This project is part of the NUS Innovation & Design Programme (iDP), undertaken as a collaboration between the National University of Singapore (NUS) and Stanford University, with Meta as the industry partner. Meta's project brief challenged the team to explore how smart glasses can be integrated into our daily activities. The core task was to discover a compelling market application for Meta's Ray-Ban smart glasses platform. We were encouraged to begin with deep user research, uncover genuine pain points, and then propose a solution where the smart glasses form factor delivers a meaningfully better experience than existing alternatives.

We identified recreational golf as a uniquely high-value application for smart glasses technology. The global golf industry represents a substantial and growing market: The global golf industry is substantial and growing: the U.S. golf courses and country clubs sector alone reached US$41.0 billion in 2024 [1], the global golf equipment market is estimated at US$9.24 billion in 2025 [2], and the golf GPS device market — the segment most directly comparable to our product — was valued at US$3.2 billion in 2024, growing at 8.5% CAGR [3], confirming that golfers already pay for on-course technology. Golf is also an ideal form-factor fit: it is a hands-occupied, eyes-up sport where pulling out a phone or rangefinder disrupts the pre-shot routine, and where invisible environmental variables (wind, temperature, altitude) directly affect every shot. This combination of a large, tech-receptive market with well-documented performance pain points made golf the strongest candidate for an AR-assisted, AI-powered smart glasses application.

"The adrenaline was pumping a little bit." [4] This was professional golfer Richard T. Lee's explanation after a stunning miss: a 2-foot putt on March 15, 2026, that cost him the Aramco LIV Golf Singapore title. If a seasoned pro can "choke" on a distance usually considered unmissable, it begs a fundamental question: why do golfers fail to consistently execute the shots they are physically capable of making?

The answer lies in a hidden battle between mental focus and environmental complexity. Golf demands instantaneous decisions based on invisible variables—wind gradients, altitude, and humidity—that drastically alter ball flight. While elite caddies help pros navigate these, most golfers are left to guess. More critically, sports psychology has identified the "Quiet Eye" (QE) phenomenon as the ultimate differentiator: elite golfers hold a steady gaze for 2.5 to 3 seconds before a putt, while amateurs typically fixate for less than one [5]. Research proves that training this specific attentional skill can shave 2 strokes per round [6], yet no consumer product currently exists to help golfers visualize and master this "pro" advantage in real-time.

Our project aims to change this. Working in conjunction with Stanford University, with Meta as our project partner, we present GolfVision: an Augmented Reality (AR) heads-up display (HUD) integrated with an AI caddie that provides an edge to golfers by allowing them to see what the elite golfers experience, with real-time, environment-adjusted playing yardage calculator and live focus feedback using a Pupils Lab Bare Metal module for measurements of quiet eye.

Interviews were conducted with four golfers of varying skill levels and one coach. These sessions were guided by an outcome table comprising specific job steps and outcome statements synthesised from online sources [7][8].

Interviews were conducted with a cohort of five golfers to identify the specific performance outcomes they aimed to achieve through GolfVision. We presented a comprehensive list of outcome statements and asked respondents to rank them based on their perceived importance and current level of dissatisfaction with existing tools.

Format of outcome statement: Direction + Unit of Measure + outcome desired [9]

To systematically identify golfers' needs, we employed the Jobs-to-Be-Done (JTBD) framework, specifically the Job mapping technique outlined by Ulwick [9].

Through structured interviews with golfers of varying levels and online sources [7][8], we mapped the core Job steps of a golfer into three phases: Think, Feel, and Play. Each reflects the cognitive, emotional, and physical stages through which a golfer progresses for each shot.

Think Box: This phase covers the pre-shot decision-making process, from assessing the lie of the ball through to visualising the intended shot and managing emotions before addressing the ball. This phase accounts for half of the job steps, indicating that the majority of a golfer's cognitive workload occurs before the club is ever swung. This aligns with existing literature that identifies pre-shot routines and decision-making as critical determinants of shot quality [10].

Feel Box: This phase captures the transition from planning to execution, in which the golfer feels the intended swing, settles into a stance, fixates on the ball, and commits to the shot. This is where the quiet eye phenomenon is most active, as the golfer's final gaze fixation on the ball directly precedes motor execution.

