.29, p<.05, η²=0.13. R² highest in z; slight decrease at the highest gain in z. - Size: Dimension F(2,44)=595.85, p<.001, η²=0.96; Size F(3,66)=492.59, p<.001, η²=0.96; Dimension×Condition F(6,132)=66.89, p<.001, η²=0.75. R² highest in z; decreased for the smallest cube (1.1 m side) in z. - Speed: Dimension F(2,44)=875.24, p<.001, η²=0.98; Speed F(3,66)=7.33, p<.001, η²=0.25. R² in z remained highest; overall R² was significantly lower at 1.8 and 2.0 m/s vs 1.4 m/s. Overall, stronger constraints (higher gain, larger space, faster speed) increased tracking difficulty (greater distances), while elbow ROM scaled down with higher gain and up with larger space and higher speeds; coupling of elbow motion with target displacement was strongest in the depth dimension.

Findings confirm that participants adapt rapidly to the 3D tracking task with improved contact time and reduced target-effector distance early in practice (supporting H1). Increasing task constraints (greater gain, larger interaction space, and faster target) consistently reduced tracking accuracy, as evidenced by increased absolute target-effector distances (supporting H2). The depth dimension emerged as the main source of difficulty, with larger distances and stronger coupling between elbow angle and target displacement along z, indicating that depth imposed unique perceptual-motor demands (supporting H3). In accordance with H4, elbow flexion/extension amplitude and coupling were particularly implicated in tracking along the depth axis; elbow ROM increased with larger interaction spaces and higher speeds, but decreased at higher gain where smaller physical movements control larger virtual displacements. These results align with prior work on constraint-induced changes in movement kinematics and underscore the sensitivity of 3D tracking performance to VR-specific parameters. The implications are twofold: for motor control research, the task reveals dimension-specific regulation strategies; for applied VR, careful tuning of gain, workspace size, and target speed can calibrate task difficulty and desired joint mobilization, especially to solicit movement in depth.

The study demonstrates that manipulating VR task constraints (movement gain, interaction space size, and target speed) systematically modulates tracking difficulty and arm kinematics in a 3D visuomotor task. Participants adapted quickly, but performance consistently deteriorated as constraints intensified. The depth dimension was pivotal, simultaneously increasing task difficulty and strengthening the coupling between elbow flexion/extension and target motion. These findings validate the 3D tracking paradigm for probing perception-action coupling and suggest practical applications in rehabilitation: constraints can be tuned to progressively mobilize the upper limb, particularly to encourage depth-oriented movements and specific joint ranges. Future research should compare virtual to real 3D tracking, refine depth cue rendering to mitigate vergence-accommodation conflicts, explore multi-joint coordination beyond the elbow and shoulder, and test clinical populations to evaluate transfer to daily activities.

- The study used a VR HMD with inherent vergence-accommodation conflict, which may degrade depth perception and contribute to increased errors along the z-axis. - The quality and weighting of depth cues were not systematically manipulated or benchmarked against real-world viewing, limiting direct comparability to real 3D tasks. - No direct real-environment 3D tracking control condition was included, preventing assessment of VR-specific performance decrements. - Participants were healthy adults and novice VR users; generalizability to clinical populations or experienced VR users is unknown. - The analysis emphasized elbow kinematics; broader multi-joint and trunk involvement (albeit minimized by seating) were not comprehensively assessed.