PaperlayerView exact revision on arXiv

Source-linked AI summary

Looking Around by Looking Around: Omnidirectional Gaze-based VR Viewport Control

Hock Siang Lee, Jinghui Hu, Florian Weidner, Haopeng Wang, Hans Gellersen

arXiv:2608.30014v1cs.HC

TL;DR

Traditional VR viewport control can demand substantial head and torso movement, limiting use in constrained or extended settings. LALA addresses this with an asymmetric omnidirectional gaze-control profile for pitch and yaw, designed around eye-movement asymmetries and comfortable gaze regions. In a user study with N=18, LALA was strongly preferred over the traditional baseline while providing competitive performance with hands-free, minimally physical interaction.

  • Problem

    Traditional VR viewport control relies on head and torso movement, which can be effortful and limiting in constrained or extended-use settings.

  • Method

    LALA uses an asymmetric omnidirectional gaze-control profile for pitch and yaw that accounts for oculomotor and perceptual asymmetries and remaps head and viewport rotations for consistent control.

  • Results

    LALA was strongly preferred over the traditional baseline while achieving competitive performance with fully hands-free interaction and minimal physical movement.

  • Takeaways & Limitations

    Gaze-only viewport control is feasible and desirable for VR contexts where comfort and low physical effort are essential.

  • Takeaways & Limitations

    The study evaluated feasibility and usability in a controlled environment, and its short segmented sessions may not generalize to prolonged, embodied, or immersive VR experiences.

Abstract

from arXiv · show

Traditional VR viewport control primarily relies on head and torso movement, which can be effortful and limiting in both constrained and extended-use settings. We introduce Looking Around by Looking Around (LALA), a gaze-based VR pitch-and-yaw viewport control technique designed for natural and effortless omnidirectional exploration via eye movements, without requiring or obstructing movement of the head, hand, or body, offering a low-effort and highly accessible interaction method. Because gaze is primarily used for perception and exhibits oculomotor and perceptual asymmetries, using it directly for control is difficult. To address this, we designed an asymmetric omnidirectional control profile for the eye, then built on it to exploit tendencies for eyes to stay within comfortable regions for viewport control. We evaluated LALA in a user study (N=18) featuring two contrasting tasks: alignment towards known directions and open-ended visual search towards unknown directions. LALA was strongly preferred over the traditional baseline, achieving competitive performance while enabling fully hands-free interaction with minimal physical movement.

1 INTRODUCTION

LALA introduces gaze-based omnidirectional VR viewport control to reduce reliance on large head and torso movements. Its asymmetric eye-control design supports hands-free exploration, with competitive performance and strong user preference.

  • Motivation: Traditional VR viewport control maps head pose directly to the virtual camera, assuming users can freely rotate their head and torso.This assumption is problematic when large movements or external accommodations are impractical.
  • Motivation: LALA uses natural eye movements as a signal for extending the viewport, building on eye–head coordination during visual exploration.The eyes often inspect peripheral regions before the head follows to bring them centrally into view.
  • Design: LALA converts gaze into an asymmetric control profile that accounts for oculomotor and perceptual asymmetries.The design targets stable viewing, smooth continuous control, and rapid reorientation during large gaze shifts.
  • Contribution: LALA enables stable omnidirectional VR exploration through eye movements alone, without requiring or obstructing head or torso movement.Its control profile is designed to steer the viewport toward any direction while maintaining comfortable viewing.
  • Evaluation: LALA achieved performance competitive with existing techniques while enabling fully hands-free interaction with minimal physical movement.Participants strongly preferred LALA and reported favorable subjective experiences during virtual-environment exploration.

2 VIEWPORT CONTROL TECHNIQUES

VR viewport control techniques reduce reliance on unrestricted head and torso rotation through amplification, body-based inputs, visual strategies, and gaze-based methods. These approaches differ in physical effort, hands-free operation, and supported viewport dimensions.

  • Head-based techniques: Head amplification reduces required physical motion by increasing virtual rotation from head movement, while retaining hands-free interaction.Constant gain preserves an absolute relationship between head pose and viewport rotation.
  • Alternative inputs: Controller-based viewport control requires handheld input, whereas body-based techniques use leaning, shoulder rotation, or arm movement.These hands-free body techniques are typically designed for locomotion rather than viewport control.
  • Visual strategies: Visual strategies can expand effective field of view without interaction by scaling panoramic views, layering perspectives, or adding pictures-in-picture.These methods change visual presentation rather than directly controlling viewport orientation.
  • Gaze-based techniques: Existing gaze-based techniques such as Gaze Gain and Gaze Pursuit primarily target 1D yaw-only control and cannot readily support combined pitch-and-yaw rotations.They use different eye movements and prioritize performance over comfort.

3 LALA DESIGN

LALA designs gaze-driven omnidirectional viewport control around the asymmetric, perception-oriented behavior of human eyes. Its modular profile and yaw, pitch, and head-remapping components aim to keep comfortable gaze unobtrusive while supporting larger reorienting shifts.