Play Box: This phase encompasses post-shot actions, observing the ball's trajectory, reflecting on the shot, and locating the ball. Although shorter, this phase is significant because it feeds back into the Think phase for the next shot, creating a learning loop.

Using the 5 Whys technique, we drilled down into the underlying reasons for performance inconsistency, specifically focusing on the 'Think Phase' and 'Feel Phase' identified through the Job mapping in Section 1.3.

To prioritise which outcomes to address, each outcome statement was scored using the Outcome-Driven Innovation (ODI) opportunity algorithm as defined by Ulwick [9]: Opportunity Score = Importance + MAX(Importance − Satisfaction, 0).

Focus-driven Execution improvement

Course Intelligence

Considering the complicated nature of golf, we split the 5 whys into 2 different branches - the think phase and the feel phase.

The execution of both 5 Whys branches successfully converged on two fundamental, unaddressed root causes that define the scope of our proposed solution.

1. Course Intelligence:
   1. Strategic Deficiency (Think): The inability to access integrated environmental data without disrupting the physical routine.
2. Psychological Deficiency (Feel):
   1. The absence of objective indicators to validate and train the Quiet Eye phenomenon during active play.

This literature review serves to validate the core problems and root causes established in the preceding primary research analysis, highlighting three overlapping challenges that set our product apart from competitors.

Inaccurate playing yardage estimation, crucial for cutting strokes, is complicated by environmental factors (wind, temperature, humidity, altitude) that most golfers ignore or vaguely estimate. While a common coaching approach suggests a simple wind-distance percentage [11], this method overlooks individual trajectory and spin sensitivity, as well as the asymmetrical effect of wind direction [12]. Research confirms that low confidence in yardage calculations leads to club selection hesitation and poor shot execution [13], highlighting the need for real-time, integrated environmental data.

Access to professional guidance is expensive and scarce. Only 9% of U.S. golf courses offer caddie services, and only 5% of golfers use them [14], costing an average of $113 USD per bag—a recurring barrier for recreational players. Coaching sessions are similarly costly and time-constrained. This market gap confirms the salient unmet need for affordable, personalized feedback during actual play.

Lastly, the mental dimension of attentional control and focus is underserved by current tools. Traditional verbal coaching to "focus on the ball" is generalized and open-loop, offering no specific, quantifiable feedback. Research validates the Quiet Eye (QE) phenomenon as a reliable performance predictor [6], showing that QE training improves putting accuracy under pressure while reducing anxiety [15]. Despite evidence that QE training can shave 1.9 putts per round [6], no accessible product provides real-time QE feedback to golfers during play.

The current market is fragmented across single-function devices that can be mapped along two axes: legacy versus emerging, and direct versus indirect competition. Legacy indirect competitors such as laser rangefinders (e.g. Bushnell Pro X3+) and GPS watches (e.g. Garmin Approach S70) address only raw distance, while legacy direct competitors like shot-tracking tags (e.g. Arccos Smart Sensors) and swing analysers (e.g. HackMotion) offer post-round or practice-only analytics with no real-time on-course utility. Emerging indirect competitors such as AI coaching apps (e.g. SWEE AI) provide personalised instruction but remain screen-bound and disconnected from live course conditions. The closest emerging direct competitors, such as CaddieVision and RANGEZ, consolidate distance data into AR glasses but still lack environmental sensing, eye-tracking, and AI coaching. PuttView X delivers AR green-reading via HoloLens but is restricted to putting practice. Across all four quadrants, a golfer seeking comprehensive support must purchase, carry, charge, and context-switch between multiple disconnected devices — and even then, no combination adjusts yardage for real-time environmental conditions, measures attentional state, or delivers strategic advice mid-round. Our product collapses this entire fragmented ecosystem into a single wearable, occupying a white space that no existing competitor — legacy or emerging — currently addresses.

