Self-Guided Attention Redirection (SGAR)

• SGAR is defined as a set of methodologies that leverage real-time attention estimates to modulate system parameters in VR and cognitive tasks.
• Techniques include dynamic gain adaptation based on head orientation and oculomotor-to-tactile feedback for self-regulation.
• Experimental results show improved metrics such as expanded translation gain thresholds and reduced gaze entropy, highlighting SGAR's practical benefits.

Self-Guided Attention Redirection (SGAR) encompasses a set of methodologies that dynamically exploit real-time estimates of user attention to drive adaptive system behavior, thereby amplifying or redirecting attentional resources toward task-relevant stimuli or system goals. SGAR has been concretely operationalized in virtual reality (VR) locomotion through dynamic gain modification using head orientation as an attention proxy (Zou et al., 30 Oct 2025), and in cognitive enhancement settings via mapping oculomotor activity onto somatosensory feedback for self-modulation of attentional state (Xu et al., 2023). Central to the SGAR approach is the closed-loop integration of perceptual/cognitive state estimation, feedback modulation, and stimulus or control adaptation.

1. Conceptual Foundations and Scope

SGAR leverages measurable correlates of attentional focus—such as head direction or oculomotor trajectories—as real-time proxies for the degree and locus of user engagement. Approaches in VR and cognitive neuroscience diverge in their control targets (locomotion gain versus tactile feedback) and attention estimation channels (head-gaze versus direct eye-tracking), yet share a core self-regulation paradigm where the user becomes both actor and sensor, situating the method at the intersection of human-in-the-loop control and embodied attention engineering.

In the VR context, SGAR maximizes the permissible manipulations of a user’s movement-to-virtual-motion mapping by associating control adaptations to moments of focused, task-diverted attention. In cognitive intervention, SGAR via eyerofeedback externalizes latent oculomotor signals, fostering self-awareness and volitional attentional modulation, particularly under high cognitive load or distractor interference.

2. Methodologies and Signal Processing

2.1 VR-based SGAR: Dynamic Gain Adaptation

SGAR for redirected walking (Zou et al., 30 Oct 2025) operates by dynamically adjusting the VR translation gain $G(t)$ in response to inferred attentional engagement with distractors:

• Attention Estimation: The system computes the angular deviation $\theta(t)$ between the user’s head-forward vector and the vector to a distractor. An engagement threshold $d=15^\circ$ delimits central focus.
• Attention Degree (A(t)): Accumulation and decay are governed by:

$$A(t+\Delta t)= \begin{cases} A(t)+a\,e^{-2c^2\Delta t/\theta(t)} & \theta(t)<d, \ A(t)-b\,\Delta t & \theta(t)\geq d, \end{cases}$$

with $(a, b, c, d) = (5000, 2000, 3.1, 15^\circ)$, $A(t)$ clipped to $[0,1]$.

• Gain Mapping: The translation gain is set by $G(t) = G_0 + (G_{\text{max}} - G_0)A(t)$, with $G_0=1.0$, $G_{\text{max}} \in [0.5,\,1.5]$ (trial-dependent).
• Real-Time Implementation: Per-frame logic tracks head orientation, updates $\theta(t)$0, and applies $\theta(t)$1 to modulate virtual displacement.

2.2 Eyerofeedback: Oculomotor-to-Tactile Mapping

SGAR in attentional self-regulation employs real-time mapping of eye gaze to body-based tactile cues (Xu et al., 2023):

• Input Acquisition: Webcam-based eye tracking (WebGazer.js, ~30 Hz). Gaze position is normalized and optionally smoothed.
• Quadrant Classification: The [0,1] × [0,1] gaze space is partitioned into four quadrants, each mapped to one of four vibrotactile motors (wrists/ankles).
• Feedback Logic:
  • Stationary mode: Continuous vibration in a quadrant when gaze remains within region.
  • Filter mode: Vibration triggered only by gaze shifts exceeding a velocity threshold ($\theta(t)$2).
• Activation Equation: $\theta(t)$3 encapsulates the linear and rotational mapping pipeline from gaze to tactile output.

3. Experimental Paradigms and Hardware

3.1 VR Locomotion Experiment Design

• Task Environment: Users escort a virtual cat through a low-cognitive-load, 3D cartoon town, with lateral distractors emerging from occluded positions when triggered by spatial proximity (1.5 m ahead of cat).
• Distractor Types: Human or dog figures moving toward the center; low frequency (~1 per 8 m walking) to avoid overload.
• Participants and Apparatus: $\theta(t)$4; Oculus Quest 2 HMD, Unity 2021.3, OpenRDW, hardware specifications as detailed (Zou et al., 30 Oct 2025).

3.2 Eyerofeedback Attentional Task Protocol

• Attention Task: Three-Choice Vigilance Task (3CVT), with target, non-target, and distractor stimuli. Sessions vary duration (short 2-5 s; long 25-35 s) and distraction condition (with/without background movie).
• Feedback/Control: Within-subject design contrasts feedback-off, stationary, and filter feedback, randomly assigned.
• Hardware/Implementation: Webcam, Arduino-driven four-motor array, JavaScript/Python for pipeline control; latency maintained $\theta(t)$5100 ms.

