VRISE: Multifaceted Applications Across Domains

• VRISE is a polysemous term defined differently across immersive VR, surveying education, explainable AI, and vision-language research.
• In VR research, VRISE quantifies adverse symptoms via the VRNQ, highlighting how hardware quality and design choices reduce nausea and disorientation.
• In surveying and AI applications, VRISE denotes innovative platforms and methods using simulation, Voronoi-based saliency, and contextual model instantiation.

VRISE is a polysemous acronym spanning several research domains. In immersive virtual reality methodology, it most commonly denotes Virtual Reality Induced Symptoms and Effects, a cluster of adverse physiological and perceptual symptoms associated with head-mounted display use, including nausea, dizziness, disorientation, fatigue, and postural instability (Kourtesis et al., 2021). In other literatures, the same acronym denotes Virtual Reality for Immersive and Interactive Surveying Education, a virtual laboratory for surveying instruction (Udekwe et al., 30 Jul 2025); Voronoi-RISE, a model-agnostic saliency method derived from RISE (Sakowicz, 2022); and, in one vision-language paper, simply refers to RISE instantiated in the vision-language setting rather than a distinct framework (Hu et al., 17 Aug 2025). The term therefore requires domain-specific disambiguation.

1. Principal senses of the acronym

The acronym is used in at least four technically distinct ways in the cited literature.

Sense	Expansion	Domain
VRISE	Virtual Reality Induced Symptoms and Effects	Immersive VR, neuroscience, neuropsychology (Kourtesis et al., 2021)
VRISE	Virtual Reality for Immersive and Interactive Surveying Education	Civil engineering and surveying education (Udekwe et al., 30 Jul 2025)
VRISE	Voronoi-RISE	Explainable AI and computer vision (Sakowicz, 2022)
VRISE	RISE in the vision-language setting	Vision-LLM training (Hu et al., 17 Aug 2025)

This terminological multiplicity has substantive consequences. In the VR HMD literature, VRISE refers to an adverse-symptom construct central to participant safety and experimental validity. In the later surveying, explainability, and vision-language papers, by contrast, VRISE is a system or method name rather than a symptom construct. A common source of confusion is the 2025 vision-language usage: that paper states explicitly that “VRISE” simply denotes RISE instantiated in the vision-language setting, while the framework itself remains RISE (Hu et al., 17 Aug 2025).

2. VRISE as Virtual Reality Induced Symptoms and Effects

In immersive HMD research, VRISE denotes adverse symptoms arising during or after VR exposure. One formulation defines VRISE as comprising nausea, dizziness, disorientation, fatigue, and postural instability, and places it in close relation to motion sickness, cybersickness, and “VR sickness” (Kourtesis et al., 2021). A broader technological review groups the syndrome under cybersickness or simulator sickness and includes nausea, oculomotor strain, disorientation, headache, dizziness or instability, and fatigue (Kourtesis et al., 2021).

The construct matters because its effects are methodological as well as clinical. Longer exposure increases both the probability and intensity of symptoms, while linear and angular accelerations can provoke intense symptoms even during short exposures (Kourtesis et al., 2021). VRISE can degrade cognitive performance, modulate physiological variables such as heart rate and body temperature, and alter neurophysiology, including cerebral blood flow, oxyhemoglobin, and EEG-derived measures, thereby threatening both participant safety and the interpretability of behavioral and neuroscientific data (Kourtesis et al., 2021).

The technological review and meta-analysis of 44 neuroscience and neuropsychology studies operationalized VRISE at the study level as a dichotomous presence or absence indicator because only six studies reported quantitative questionnaire-based symptom data (Kourtesis et al., 2021). This reporting limitation is itself significant: it indicates that, historically, VRISE was often acknowledged as a practical problem without being measured with sufficient granularity for comparative synthesis.

3. Measurement, validation, and mitigation in immersive VR research

A central development in the quantitative study of VRISE is the Virtual Reality Neuroscience Questionnaire (VRNQ), designed to assess both VR software quality and VRISE intensity (Kourtesis et al., 2021). The VRNQ contains 20 items across four 5-item subscales: User Experience, Game Mechanics, In-Game Assistance, and VRISE. Responses use a 7-point Likert scale, and the scoring is defined as

$$S_{\text{total}} = \sum_{i=1}^{20} x_i,\qquad S_{\text{domain}}(d) = \sum_{j=1}^{5} x_{d,j}.$$

Subscale scores therefore lie in $[5,35]$ and total scores in $[20,140]$ (Kourtesis et al., 2021).