Recreational golfers face performance inconsistency because existing tools fail to provide two critical needs: integrated, real-time environmental data for accurate playing yardage, and objective, gaze-based feedback on attentional focus (Quiet Eye). This fragmentation and lack of a hands-free, heads-up solution disrupts the pre-shot routine, preventing amateurs from mastering the mental and environmental complexities of the game.

Our project develops GolfVision to solve performance inconsistency among recreational golfers by fusing two previously fragmented needs into a hands-free platform: delivering real-time, environment-adjusted playing yardage (Course Intelligence) and providing objective, gaze-based feedback on attentional control (Quiet Eye).

Based on the synthesis of primary research, the 5 Whys analysis, and the market gaps validated by the secondary research and competitor analysis, we now formalize our solution's positioning. To validate that our products effectively address the root causes of the Think and Feel phases, we applied the Value Proposition Canvas framework. This methodology maps the customer profile's jobs, pains, and gains (right side) against our product’s value map (left side), which details the pain relievers and gain creators designed to deliver our distinct competitive advantage.

The scope of this project is confined to the 'Think Box' and 'Feel Box' phases of the golfer's workflow, as defined in the primary research. Specifically, the technical development focuses on delivering 'Course Intelligence' within the Think Box and enhancing 'Attentional Control' within the Feel Box via gaze-based Quiet Eye feedback. While the 'Play Box' (post-shot analysis and learning loops) remains a critical component of the overall golfer journey, it is formally deferred to future iterations of the platform. Adhering to the collaborative mandate with Meta, the implementation is strictly constrained to the smart glasses form factor. Additionally, for the hardware prototyping, we will utilize commercial off-the-shelf technology to enable GolfVision’s functionalities, while relying on Meta’s specialized engineering expertise for the final industrial integration and hardware refinement.

From the primary interview with the golfers, we are able to generate the user needs and system requirements.

Based on the system requirements mentioned in Section 1.9, we formed three major subsystems to enable our GolfVision: Hardware Subsystem, Application Subsystem and Backend Software Subsystem. The Hardware Subsystem will focus on collecting the gaze information from the user and relaying the information to the user. The Application Subsystem will focus on the user interface. Finally, the Backend Software Subsystem will focus on creating an AI caddie that will integrate the environmental data and gaze data to provide relevant insights that the golfers can use to improve their putting experience.

The first major design decision for the Hardware Subsystem was selecting the mode of feedback: Visual or Audio. We chose a visual Heads-Up Display (HUD) approach based on the comparative analysis below.

The visual HUD was selected as the optimal relay method based on human factors principles of cognitive workload and attentional focus. Unlike audio feedback, which is transient and requires high levels of auditory processing that can conflict with a golfer's internal dialogue, the HUD integrates complex, multi-modal data—including shifting environmental variables and millisecond-level Quiet Eye metrics—directly into the user's natural line of sight. This design alignment fulfills UR3 by minimizing the cognitive load required to access and mentally integrate disparate data points. By placing actionable intelligence within the primary visual field, the system ensures that golfers can verify their attentional readiness and course conditions without disrupting the flow of their critical pre-shot routine or breaking gaze-action coupling.

Following the selection of a visual design for information relay, the next step was to select the specific hardware components for the display and the eye tracking system. The goal was to ensure a high sampling rate for accurate gaze measurement and a lightweight, bright display suitable for outdoor use.

Comparative Analysis of Hardware

The table below compares the leading options for the two critical components: the Eye Tracking Sensor and the Visual Display.

Eye Tracker Module: The Pupil Labs Bare Metal was selected. It provides the necessary high sampling rate (200Hz vs. HoloLens 2's 90Hz and Tobii's 60Hz), superior accuracy (1.2° vs. 1.5° for HoloLens 2), and crucial full access to raw eye-tracking data, which is essential for integration with our AI coach. The Tobii Pro Spark was dismissed due to its requirement for off-head placement (45cm away), making it unsuitable for mobile, outdoor golf use.

