Can AR glasses make
grocery shopping better?
A controlled within-subjects evaluation of head-worn augmented reality for product navigation, information display, and comparison in grocery retail - measuring task performance, usability, and cognitive workload on Meta Quest 3.
Grocery shopping is
surprisingly broken.
Shoppers navigate large, unfamiliar stores using static signage and their own memory - losing significant time just finding products. Contextual data like allergens, nutritional info, and price comparisons is scattered across packaging, shelf labels, and separate apps.
When choosing between similar products, shoppers must mentally compare information from multiple sources simultaneously - increasing cognitive load and slowing decisions.
Head-worn AR overlays digital information directly onto the user's visual field. In principle, it could address all of these frictions at once. But does it actually work? No controlled study exists.
Navigation inefficiency
Locating products requires store familiarity or staff assistance. Most shoppers don't have either in an unfamiliar store.
Information fragmentation
Ingredients, allergens, nutrition, pricing - all scattered across packaging, shelf labels, and separate apps.
Comparison overload
Choosing between alternatives demands mentally juggling multiple data points from multiple physical sources.
The literature divides.
Marketing vs HCI.
The existing academic literature on Augmented Reality in retail is distinctly bifurcated. The vast majority evaluates AR through consumer behavior frameworks (focusing on subjective constructs like immersion and purchase intention), completely omitting objective HCI evaluations such as task completion time, spatial navigation efficiency, or cognitive workload.
The Context Gap
A 2025 meta-analysis encompassing 88 empirical AR wayfinding studies revealed that while AR excels in campuses and hospitals, retail and grocery contexts account for less than 1% of published research.
The Hardware Penalty
Research shows navigation performance is strictly bound by modality. A benchmark study (Neeson et al., 2025) proved head-mounted displays outpace smartphones during indoor navigation, as mobile "head-down" interactions actively impair spatial tracking.
Physical Environment Neglect
Spatial computing introduces critical perceptual risks. In dynamic physical spaces, an interface that commands disproportionate visual attention could cause users to ignore physical pricing, static aisle labels, or physical obstacles entirely.
Three research streams.
None answer this question.
No published peer-reviewed study has conducted a controlled HCI evaluation of a head-worn AR shopping assistant - combining list-based navigation, contextual product information display, and quick-comparison overlays - against an unassisted baseline in a grocery retail scenario, using task-performance metrics, standardized usability scales, and cognitive workload measures.
Qiu et al.'s (2025) systematic review of 88 AR wayfinding studies found zero situated in retail. Wolniak et al. (2024) explicitly notes that AR in grocery “has not been studied in the academic field yet.” This thesis fills that gap - a narrow empirical contribution, positioned precisely at the intersection that existing work skirts but never occupies.
Four questions.
Measured, not assumed.
To what extent does a head-worn AR shopping assistant improve task completion time, items found correctly, and product selection accuracy compared to unassisted shopping?
How do users rate the usability of the AR system, as measured by the System Usability Scale (SUS)?
Does the AR system change perceived cognitive workload compared to unassisted shopping, as measured by Raw NASA-TLX?
What interaction breakdowns, usability barriers, and design trade-offs emerge - particularly around visual clutter, information placement, and social comfort?
Three features.
One shopping journey.
Built in Unity 6 for Meta Quest 3 (video pass-through mode). The simulated store is a physical lab space with real shelves and real products - augmented with virtual wayfinding cues and information panels. Participants physically walk through the mapped space.
List-Based Navigation
Floor-path wayfinding + product highlighting
The user loads a 6-item shopping list. The system calculates a route through the store and renders a NavMesh floor-path line guiding them to each item. On arrival, the target product is visually highlighted on the shelf.
Why: 12/20 survey respondents identified "difficulty finding a specific product" as their primary pain point.
Contextual Info Overlay
On-demand product information
When the user looks at or approaches a product, a world-anchored label displays price, nutritional data, and allergen flags. Triggered by proximity or gaze - not persistent for all visible products - to minimise visual clutter.
Why: Survey: 17/20 requested price info. Design follows Hoffmann et al. (2022): on-demand outperforms fully user-controlled disclosure.
Quick-Comparison Overlay
Side-by-side attribute comparison
For tasks requiring a choice between alternatives (e.g., "find a gluten-free pasta"), the system displays 2–3 relevant products side by side with key differentiating attributes. Disappears once a product is selected.
Why: Extends ARShopping (Xu et al., 2022) and Álvarez Márquez & Ziegler (2023) - integrating comparison into a complete shopping workflow rather than studying it in isolation.