The instrument was validated on 40 participants across three VR sessions. Internal reliability was excellent for all subscales, with Cronbach’s $\alpha = 0.89$ for User Experience, $\alpha = 0.89$ for Game Mechanics, $\alpha = 0.90$ for In-Game Assistance, and $\alpha = 0.89$ for VRISE. Confirmatory factor analysis on 120 observations yielded good fit, with $\chi^2/df = 1.61$, CFI $= 0.954$, TLI $= 0.938$, SRMR $[5,35]$0, and RMSEA $[5,35]$1 (Kourtesis et al., 2021). The questionnaire also defines two median-based suitability thresholds: minimum cut-offs of at least 25/35 per subscale and 100/140 total, and parsimonious cut-offs of at least 30/35 per subscale and 120/140 total.

The same study reported that immersive VR sessions can be scheduled for 55–70 minutes without pertinent adverse symptomatology provided that the software meets or exceeds the parsimonious VRNQ cut-offs and users are appropriately familiarized with the system (Kourtesis et al., 2021). Session duration correlated positively with total VRNQ score ($[5,35]$2, $[5,35]$3, $[5,35]$4), and VRISE scores were predominantly 6 or 7, corresponding to very mild or absent symptoms. Deeper immersion, better graphics and sound, and more helpful instructions and prompts were all associated with lower VRISE intensity, while age and education were not correlated with VRNQ scores or session duration; gaming experience affected early exposure but not session duration after familiarization (Kourtesis et al., 2021).

These findings align with the technological review’s hardware and interaction recommendations. Minimum criteria identified there include diagonal FOV $[5,35]$5, refresh rate $[5,35]$6 Hz, per-eye resolution $[5,35]$7 sub-pixels, OLED or improved LCD displays, precise 6DoF tracking, teleportation-based locomotion, ergonomic controllers or direct-hand interaction, and appropriately calibrated spatialized audio (Kourtesis et al., 2021). In the meta-analysis, new-generation commercial HMDs induced significantly fewer VRISE than old-generation systems ($[5,35]$8), and 11 studies using commercial-version new-generation HMDs with ergonomic interactions reported 0 VRISE and 0 dropouts among 546 participants (Kourtesis et al., 2021).

A software-development case study, VR-EAL, operationalized these principles in neuropsychological assessment. Its final version achieved VRNQ medians of 128/140 total, 31/35 for User Experience, 32/35 for Game Mechanics, 32/35 for In-Game Assistance, and 33/35 for VRISE, thereby exceeding all parsimonious cut-offs during an average immersion of 62.2 minutes. The paper attributes the near-eradication of VRISE to stable performance at approximately 120–140 fps, teleportation, 6DoF interactions, spatialized audio, haptics, lightmapping, mesh baking, and expanded in-game assistance (Kourtesis et al., 2021).

4. VRISE as a virtual laboratory for surveying education

In civil engineering education, VRISE denotes Virtual Reality for Immersive and Interactive Surveying Education, a VR laboratory intended to reproduce core ground-based and aerial surveying workflows with task-specific interaction design, adaptive supports, and real-time feedback (Udekwe et al., 30 Jul 2025). The platform targets persistent constraints in surveying instruction, including weather dependence, equipment availability, limited repetition opportunities, and the high cognitive and motor demands of precise instrument handling.

The system was implemented for the Meta Quest 3 and developed in Unity 2022 LTS with Universal Render Pipeline, XR Plugin Management, OpenXR, and the Unity XR Interaction Toolkit. Runtime configuration used OpenGLES3, Linear color space, 2× or 4× MSAA, baked lighting, ASTC texture compression, and a target frame rate of 72–90 FPS to minimize motion sickness. The virtual environment consists of an Orientation Scene and a Surveying Field Scene, and an in-scene tablet allows live adjustment of graphics quality, post-processing, audio, grass density, and pedestrian or motorist visibility (Udekwe et al., 30 Jul 2025).