Visual Display: The Xreal Air 2 Ultra was chosen. Key selection criteria included outdoor functionality, spatial mapping, AR capabilities, and comfort. The Xreal Air 2 Ultra features a built-in screen, electrochromic lenses for direct sunlight visibility, and a lightweight design (83g vs. HoloLens 2's 566g). This lightweight form factor is critical for comfort over extended periods, as stated in PR7. The HoloLens 2 and transparent OLED options were rejected due to brightness issues in direct sunlight, the difficulty of integrating the Pupil Labs Bare Metal with the HoloLens 2's shape, and the HoloLens 2's weight.

Integrating the Pupil Labs Bare Metal eye-tracking module with the Xreal Air 2 Ultra display was a core engineering challenge necessary to create a cohesive prototype for real-time gaze tracking and HUD. This required overcoming several physical and optical constraints to ensure the system remained functional and comfortable. The integration effort focused on solving three key challenges:

1. Accurate Positioning: Ensuring the Bare Metal was positioned at the correct eye level and angle to capture precise eye movement.
2. Secure Attachment: Developing a snug attachment mechanism to prevent movement and sustain fast head movements during a golf swing.
3. Display Visibility: Maintaining the user’s clear view of the Xreal glasses’ augmented reality display.

The iterative design process is presented in Section 3.1.

A high-level architecture diagram of our system is as shown. We were required to do hardware and software integrations to achieve the operational and performance requirements as set out by our design specifications.

Drawing from the design specifications and conceptual frameworks established in Section 2, the project proceeded to the development of a functional prototype. This phase was characterized by an iterative design approach across the hardware, application, and backend software layers to ensure system integration and performance.

The hardware integration phase focused on securing the Pupil Labs sensor to the Xreal glasses while maintaining precise optical alignment and user comfort. Through five distinct design iterations, the mounting mechanism was refined to achieve high-accuracy gaze tracking (PR3) and an ergonomic, lightweight form factor (OR5, PR7) suitable for active on-course play.

The application layer serves as the critical interface between hardware telemetry and user visualization. Developed using Unity, this layer enables the rendering of AR features and real-time data overlays (OR1).

The backend software architecture provides the computational logic required to fulfill operational requirements, divided into five specialized modules.

The backend software layer can be split into 5 main modules.

1. Remote control module
2. Eye tracking module
3. AI caddie module
4. Weather data module
5. Yardage calculation module

This module facilitated manual data override for testing. Using a "Wizard of Oz" methodology, the team could simulate automated responses to validate the user experience prior to full algorithm automation.

The Eye Tracking Module processes high-frequency gaze data to provide actionable feedback on attentional stability, specifically targeting the Quiet Eye (QE) duration.

The following iterations fulfilled OR2, OR3, PR3, PR4 and PR5 of our requirements.

The AI Caddie Module utilizes Large Language Models (LLMs) to provide contextualized recommendations and coaching based on environmental data and historical player performance.

The AI caddie module fulfilled OR7 and PR10 as part of our design specifications.

The Weather Data Module automates the retrieval of environmental variables, fulfilling OR1 and PR1 to inform tactical decision-making.

This module calculates "Playing Yardage" by applying ball-flight physics to raw distance, factoring in wind, temperature, humidity, and altitude to eliminate mental estimation errors (OR1, PR2).

We verified our prototype by mapping it to the list of operational requirements and performance requirements in our design specifications, in line with the verification and validation framework [4]. A checklist of operational requirements and performance requirements was used to verify if the system built met the requirements that were set out.

Two golfers participated in the on-course evaluation: a sponsored professional athlete with an official handicap of -2 and an amateur-level player with an approximate handicap of 20. This pairing was intentional, as it allowed the prototype to be assessed across distinctly different player profiles and skill levels.

To simulate a seamless, hands-free user experience, a pouch was used to store the phones on the golfers' persons, keeping the hardware out of sight during play. Additionally, the AI interaction was Wizard-of-Oz's - the golfer triggered the system by saying "Hey Meta," replicating the voice-activation flow of the Meta Ray-Ban glasses to ensure the testing experience closely mirrored the intended final product interaction.

Testing was conducted on Holes 3, 6, and 8 of the Mandai Executive Golf Course. These holes were specifically selected because they are all Par 3 holes, meaning the golfer's objective on each tee shot is to reach the green directly. This eliminates the variable of course management strategy. There is no need for the golfer to decide between laying up, shaping a shot around a dogleg, or targeting an intermediate position. Instead, the target is fixed (the pin), and the distance from white tee to green is known. This made Par 3 holes ideal for evaluating the system's core functionality since all key variables such as distance, target, and shot intent are clearly defined from the outset.