  • 3.2 Viewport Control: LALA applies relative gaze-based yaw control and absolute gaze-based pitch control, with pitch using an amplified eye-in-world angle to define a goal direction.The pitch formulation is designed around gravitational and proprioceptive reference cues, while yaw continuously rotates toward the horizontal gaze direction.
  • 3 LALA DESIGN: LALA explicitly grounds interaction design in vision science by incorporating physiological and behavioral properties of eye movements and visual perception.The authors identify this explicit use of vision science as a relatively rare contribution in HCI and VR interaction research.
  • 3.1 Asymmetric Omnidirectional Gaze Control Profile: LALA uses an asymmetric omnidirectional gaze control profile to convert eye-in-head positions into activation values for viewport control.The profile is intended to support stable viewing, smooth continuous control, and rapid reorientation.
  • 3.1 Asymmetric Omnidirectional Gaze Control Profile: The profile keeps activation near zero for central fixations, rises through the comfortable mid-range for smooth pursuits, and increases sharply in the peripheral region.A sigmoid formulation provides smooth transitions between these interaction regions.
  • 3 LALA DESIGN: Oculomotor and perceptual asymmetries require distinct treatment of horizontal yaw and vertical pitch rather than simple extrapolation across axes.The paper notes that smooth-pursuit performance is generally superior for horizontal yaw than vertical pitch.
  • 3.3 Head-Movement Remapping: Because gaze rotation is partially decoupled from the HMD, LALA remaps head movements to avoid inconsistent perceived yaw and pitch behavior.The complete design combines the gaze profile, yaw control, pitch control, and head-movement remapping as modular components.

4 EVALUATION

The evaluation compared LALA with head-based techniques in two contrasting VR tasks: known-direction alignment and open-ended search toward unknown directions. Participants selected large, distant targets using gaze plus a button press while seated in a stationary chair.

  • Tasks: The study contrasted efficient target acquisition in known directions with open-ended visual search in unknown directions.The two tasks were chosen to represent different gaze patterns and cognitive demands.
  • Procedure: Targets were selected through eye gaze and an intentional button press rather than gaze dwell.This design emphasized viewport alignment and reduced unintended selections from the Midas Touch problem.
  • Study Environment: The experiment used a 30×30×30m colored virtual cube to provide orientation cues while limiting visual distractions.Subtle wall gradients became lighter in the forward-facing direction.
  • Alignment Task: Alignment targets were placed 10m away in 15-target rings with radiuses of 45°, 90°, and 135°.The radiuses represented increasing head–eye coordination demands, from initially visible targets to directions difficult to reach through head movement.
  • Search Task: Search targets were placed 10m away in 14 directions extending toward cube faces and corners, requiring exploration in all directions.The arrangement included targets directly behind, above, and below the participant.
  • Techniques: The compared techniques were 1xHead, 2xHead with doubled head rotation, and LALA gaze-based pitch-and-yaw viewport control.Naive pitch-and-yaw extensions of prior gaze techniques were excluded because pilots found them disorienting or excessively straining.

5 RESULTS

Results showed technique-dependent completion times and rotation behavior in the Alignment Task, while error rates remained similar across conditions. LALA also reduced subjective physical demand and effort relative to the traditional baseline.

  • Completion Time: A significant Technique × Radius interaction showed that technique affected completion time differently across radius levels.All technique pairs differed significantly within each radius except 1xHead versus 2xHead, while all radius pairs differed within each technique.
  • Error Rate: Error Rate showed no significant Technique × Radius interaction or main effects, with an overall mean of µ = 0.070 (95% CI = ±0.015).Missed selections were similar across all conditions.
  • Physical Movement: A significant Technique × Radius interaction affected cumulative eye, head, and torso rotations.Technique pairs differed significantly for cumulative eye rotation at each radius, and radius levels differed within each technique.
  • Subjective Measures: 1xHead produced significantly higher Physical Demand than both 2xHead and LALA.The comparison was significant for 1xHead versus 2xHead (V=139, p=.010) and 1xHead versus LALA (V=165, p<.001).
  • Subjective Measures: 1xHead also produced significantly higher Effort than LALA, while other subjective metrics showed no significant technique effects.There was no significant effect of technique on VRSQ subscales.

5.2 Search Task Results

In the search task, technique influenced completion time and several subjective workload measures, while error rates did not differ significantly. LALA was most often preferred despite slower completion than head-based techniques.

  • Completion Time: 2xHead was fastest, followed by 1xHead, while LALA had the slowest completion time.2xHead was significantly faster than 1xHead, which was significantly faster than LALA; 2xHead also outperformed LALA.
  • Error Rate: Technique had no significant effect on Error Rate, with an overall mean of μ = 0.070 (95% CI = ±0.014).
  • Subjective Measures: 1xHead was significantly worse than 2xHead for Physical Demand, Performance, Effort, and Frustration.1xHead also had significantly worse Physical Demand than LALA.
  • Subjective Measures: No significant technique effect was found for the remaining subjective metrics, and all effect sizes were small.
  • Preference: LALA was rated the best technique by 10 participants, while 1xHead was rated the worst by 11 participants.LALA and 1xHead were each selected by an absolute majority for best and worst, respectively.