Running price calculator · stock availability · recommendation engine · social features · voice interaction · live IoT backend. Data is pre-loaded as local JSON. Scoped to keep core interaction patterns - navigate, find, inspect, compare - evaluable without confounding.
Controlled.
Within-subjects.
Counterbalanced.
Each participant completes the same shopping task in both conditions - AR prototype and unassisted baseline. Order is counterbalanced to control for learning effects (half do Condition A then B, half do Condition B then A), preceded by a small pilot group (n=2–3) to refine the protocol. To simulate everyday lightweight AR glasses despite current hardware constraints, Meta Quest 3 is used as a prototyping proxy. Task sets use different but equivalent product lists to prevent spatial memory carryover.
Unassisted Shopping
Paper shopping list. No AR navigation or information assistance. Natural smartphone use permitted - checking lists, looking up info, any normal phone behaviour.
AR Shopping Assistant
Meta Quest 3 running the prototype. Video pass-through of the physical lab space, augmented with floor-path wayfinding, product information overlays, and quick-comparison panels.
Task Structure
Complete a 6-item shopping list in a simulated grocery store - a physical lab space with real shelves and real grocery products. For 2 of the 6 items, an additional decision sub-task: choose between alternatives based on a given criterion.
Session Flow
n = 20 regular shoppers.
Clear tensions emerged.
A qualitative requirements survey (March 2026) explored shopping habits, pain points, attitudes toward AR, and information preferences. Findings shaped every design decision in the prototype.
Participate in our ongoing survey here →
Identified "difficulty finding a specific product" as their primary pain point
Selected "finding products quickly" as the single most valuable AR feature
Requested detailed product information (allergens, nutrition, etc.) as an essential feature
"Similar products are all over the store which makes it difficult to see and compare similar items in one place." - Survey Respondent
Shoppers expressed high cognitive load and physical frustration from having to sequentially pick up multiple items just to compare nutrition or allergen facts.
Static overhead signs are inadequate for locating specific niche items, directly validating the prototype's dynamic NavMesh list-based routing.
How findings shaped the prototype
Core features prioritize product finding and information display - not social or recommendation features
Product information overlay uses on-demand, proximity-triggered activation - not persistent display for all visible products
Baseline condition uses unassisted shopping with natural phone access - reflecting how most respondents actually shop today
Quantitative and qualitative.
Rigorously instrumented.
Task Performance
Usability
Cognitive Workload
Qualitative
Shapiro-Wilk normality test → paired t-test or Wilcoxon signed-rank test. Effect sizes reported (Cohen's d / r). n = 16–20 in a within-subjects design is powered to detect medium-to-large effects. Qualitative: lightweight thematic analysis with deductive codes supplemented by emergent inductive codes.
Phase 1 complete.
Prototype in development.
Requirements Survey
Weeks 1–2Completedn = 20 qualitative requirements survey with regular grocery shoppers. Completed March 2026.
Design & Prototyping
Weeks 2–7In progressLo-fi wireframes, hi-fi Unity prototype on Meta Quest 3, simulated store setup with real shelves and products.
Pilot Test
Weeks 7–82–3 pilot participants. Bug fixing, protocol refinement, timing calibration.
Controlled User Study
Weeks 8–1216–20 participants. Full within-subjects study with data collection.
Analysis
Weeks 12–14Quantitative analysis, interview transcription, thematic coding.
Writing
Weeks 14–16Results, discussion, design implications, final thesis document.
What this research
adds to the field.
Empirical Evidence
The first controlled HCI evaluation of a head-worn AR grocery assistant measuring task performance, usability, and cognitive workload against an unassisted baseline.
Design Knowledge
Documented design implications grounded in observed user behaviour: navigation mode transitions, overlay clutter, comparison usage, social comfort thresholds.
Design Guidelines
Practical guidelines for head-worn AR in retail: information placement, progressive disclosure in dense shelf environments, cognitive load management.
Research Artifact
A functional prototype with architecture documentation and interaction design rationale - usable as a research platform for follow-up studies.
What this thesis does not claim: A generalizable framework, a novel interaction technique, or a theoretical contribution. It does not claim AR is universally better than non-AR shopping, nor ecological validity for real grocery stores. Its value is empirical and situated. Even null or negative results are explicitly planned for - a finding that AR is slower or increases workload is equally valuable with careful analysis of why.
Ask the Research Assistant
Have detailed questions about the methodology, the preliminary survey data, or the technical implementation? Ask the AI, which is securely restricted to answer only based on the thesis proposal files.