Out of 17 requirements, 12 were fully met, 3 were partially met, and 2 were not met. The two unmet requirements, PR6 (auto-HUD removal) and PR8 (IP67 rating), were bound by time and hardware constraints rather than fundamental technical limitations. Importantly, every requirement tied to our primary value proposition of real-time Quiet Eye feedback and AI-powered coaching was fully met during the on-course evaluation.

To validate the QET component of GolfVision, testing was conducted across two phases: a controlled putting experiment (Phase 1) and an on-course system evaluation at Mandai Executive Golf Course (Phase 2).

Phase 1: Controlled Experimental Validation

Controlled 8ft putting trials established that QET intervention produced statistically meaningful improvements in proximal performance metrics. Precision and distance control improved by 10-15cm. This divergence validated the necessity of GolfVision as a continuous, on-course feedback architecture rather than a discrete coaching intervention, directly addressing the requirement for real-time gaze-action coupling stabilization.

As seen in the figure below the introduction of QET helped the golfer to putt nearer to the hole, which is optimal as golfers usually aim to get the ball as close to the hole as possible in order to finish it off on the second putt.

Phase 2: On-Course System Integration (Mandai)

On-course deployment confirmed the system's ability to extract and process high-frequency QE data via the Pupil Labs module within the required 5-second latency threshold (PR5). Gaze patterns aligned with skill-differentiated benchmarks (2.0-3.5s for the professional; ~1.5s for the amateur), with the amateur exhibiting an adaptive upward trend in fixation duration following real-time feedback. This serves as a functional proof-of-concept for the primary value proposition: transforming an internal cognitive state into a quantifiable, actionable HUD metric to improve execution consistency during live play.

During the on-course evaluation across the three selected Par 3 holes, the system dynamically sourced localized environmental parameters via the OpenMeteo API to compute an adjusted "Playing Yardage," factoring in wind velocity, direction relative to the golfer, temperature, and humidity. Baseline raw distances were obtained using the hardcoded values of the holes from the mandai golf data sheet. Crucially, the final computed Adjusted Playing Yardage values across all three holes closely approximated the professional golfer's experienced mental assessment, providing strong initial validation for the algorithm’s real-world accuracy.

While the system weight met the 200g target (PR7), testers reported the assembly felt front-heavy due to the forward-mounted sensor and 3D-printed bracket. This created pressure points and fatigue over time. Future designs should redistribute mass, potentially by moving companion electronics to the rear of the head strap.

Testers noted depth perception issues during putting, where objects appeared closer than their actual distance. This results from the Xreal display's fixed focal plane conflicting with the 1–2 meter convergence required for putting—a conflict less noticeable at longer distances. Future designs must mitigate this optical constraint by dynamically adjusting the virtual image plane or disabling the HUD during the putting routine.

Both testers preferred a wireless system, as the current tethered setup felt bulky and restricted movement (OR5 partially met). While the USB-C connection was necessary for high-bandwidth raw data streaming from the Pupil Labs sensor, moving to a wireless solution introduces challenges in balancing power consumption with the low latency required for real-time feedback (PR5). Future development will focus on integrated, power-efficient wireless modules.

During the Mandai evaluation, the AI Caddie's wind-adjusted club recommendation directly caused the professional golfer to change his club selection on at least one hole — a moment captured on video. This demonstrates that the system's output was actionable: an experienced professional trusted the AI's recommendation over his initial instinct, indicating that the fused data (wind, playing yardage, and personal club profiles) provided value beyond what even a trained golfer could mentally compute.

A limitation was observed in context sensitivity. The professional golfer expressed interest in receiving Quiet Eye feedback specifically for putting, but considered it less relevant for full shots. The current system does not distinguish between shot types when delivering gaze-based coaching — it provides the same feedback for putts, approach shots, and tee shots. This lack of shot-context awareness reduces personalisation and represents a gap relative to the nuanced, situation-dependent advice a human caddie would provide.