4. Quantitative Effects and Statistical Outcomes

4.1 VR SGAR: Redirection Threshold Expansion

• Detection Thresholds: Introduction of SGAR yields a widened translation gain imperceptibility window:
  • With distractors: LDT=0.84, UDT=1.32, PSE≈1.08
  • No distractor baseline: LDT=0.91, UDT=1.15, PSE≈1.03
  • Gain-switch control: LDT=0.92, UDT=1.12, PSE≈1.02
• Statistical Analysis: Three-group ANOVA on raw data reveals main effects of gain ($\theta(t)$6, $\theta(t)$7, $\theta(t)$8), group ($\theta(t)$9, $d=15^\circ$0, $d=15^\circ$1), and gain×group interaction ($d=15^\circ$2, $d=15^\circ$3, $d=15^\circ$4). Post-hoc comparisons confirm significant window expansion under SGAR.

4.2 Eyerofeedback: Behavioral and Entropy Improvements

• Response Time: Repeated-measures ANOVA ($d=15^\circ$5, $d=15^\circ$6) and interaction with block duration demonstrated significant main effects and improvements under feedback, particularly for the filter mode during high load/distraction.
• Missed Trials: Reduction with feedback ($d=15^\circ$7, $d=15^\circ$8); effects primary under long/distraction conditions.
• Gaze Entropy: Filtered feedback leads to substantial entropy reduction in gaze patterns under high load ($d=15^\circ$9 z-units, $$A(t+\Delta t)= \begin{cases} A(t)+a\,e^{-2c^2\Delta t/\theta(t)} & \theta(t)<d, \ A(t)-b\,\Delta t & \theta(t)\geq d, \end{cases}$$0), indicating enhanced attentional focus and control.

Measure	VR SGAR (w/ distractor)	Eyerofeedback (filter, long/distraction)
Gain threshold range	0.84 – 1.32	N/A
RT ($$A(t+\Delta t)= \begin{cases} A(t)+a\,e^{-2c^2\Delta t/\theta(t)} & \theta(t)<d, \ A(t)-b\,\Delta t & \theta(t)\geq d, \end{cases}$$1 z-score)	N/A	1.111 vs 1.479 (filter vs silence)
Gaze entropy ($$A(t+\Delta t)= \begin{cases} A(t)+a\,e^{-2c^2\Delta t/\theta(t)} & \theta(t)<d, \ A(t)-b\,\Delta t & \theta(t)\geq d, \end{cases}$$2 H)	N/A	−1.24 (filter vs stationary)

5. Subjective and User-Experience Metrics

5.1 Simulator Sickness and Presence in VR

• Simulator Sickness Questionnaire (SSQ): No significant group or gender effects; TS and subscales stable across conditions.
• Igroup Presence Questionnaire (IPQ): SGAR condition outperformed controls on Spatial Presence (F=5.967, p<.05), Involvement (F=3.466, p<.05), and Experienced Realism (F=20.858, p<.001). Mean PRES (overall presence) was higher under SGAR (5.15 vs 5.04/4.81 in controls).

5.2 Subjective Effects of Eyerofeedback

• Self-Reports: Both stationary and filter feedback improved subjective focus and central fixation over silence. Filter mode favored for daily use and perceived effectiveness, stationary feedback sometimes reported as distractive.
• Interoceptive Awareness: Heightened subjective awareness of gaze and enhanced gaze regulation suggest engagement of interoceptive-inference mechanisms.

6. Implementation Guidelines and Practical Considerations

• VR SGAR Implementation: Distractor placement at lateral edges, triggered by proximity; head direction threshold of 15°, hyper-parameters (a=5000, b=2000, c=3.1), gain cycles lasting ~550 ms. Suitable for low-load goal-directed environments.
• Eyerofeedback Implementation: Multi-point vibrotactile arrays to refine spatial resolution; adaptive thresholding in filtered feedback to reduce habituation; maintain closed-loop latency $$A(t+\Delta t)= \begin{cases} A(t)+a\,e^{-2c^2\Delta t/\theta(t)} & \theta(t)<d, \ A(t)-b\,\Delta t & \theta(t)\geq d, \end{cases}$$3100 ms; recommend on-body processing for wireless, reduced-latency setups.

A plausible implication is that further refinement of spatial and temporal feedback resolution in SGAR systems could extend their reach to more complex, ecologically valid tasks (e.g. search in naturalistic scenes, clinical monitoring). Personalization of mappings (gains, rotational offsets) and minimizing cognitive/emotional load from feedback channels are critical for practical deployment.

7. Theoretical and Applied Implications

SGAR constitutes a closed-loop paradigm in which user-state inference modulates the parameters of external feedback or virtual control, fostering a real-time, bidirectional interaction between user and system. In VR, this widens the manipulation envelope for spatial redirection without degrading experience; in cognitive intervention, it supports volitional control and enhancement of attention, especially under challenging load or distractor conditions.

Both lines of research converge on the principle that making latent physiological and attentional signals salient—through ecological mapping to system adaptations or somatosensory feedback—enables individuals to actively regulate their cognitive state in real time. Future directions include higher-resolution, multimodal feedback, testing in clinical and applied populations, and optimization of closed-loop adaptation to individual differences in attention and perception (Zou et al., 30 Oct 2025, Xu et al., 2023).