Two instructional modules are central. The first is ground-based differential leveling, including tripod positioning, rough setup, fine leveling via tribrach screws, target acquisition, and backsight/foresight reading. Core computations follow standard leveling relations:

$[5,35]$9

The reported error metric is the elevation relative error at point B,

$[20,140]$0

The second is aerial surveying with waypoint-based navigation, in which students fly a virtual sUAS along straight and curved waypoint sequences using HUD telemetry and joystick-based control (Udekwe et al., 30 Jul 2025).

A distinctive technical feature is controller-input stabilization through Single Exponential Smoothing with $[20,140]$1:

$[20,140]$2

A deadzone rule suppresses residual micro-movements by setting $[20,140]$3 when $[20,140]$4 (Udekwe et al., 30 Jul 2025). The stated purpose is reduction of high-frequency jitter during aiming and manipulation.

Evaluation was explicitly preliminary. A single participant with basic familiarity with both VR and surveying completed five differential-leveling attempts and five attempts for each of two aerial paths. In differential leveling, $[20,140]$5 decreased from 0.4% on Attempt 1 to 0.05% on Attempt 5, task time declined from approximately 320 s to approximately 265 s by Attempt 4, and interaction counts fell from 30 to 15 actions, despite a transient disruption on Attempt 3. In aerial navigation, Path 1 showed improving accuracy and stable timings of approximately 110–127 s, while Path 2 remained longer and more variable, reflecting greater curvature and control complexity. No inferential statistics were reported because $[20,140]$6 (Udekwe et al., 30 Jul 2025).

The platform’s significance lies less in established efficacy than in feasibility. The paper presents VRISE as an accessibility-oriented virtual laboratory aligned with Universal Design for Learning principles, but also states that broader user studies, additional modules, and comparative field-performance data remain future work (Udekwe et al., 30 Jul 2025).

5. VRISE as Voronoi-RISE in explainable computer vision

In explainable AI, VRISE denotes Voronoi-RISE, a perturbation-based, model-agnostic saliency method that modifies RISE by replacing square occlusion grids with convex polygonal occlusions generated from random Voronoi tessellations and by adding an informativeness guarantee to the mask generator (Sakowicz, 2022). The motivation is twofold: to reduce grid-induced “feature slicing” and to accelerate convergence when standard threshold-based mask sampling produces uninformative all-zero or all-one masks.

The baseline RISE aggregation for class $[20,140]$7 and image $[20,140]$8 is

$[20,140]$9

where $\alpha = 0.89$0 are soft masks and $\alpha = 0.89$1 is the nominal visible-cell probability (Sakowicz, 2022). VRISE renders masks at native resolution from Voronoi polygons, applies Gaussian blur, and weights each mask by its realized fill rate:

$\alpha = 0.89$2

This per-mask normalization is intended to compensate for deviations from the independent Bernoulli assumption when guaranteed generators are used (Sakowicz, 2022).

The method uses random seeds inside an inspected area together with “fencepost” seeds placed outside the area along diagonal extensions to ensure that all relevant Voronoi cells are finite. Multiple independent meshes replace RISE’s random grid shifts. VRISE also introduces guaranteed occlusion selectors, including coordinate-based and permutation-based variants, plus a hybrid fix-on-demand generator. The analytic probability that threshold sampling yields an uninformative mask is

$\alpha = 0.89$3

When $\alpha = 0.89$4 is non-negligible, eliminating empty or full masks improves early sample efficiency (Sakowicz, 2022).

Quantitative evaluation on the ILSVRC2012 validation split used ResNet-50 and VGG-16, a saliency-guided insertion/deletion “Alteration Game,” and the Pointing Game. For ResNet-50 with $\alpha = 0.89$5, $\alpha = 0.89$6, $\alpha = 0.89$7, $\alpha = 0.89$8, and meshcount $\alpha = 0.89$9, VRISE decreased Remove AUC by approximately 7.1%, increased Sharpen AUC by approximately 1.2%, and decreased Pointing Game accuracy by approximately 1.5%. For VGG-16, Remove AUC improved by approximately 8.4%, Sharpen AUC increased by approximately 1.2%, and Pointing Game decreased by approximately 1.8%. Improvements were larger in multi-object images, but gains were not universal, particularly for Pointing Game localization (Sakowicz, 2022).