The professional golfer confirmed that all three displayed weather variables — temperature, humidity, and wind speed — were relevant to his decision-making, validating the design decision in Section 3.3.4 to limit the weather display to these three metrics.

The pro golfer noted that having real time wind data with respect to his gaze direction was very helpful in helping him decide how to take his shot.

A practical usability finding emerged regarding units of measurement. The professional expressed a preference for mixed units depending on context: wind speed in miles per hour, distance in metres for approach shots, but feet for close-range putting (e.g., "a 10-foot putt" versus "a 20-metre approach"). The current system uses fixed units, requiring the golfer to mentally convert — a minor but unnecessary cognitive overhead that works against the intent of UR3.

Testers also noted that while they appreciated the current display layout and information hierarchy, the overall UI required further visual refinement before it would be suitable for a broader user cohort.

Both testers provided unsolicited feedback on features beyond the current prototype scope. The most requested feature was a minimap-style hazard overlay — described by the testers as a "video game-like map" - displaying straight-line distances to bunker edges (front and back) and green boundaries. Testers emphasised that understanding the size of their landing zone relative to surrounding hazards was critical to course management, and that an in-HUD overlay would eliminate the need for separate course maps.

Additional requests included green scanning to overlay the intended shot line and ball tracking post-impact. While outside the current project scope, this feedback validates the broader market demand identified in Section 1.1.2 and informs the Stanford development roadmap (Section 6.2).

Sections 4.2 and 5.1 verified that individual components met their specifications and analysed their on-course performance. This section maps those findings back to the four User Requirements (UR1–UR4) from Section 1.9, assessing whether the integrated system addresses the golfer's actual needs as identified through the JTBD framework.

Three of four user requirements (UR1, UR2, UR4) were validated. UR3 was partially validated - notably, both fully unmet performance requirements (PR6, PR8) fall under this user need, confirming that minimising cognitive load remains the primary development gap. The validated requirements correspond to the system's core value proposition (course intelligence and gaze coaching), while UR3 relates to ergonomic and interaction design - suggesting the concept is sound but the form factor needs further iteration.

The development of GolfVision posed significant engineering challenges, requiring the seamless integration of high-frequency eye-tracking hardware with real-time AR visualization and AI-driven analytics. This project demonstrates a high-value application for smart glass technology, particularly as Meta integrates eye-tracking sensors into future consumer hardware. By providing golfers with previously invisible environmental data and elite-level gaze training, GolfVision transforms a standard accessory into a powerful performance-enhancing tool.

Software Compatibility and Stability: The most significant challenge encountered was the shifting software ecosystem for the Xreal Air 2 Ultra. During development, Xreal transitioned from the Nebula OS to AndroidXR, deprecating support for legacy Unity integrations. This transition required sourcing third-party middleware to run the custom Unity application, which introduced periodic system glitches and limited the ability to leverage native features like high-fidelity hand tracking, ultimately impacting the fluidity of the user interface.

Hardware Complexity and Portability: The current prototype requires a fragmented ecosystem of five distinct devices to function—including two smartphones, a laptop for backend processing, the eye-tracking module, and the AR glasses. This multi-device dependency creates significant friction for on-course play, increasing bulk and reducing the portability expected of a consumer-grade sports accessory. A streamlined, wireless architecture that integrates processing into a single companion device is necessary for market viability.

Optical Convergence and Depth Perception: Testing revealed a persistent depth perception error caused by the prism optics of the AR display. Golfers reported difficulty gauging the exact spatial relationship between the club head and the ball during the putting stroke, as the virtual overlay's fixed focal plane competed with the close-range physical target. This optical conflict necessitates frequent recalibration and suggests that the design of the AR glasses must maintain the natural visual focus of the golfer.

While the current prototype demonstrates core functionality, several advanced features remain for future development in collaboration with Stanford University. These include hazard detection, stress management, and course management strategy—capabilities intended to transition the system from Par 3 evaluation to comprehensive full-course decision support. This evolution aims to provide a robust digital alternative to traditional caddie services by addressing a broader spectrum of on-course challenges. Due to the limited duration of initial testing in Singapore, a subsequent phase of rigorous user validation is scheduled in the United States. This phase will involve a larger cohort of athletes to gather diverse technical feedback and refine the system for global market requirements.

