Virtual Reality Selections Shift during Confirmation, Mimicking Quantum Uncertainty Principles

Target selection represents a core interaction paradigm within virtual reality, yet the very act of confirming that selection can inadvertently disturb the user’s aim, a phenomenon known as the Heisenberg Effect. Linjie Qiu, Duotun Wang, and Boyu Li, from The Hong Kong University of Science and Technology (Guangzhou), along with Jiawei Li, Yulin Shen, and Zeyu Wang, investigated this effect across different input methods and selection techniques. Their research addresses a significant gap in understanding how the Heisenberg Effect manifests with bare hand input, and whether score-based selection can reduce disturbances compared to direct selection, particularly when considering spatial variations. By conducting a within-subject study, they demonstrate that bare hand input is more prone to this effect, and introduce weighted VOTE, a novel intention accuracy model, which improves selection precision by reweighting recent user intent to counteract input disturbances.

This model dynamically reweights recent interaction intent, effectively counteracting disturbances during input. Evaluation confirms that Weighted VOTE substantially improves selection accuracy compared to baseline techniques, particularly for score-based hand input, reducing error rates from 21.96% to 7.54%, achieving performance comparable to controller-based systems. Participants, a cohort of 24 individuals, completed target selection tasks using both handheld controllers and bare-hand tracking. Data acquisition involved tracking participant pose and interaction data throughout the selection process. The study recorded the initial target, the confirmed target, and the time taken to complete each selection.

A high-precision motion tracking system, operating at 120Hz, captured hand and controller positions with sub-millimetre accuracy. Temporal analysis of interaction data identified the point of confirmation, the button press or pinch gesture, and its correlation with pose disturbances.
Building upon prior vote-oriented techniques, researchers introduced weighted VOTE, a novel history-based intention accuracy model. Weighted VOTE reweights recent interaction intent to counteract input disturbances, effectively smoothing out pose fluctuations during confirmation. The model assigns higher weights to more recent interaction data, prioritising the user’s immediate intention while mitigating the impact of transient disturbances.

Evaluation of weighted VOTE involved comparing its selection accuracy against baseline techniques, demonstrating an improvement in performance metrics. Score-based controller selection exhibited the effect in 47.50% of errors, while score-based hand selection showed it in 46.86% of errors.

These findings offer a novel explanation for the observed inferiority of hand tracking selection accuracy compared to controller tracking. The study revealed that direct selection is more sensitive to target width, influencing performance as predicted by Fitts’s law. To model intention shifts and accuracy trends, a third-degree polynomial was fitted to the probability of correctly matching intended objects to targets.

Evaluations indicated that while both direct and score-based hand input yielded lower final selection accuracy, temporal records revealed intermittent moments of high precision. Building upon existing vote-oriented techniques like VOTE and BackTracer, a weighted VOTE method was introduced, refining historical data contributions to selection decisions.

Ablation studies demonstrated that all input methods benefited from weighted VOTE, with the most substantial improvement observed in score-based hand input. To counteract these disturbances, a weighted VOTE method was introduced, which reweights recent interaction intent based on historical accuracy.

Evaluation of this method showed improved selection accuracy, particularly for score-based selections using bare-hand input, bringing performance closer to that achieved with controllers. The study acknowledges a limitation in not assessing user experience through subjective measures like NASA-TLX or SUS.

Future research could explore adaptive variants of weighted VOTE by incorporating behavioural inputs such as eye-tracking data and real-time error detection to enable dynamic adaptation. Further development of adaptive selection methods, incorporating additional behavioural data, promises to refine these techniques and enhance the overall user experience.