6 DISCUSSION

The discussion frames LALA as a low-effort alternative that trades speed for comfort and preference. Its controlled evaluation supports feasibility, but its scope is limited by the abstract setting and short, interrupted exposure.

  • Overall Trade-off: Across both tasks, LALA was slower but reduced viewing effort relative to 1xHead and was the most preferred technique overall.Participants preferred LALA over 2xHead despite 2xHead’s faster performance; the techniques produced similar effort and error levels in that comparison.
  • Performance Spectrum: 2xHead enabled faster alignment for large angular displacements, but high control-display gain could cause overshooting and make precise alignment harder.
  • LALA’s Design Trade-off: LALA’s slower performance was intentional: its velocity profiles prioritized smooth motion, stability, and comfort over speed.The discussion states that increasing velocity would likely induce cybersickness, lower stability, and reduce comfort.
  • Design Contribution: LALA uses an asymmetric omnidirectional gaze profile to support stable, hands-free viewport control without constraining head movement.The profile abstracts eye-movement complexity from downstream viewport technique design.
  • Limitations: The study was an abstract feasibility evaluation in a controlled environment, so its findings may not generalize to longer, more embodied, or immersive experiences.Participants spent approximately 40 minutes in VR across six segments with breaks, limiting naturally accumulating fatigue effects.
  • Application Scope: The authors position 1xHead for highly interactive experiences and LALA for prolonged viewing or constrained-mobility contexts.They describe these as context-dependent trade-offs rather than a universal replacement of head-based control.

7 CONCLUSION

LALA provides omnidirectional gaze-based VR viewport control designed to minimize physical movement and emphasize intuitiveness. The study found strong preference for LALA alongside hands-free interaction and reduced physical effort.

  • LALA combines an asymmetric omnidirectional eye-control profile with comfortable-region gaze behavior for pitch-and-yaw viewport control.
  • In a user study with N=18, LALA was strongly preferred over the traditional baseline while enabling fully hands-free interaction with minimal physical movement.
  • The findings suggest that intuitiveness and reduced physical strain can outweigh efficiency in shaping viewport-control preferences.

Source Code for LALA and the study

The paper states that source code for LALA and the entire project is available at a location redacted for anonymity.

  • Source code for LALA and the entire project is listed as available, but the repository location is redacted for anonymity.

Pseudo Code for rotation remapping

The supplied passages document subjective and quantitative result tables and figures for the Alignment and Search studies, including statistical-analysis conventions. They do not provide pseudocode or rotation-remapping details.

  • Table 5 reports Friedman statistics for subjective data in the Alignment Task, with effect sizes given as Kendall’s W.
  • Figure 10 presents VRSQ by technique for the Alignment Task, where lower values are better and error bars show 95% confidence intervals.
  • Table 6 and Table 7 document Search Task quantitative and subjective statistics, while Figure 11 presents Search Task VRSQ by technique.
  • The quantitative analyses used two-way repeated-measures ANOVA, with Greenhouse–Geisser corrections for sphericity violations and ART when normality was violated.

Search Study Results by Direction

Search results varied by direction: LALA and 2xHead were largely direction-independent, whereas 1xHead degraded in the rear hemisphere. Completion-time and error-rate comparisons further showed direction-specific technique effects.

  • LALA and 2xHead were largely direction-independent, while 1xHead showed significant performance degradation in the rear hemisphere.
  • Completion Time: 1xHead had significantly slower alignment times in Back than at side or frontal directions.
  • Figure 11 reports Search Task VRSQ by technique with 95% confidence intervals, and lower values indicate better scores.
  • Completion Time: 2xHead was significantly faster than both 1xHead and LALA in Back and Up directions, whereas 1xHead outperformed LALA in Front.
  • Error Rate: Direction produced a significant Technique × Direction interaction for Error Rate, but no individual condition differences were significant.

Max Eye, Head, and Torso Rotations

Direction affected eye and head rotation differently across techniques, while torso rotation showed broadly similar pairwise outcomes. LALA and 2xHead generally maintained directionally stable movement patterns compared with 1xHead.

  • Max Eye Pitch: Only 2xHead showed no significant between-direction differences in Max Eye Pitch.
  • Max Head Pitch: Both LALA and 2xHead had lower Max Head Pitch than 1xHead across all directions.
  • Figures 12–14 visualize completion time, maximum eye/head/torso pitch and yaw, and cumulative eye/head/torso rotation by cardinal direction.
  • Max Head Yaw: 1xHead showed higher Max Head Yaw than LALA across all directions and higher values at Back than Front.
  • Max Torso Rotation: Max Torso Pitch had no significant Technique × Direction interaction, while Max Torso Yaw showed a difference only for 1xHead, which was higher at Back.
  • Cumulative Eye Rotation: LALA and 2xHead showed no significant between-direction differences in cumulative eye rotation, unlike 1xHead in upper-rear directions.
  • Cumulative Head Rotation: 1xHead produced more cumulative head rotation than LALA across nearly all directions and more than 2xHead in Up, Down, and rear directions.
  • Cumulative Torso Rotation: Cumulative torso rotation showed no significant pairwise differences, with an overall mean of M = 7.116° (95% CI = ±1.727°).
Loading 2608.30014v1…