Based on hardware ergonomics feedback (Section 5.1.3), future work will also focus on redistributing the mass of the glasses to address the front-heavy feel. Additionally, we will explore ways to dynamically disable the lens or move the HUD out of the way to mitigate the optical convergence error observed during putting.

Other sports application

GolfVision represents a pioneering application of real-time gaze-action coupling training in a mobile sporting environment. The underlying architecture—fusing high-frequency eye-tracking data with contextual environment analytics—possesses significant transferability to any target-oriented sport where attentional control is a primary performance determinant. In disciplines such as hockey, archery, and soccer, the system can provide coaches with a precise, objective quantification of an athlete’s visual search strategy and fixation stability. By identifying suboptimal gaze patterns during high-pressure execution, the platform enables targeted interventions to enhance team dynamics and individual precision.

To validate the transferability, testing was done with field hockey goalkeepers from the Singapore Women’s National Hockey Team as well as a Premier League goalkeeper. The same Pupil Labs eye tracker used on GolfVision was deployed and we found that elite goal keepers have a longer fixation duration and more structured gaze patterns compared to non-elite ones. A follow-up interview with the Singapore Women’s goalkeeping coach confirmed that eye tracking data would be highly valuable for coaching as it will allow coaches to quantify goalkeeper attentional focus which is currently assessed through subjective observation. These findings suggest that the sensing and feedback architecture we developed for golf can be applied on sports such as hockey as well.

Beyond the athletic arena, the technology’s capacity to track and guide visual focus offers transformative potential for high-stakes professional training. Empirical research has already demonstrated that Quiet Eye training significantly improves the performance of junior surgeons under pressure, particularly in precision tasks like surgical knot-tying [23]. By adapting this system to educational contexts, we can accelerate the motor-skill acquisition phase for complex manual tasks, providing learners with real-time feedback on where and when to focus their visual attention to mirror expert-level performance.

The integration of advanced biometric tracking within a lightweight AR form factor positions GolfVision at the forefront of the global sports technology market, currently valued at approximately USD $93.81 billion [24]. As hardware manufacturers like Meta continue to embed eye-tracking sensors into consumer-grade smart glasses, the modular software framework developed in this project can be scaled across various industrial sectors to enhance productivity, safety, and operational precision.

[1] “2025 Golf Courses and Country Clubs Industry Market Research Report,” Kentley Insights. [Online]. Available: https://www.kentleyinsights.com/golf-courses-and-country-clubs-industry-market-research-report/. [Accessed: 03-Apr-2026].

[2] MordorIntel, “Golf Equipment Market Size, Share & Industry Report, 2031,” Mordor Intelligence. [Online]. Available: https://www.mordorintelligence.com/industry-reports/golf-equipment-market. [Accessed: 03-Apr-2026].

[3] “Golf GPS Watches Market Size & Industry Growth 2030,” FutureDataStats. [Online]. Available: https://www.futuredatastats.com/golf-gps-watches-market. [Accessed: 03-Apr-2026].

[4] R. Camacho, “Requirements Traceability Matrix Templates & Examples,” Parasoft, Apr. 9, 2020. [Online]. Available: https://www.parasoft.com/blog/requirements-management-and-the-traceability-matrix/. [Accessed: 31-Mar-2026].

[5] J. N. Vickers, “Neuroscience of the Quiet Eye in Golf Putting,” International Journal of Golf Science, vol. 1, no. 1, pp. 2–19, 2012.

[6] S. J. Vine, L. J. Moore, and M. R. Wilson, “Quiet eye training facilitates competitive putting performance in elite golfers,” Frontiers in Psychology, vol. 2, article 8, 2011.

[7] Rick Shiels Golf, “THE PERFECT GOLF PRE SHOT ROUTINE = More Consistency,” YouTube, Feb. 22, 2021. [Online]. Available: https://www.youtube.com/watch?v=kdR6aF2CxtQ. [Accessed: 20-Mar-2026].

