ReAlign: Closed-Loop Support for Continuous Virtual Reality Tasks

Continuous virtual reality (VR) tasks require sustained interaction, but human task behavior naturally fluctuates. We frame support in these tasks as a closed-loop policy problem: passive head-mounted display signals provide an online, operational estimate of task-directed stability (how consistently ongoing behavior remains aligned with the task), and declines in this estimate trigger brief, response-free cues. Because such a policy cannot be judged by average performance alone, we propose a three-axis evaluation framework covering event-locked recovery, efficacy maintenance across repeated exposures, and bounded false-alarm cost. We instantiate this framing in the ReAlign pipeline, an engine-aware system driven by gaze dispersion and head-motion jitter, which governs a state-triggered Adaptive policy. In a within-subjects study ($N$ = 48), the Adaptive policy was associated with faster event-locked recovery than a perceptually matched fixed-interval Periodic policy (median Time-to-Realignment 6.2 vs. 10.8 s), flatter efficacy decay across repeated cue deliveries, and lower perceived disruption; catch trials delivered during high-stability periods incurred a reaction-time delay within a prespecified ±25ms equivalence margin. These findings suggest that closed-loop support can serve as lightweight task scaffolding. Beyond this system, the framework offers reusable policy-level evaluation logic for adaptive extended reality (XR) support.