Catching an object requires the hand to meet a future position, not simply follow the object’s latest visible location. My MSc thesis with Sang-Hoon Yeo uses virtual reality to examine how changing a ball’s downward acceleration affects that prediction, eye tracking and interception performance.
Why Change Gravity in VR?
Everyday experience gives us strong expectations about falling motion. A virtual environment can present a familiar-looking ball while changing the acceleration that governs its descent. This creates a controlled way to ask how visual evidence and expectations about gravity contribute to interception.
The task launches a ball upwards before it reaches a peak and descends. Five downward accelerations range from 4/6g to 8/6g, including normal gravity at 6/6g. Twelve participants aged 20–26 perform the task while gaze and interception data are recorded.
VR changes displayed dynamics; it does not expose the participant’s whole body to altered physical gravity. That distinction is important when considering applications to unusual gravitational environments.
Seeing, Predicting and Moving Are Different Processes
Eye movements help gather information about the ball’s trajectory. Smooth pursuit can track a moving target, while rapid saccades reposition gaze. The hand must then be timed and positioned to intercept, using both current information and a prediction of what happens next.
Accurate gaze is therefore potentially useful without being identical to accurate interception. A person can look close to the ball yet mistime the hand movement; another can briefly look away but still make a useful prediction from earlier motion.
graph TD
A[Virtual ball rises and descends] --> B[Visual motion under selected acceleration]
B --> C[Gaze tracking]
B --> D[Prediction of future ball position]
E[Prior experience of falling motion] --> D
D --> F[Hand interception]
C --> G[Gaze accuracy measurement]
F --> H[Interception error measurement]
G --> I[Compare across acceleration conditions]
H --> I
Original explanatory diagram of the study’s measurements and interpretation, not a model fitted by the thesis or a reproduced paper figure.
What the Results Show
The mean interception error was closest to zero at 5/6g. Gaze accuracy decreased as acceleration increased. Correlations between gaze accuracy and interception error were generally weak, with a stronger association at 7/6g when considering only data above 75% gaze accuracy.
These findings suggest that performance in the tested VR task does not reduce to a simple rule that normal gravity always gives the smallest mean error. They do not establish that people generally prefer weaker gravity or that 5/6g is optimal for every interception task.
The restricted correlation deserves particular care. Selecting trials above a gaze-accuracy threshold changes the analysed sample. The relationship observed in that subset is not automatically the relationship across all trials or all participants.
Bias Is Not the Same as Precision
An error measure can describe direction as well as size. Opposing signed errors may cancel in a mean, so an average close to zero indicates low average bias rather than perfect performance in every attempt.
This is why the interpretation should consider the distribution of errors and the precise outcome definition. Gaze accuracy, mean interception error and variation between attempts each capture a different aspect of coordination. More repeated trials also do not turn a small participant group into a large independent sample.
What Reproduction Would Need
The original thesis is the authority for the apparatus, task and analysis. Reproducing its human results would require the VR scene and ball dynamics, starting conditions, display timing, gaze calibration, synchronised sampling, participant protocol and error definitions.
Matching the kinematic equations alone would reproduce a simulated trajectory, not the perceptual result. Display and tracking delays can affect when a participant sees the ball and when the system records the response, so timing belongs in the measurement chain.
The findings can inform research into virtual training and visuomotor adaptation. Rehabilitation is a possible future application, not an outcome demonstrated by this study of a small, young participant group; no clinical benefit is established here.
Testing and Real-World Use
This could be tested in a calibrated VR task by changing only the ball’s acceleration and comparing gaze accuracy, interception bias and error variability across matched conditions. In virtual training research, the results could help designers understand when simulated motion feels predictable and how altered dynamics affect eye–hand coordination.
Thesis and Original Figures
The September 2023 thesis on Zenodo includes the full text, experimental illustrations and results. The earlier eye-tracking project page offers related development context, while gaze-guided robot teleoperation is a separate project.