[8] The Scoring Method, “Proven Pre-Shot Routine to BREAK 80!,” YouTube, Apr. 4, 2024. [Online]. Available: https://www.youtube.com/watch?v=aNQBTKTZMes. [Accessed: 20-Mar-2026].

[9] A. W. Ulwick, Jobs to Be Done: Theory to Practice. Houston, TX: Idea Bite Press, 2016. [Online]. Available: https://jobs-to-be-done-book.com/thank-you/.

[10] J. N. Vickers, Perception, Cognition, and Decision Training: The Quiet Eye in Action. Champaign, IL: Human Kinetics, 2007.

[11] Golf Monthly, “How to Calculate Distance in the Wind,” GolfMonthly.com, 2025.

[12] S. Malik et al., “Effects of wind on ball flight,” as cited in Golf Science Journal, 2018.

[13] “Trust in Distance Measuring Devices (DMDs) Automation in Golf,” International Journal of Golf Science, 2025.

[14] “Don’t Underestimate the Demand for Caddies,” Golf Digest, 2025.

[15] “The effect of quiet eye training on golf putting performance in pressure situation,” Scientific Reports, vol. 14, article 55716, 2024.

[16] “Neon - Technical Specifications - Neon eye tracking module and frames,” Pupil-labs.com. [Online]. Available: https://pupil-labs.com/products/neon/specs. [Accessed: 17-Mar-2026].

[17] MSBryan, “HoloLens 2 hardware,” Microsoft.com. [Online]. Available: https://learn.microsoft.com/en-us/hololens/hololens2-hardware. [Accessed: 17-Mar-2026].

[18] MSBryan, “Eye tracking overview - Mixed Reality,” Microsoft.com. [Online]. Available: https://learn.microsoft.com/en-us/windows/mixed-reality/design/eye-tracking. [Accessed: 17-Mar-2026].

[19] Tobii.com. [Online]. Available: https://go.tobii.com/tobii-pro-spark-brochure. [Accessed: 01-Apr-2026].

[20] “Blue 1.51" Transparent OLED Display 128x56 Controller SSD1309 SPI,” Buydisplay.com. [Online]. Available: https://www.buydisplay.com/blue-1-51-inch-transparent-oled-display-128x56-controller-ssd1309-spi. [Accessed: 02-Apr-2026].

[21] “XREAL - Building Augmented Reality for Everyone,” XREAL Ltd. [Online]. Available: https://www.xreal.com/. [Accessed: 17-Mar-2026].

[22] K. Dalton, “The Vision Strategy of Golf Putting,” Aston Publications Explorer (Aston University), Jan. 1, 2013. [Online]. Available: https://doi.org/10.48780/publications.aston.ac.uk.00019543. [Accessed: 10-Jan-2026].

[23] J. Causer, J. N. Vickers, R. Snelgrove, G. Arsenault, and A. Harvey, “Performing under pressure: quiet eye training improves surgical knot-tying performance,” Surgery, vol. 156, no. 5, pp. 1089–1096, 2014.

[24] “Wearable Devices Technology in Sport Market - Companies & Trends,” Mordor Intelligence. [Online]. Available: https://www.mordorintelligence.com/industry-reports/wearable-devices-in-sports-market. [Accessed: 02-Apr-2026].

Jobs to be done steps

THINK BOX

1. Figure out the lie of the ball
2. Decide the direction I want to be aiming
3. Determine the Ground Yardage
4. Find the wind Direction
5. Determine the Elevation
6. Determine the Playing Yardage
7. Decide how to play the shot
8. Decide what club to hit with
9. Decide the shot shape
10. Visualise the golf shot
11. Keeping yourself emotionally steady before addressing the ball
12. Get comfortable

FEEL BOX

13. Feel the swing, I want to swing
14. Get into a stance
15. Focus on the ball
16. Just react and pull the trigger

PLAY BOX

17. Observe the trajectory of the ball
18. Reflect on the previous shot
19. Proceed to find the ball
20. Just react and pull the trigger

Comparison of 4 Sports Vision Techniques