The paper also identifies two characteristic failure modes. Feature slicing occurs when mask boundaries divide a salient part and reduce its recognizability. Saliency misattribution occurs when salient and non-salient areas co-occur within the same mask, causing score mass to spread to pixels that are not truly causal. The proposed mitigations include increasing meshcount, using moderate Gaussian blur, and keeping $\alpha = 0.89$0 in the 20–50% range (Sakowicz, 2022). This makes VRISE a targeted refinement of RISE rather than a uniformly dominant replacement.

6. VRISE in the vision-language literature

A separate 2025 paper uses “VRISE” in a narrower and explicitly non-independent sense: it denotes RISE instantiated in the vision-language setting (Hu et al., 17 Aug 2025). The underlying framework is RISE, expanded as Reason–Inspire–Strengthen–Expertise, a self-supervised reasoning-centric training procedure for vision-LLMs on complex image annotation tasks such as emotion classification and context-driven object detection.

The method has two stages. In RISE-CoT, the model generates a chain of thought $\alpha = 0.89$1 from image $\alpha = 0.89$2 and annotation $\alpha = 0.89$3, then reconstructs the annotation $\alpha = 0.89$4 from $\alpha = 0.89$5, and receives a reward

$\alpha = 0.89$6

For classification, the similarity term combines KLD, MSE, and a soft constraint that predicted probabilities sum to approximately 1; for detection, it uses IoU under Hungarian matching. The policy objective is optimized with GRPO:

$\alpha = 0.89$7

This yields an enriched dataset of verified chains of thought (Hu et al., 17 Aug 2025).

In RISE-R1, high-quality chains of thought with reward $\alpha = 0.89$8 and $\alpha = 0.89$9 are used for supervised fine-tuning on unified “think–answer” targets, followed by reinforcement fine-tuning with a reward based on end-to-end answer quality and format compliance. Implemented on Qwen2-VL-2B, the method outperformed standard SFT and Visual-RFT on Emotion6 and LISA, and also improved ImageNet-Sub and COCO-Sub in full-shot settings. On Emotion6, RISE obtained JSD values of 0.168, 0.133, and 0.071 for 4-shot, 16-shot, and full-shot protocols respectively; on LISA, it obtained [email protected] values of 0.195, 0.271, and 0.404 (Hu et al., 17 Aug 2025).

The important terminological point is that this usage does not define a new method called VRISE. The paper states that “VRISE” is merely the vision-language instantiation of RISE. In encyclopedic treatment, it is therefore better classified as a local naming convention than as a separate acronymal tradition (Hu et al., 17 Aug 2025).

7. Ambiguity, misconceptions, and domain-specific significance

The main interpretive challenge surrounding VRISE is acronymal ambiguity. In immersive VR methodology, VRISE is a symptom construct with direct implications for ethics, safety, and inferential validity. In surveying education and explainable AI, it is the name of a platform or algorithm. In vision-language modeling, it is only a contextual label for RISE (Kourtesis et al., 2021).

Several recurrent misconceptions can be addressed directly from the cited literature. One is that VRISE in immersive VR is an unavoidable consequence of HMD use. The evidence instead indicates that symptom intensity is strongly modulated by hardware generation, display and tracking characteristics, locomotion design, assistance features, rendering stability, and user familiarization; under commercial-version new-generation HMDs with ergonomic interactions, the meta-analysis found zero VRISE and zero dropouts in 11 studies (Kourtesis et al., 2021). Another is that the surveying platform already has broad efficacy evidence; the published evaluation is explicitly a single-participant feasibility study without inferential statistics (Udekwe et al., 30 Jul 2025). A third is that Voronoi-RISE uniformly improves on RISE across metrics; the reported improvements are strongest for coarse meshes and multi-object images, while Pointing Game accuracy can decline slightly (Sakowicz, 2022).

Across these domains, the acronym consistently signals a concern with interaction between representation and evidence: symptom formation under immersive display conditions, procedural learning in simulated surveying, spatial attribution in saliency maps, and reasoning-grounded annotation in vision-LLMs. The commonality is lexical rather than theoretical. The substantive meaning of VRISE is determined entirely by disciplinary context.