PART ONE
Project Overview
PROJECT OVERVIEW
The Opportunity
Many organisations invest heavily in finding ways to differentiate themselves through ideas, concepts, and
strategies. One way of doing this is through encouraging more face-to-face interactions amongst workers.
Encouraging face-to-face interactions through design is often top of mind for architects yet the tools required to testquantitatively how effective these spaces, do not exist. For years, organisations invested in hot-desking and open studios to spark collaboration, yet lacked trustworthy post-occupancy tools to validate impact. Decisions were driven by anecdotes, fit-outs were hard to justify, and valuable spaces were under-used—limiting spontaneous connection and eroding confidence in workplace strategy.
The main question remained: What made a good, flexible, collaborative work environment?
ARUP | BVN | UNSW
Role: Product Design Lead
Key Responsibilities
- Developed an Algorithm to identify Face-to-face interactions
- Design Hardware for low-cost commodity hardware Beacons and receiever Stations
- Preliminary testing for proof of concept
- Experimentation designer and instigator
- Stakeholder relations and funding coordinator
- Visualisation using Python and Grasshopper
PROJECT OVERVIEW
Background
Many companies have seen great improvement from rethinking their office spaces with impromptu collaboration in mind such as Google, Pixar, Disney and many others. The engineers and the artists are no longer seated in offices across the campus but can chat over the water cooler or use a communal kitchen during the day.
Image taken from Propmodo Decoding Realestate of the Google Headquarters Office.
Observational Studies are...
awkward?
Existing research in this area has relied heavily on the manual collection of data, which made the repetition of experiments both time-consuming and laborious. Compounding this issue, GPS proved unreliable in indoor environments, limiting its usefulness for precise tracking. Furthermore, many of the available technologies required expensive equipment and significant manual effort, creating barriers to scalability and widespread adoption.
PROJECT OVERVIEW
The Mission
Our mission was to turn open-plan intent into measurable outcomes by building a low-cost, real-time
indoor positioning system that quantifies face-to-face interactions.
By tackling this gap head-on, our goal was to provide a privacy-first, evidence-based way to
test and iterate workplace design combining commodity beacons and receiver stations,
interaction-detection algorithms, and clear visualisations—to rebuild trust in design decisions,
align stakeholders, and measurably increase in-person collaboration. In partnership with BVN, ARUP,
and UNSW, we proved feasibility through a practical proof-of-concept and created a path to scale.
The Impact
Not only did the research result in two publications to the CAADRIA technology paper presented in New Zealand, but ARUP and BVN now have a streamline way of gathering qualitative data for their post construction usability studies.
01 Delivered a proof-of-concept prototype to track and analyse workplace interactions.
01 Captured over 12,000 data points during a 6-week pilot, mapping patterns of collaboration, focus time, and informal conversations.
01 Discovered that 37% of interactions occurred in unplanned breakout areas, highlighting the need for more flexible collaboration zones.
01 Identified under-utilised meeting rooms and peak times of cross-team engagement, providing clear opportunities to optimise space.
01 Equipped the design team with evidence-based insights to shape their own workplace strategy.
01 Extended impact beyond the pilot — findings are now influencing future client workplace designs, enabling smarter, human-centred environments.
PART TWO
Foundational Research
FOUNDATIONAL RESEARCH
Technology Review
Our first step was a technology review to determine whether an existing solution already met the need, or if we would have to build one using commodity hardware.
What we knew could work...
Device sends a signal to connect
Tracking device listens for request
Sniffing software measures traffic
Data records to database
Video surveillance evaluates accuracy
The combination of Wi-Fi and Bluetooth improves accuracy: Wi-Fi helps with general location, while Bluetooth can provide fine-grained proximity in smaller areas.
How the tracker would work...
Acts as the core computer of the system
Serves as the primary storage for the Raspberry Pi’s operating system (like Raspberry Pi OS)
Scans nearby Wi-Fi access points and records their signal strengths (RSSI) and identifiers (SSID/MAC)
Detects nearby Bluetooth devices or beacons (like BLE beacons).
FOUNDATIONAL RESEARCH
Assumptions
01 Those not carrying a device and multiple devices: This can cause a bias in our results as they enter the space. The video camera acts as a control to evaluate the impact of the data during our experiments.
02 Discoverable mode switched off: A person will be ‘invisible’ if:
- Their device has WiFi or Bluetooth turned off
- The device has never accessed BVN’s WiFi before
03 Environmental Complexities: In an office, signal loss is a significant factor for unreliable data. There are many things that can interfere the line of sight between emitter and receiver:
- Different building materials e.g. concrete columns Walls
- Furniture
- People
- Appliances e.g. microwaves, fridges
Queried & Discovered
When devices are discoverable, they can be queries using the software and hardware with the individuals device information captured.
However, the tracker couldn’t detect outgoing traffic, which assumes that a person is still in the area from the moment they were detected until the end of the experiment. Sniffing software was used to collect more information about every detected device and to have some assemblance of when a individual might enter or leave but it wasn’t enough.
FOUNDATIONAL RESEARCH
Exhaustive Review — Commodity Hardware Comparison
A compact technology review of common IPS approaches you can build with commodity hardware (Raspberry Pi, USB adapters, smartphones, cheap beacons, cameras, etc.).
| Method | Typical Accuracy | Typical Cost | Pros | Cons / When to Avoid |
|---|---|---|---|---|
| Wi-Fi Fingerprinting | 2–8 m | Low–Medium (no infra cost if APs exist) |
• Uses existing Wi-Fi infrastructure. • Works well in dense AP environments. • Easy to prototype with Raspberry Pi / phones. |
• Sensitive to environmental & AP changes (recalibration needed). • Fingerprint collection is labour-intensive for large spaces. |
| Wi-Fi Trilateration / RSSI | 3–15 m | Low–Medium |
• Conceptually simple (distance estimates to APs). • No additional beacons required if AP locations known. |
• RSSI↔distance mapping is noisy indoors; poor accuracy without calibration. • Requires multiple visible APs with known locations. |
| BLE Beacons (iBeacon / Eddystone) | 1–5 m (proximity) / ~1–3 m with fingerprinting | Low (\$5–\$30 per beacon) |
• Inexpensive, easy to deploy. • Good for proximity and zone detection. • Works with phones and Pi Bluetooth dongles. |
• Battery maintenance for beacons. • RSSI still noisy; careful placement required for consistent zones. |
| RFID / NFC | < 1 m (proximity) | Low (tags & readers cheap) |
• Very reliable for explicit proximity / checkpoints. • Simple to implement for access control or object tagging. |
• Not suitable for continuous tracking across a whole floor. • Requires users/devices to be near tags/readers. |
| Cellular (GSM/LTE) | 10–100s m (coarse) | Low (uses phones) |
• No extra infrastructure when using phones. • Useful as coarse fallback outdoors or mixed environments. |
• Very coarse indoors — typically not useful alone for room-level accuracy. |
| UWB (Ultra-Wideband) | 0.1–0.5 m | Medium–High (anchor kits & tags) |
• High precision and low latency. • Great for real-time tracking (assets, people). |
• Requires UWB hardware (not as ubiquitous though becoming common). • Higher cost and setup complexity relative to BLE/Wi-Fi. |
| Camera + Visual SLAM | 0.1–1 m (depending on setup) | Low–Medium (cameras + compute) |
• High spatial detail; builds maps and tracks without beacons. • Good for robotics and AR use cases. |
• Sensitive to lighting and visual occlusion. • Privacy concerns if capturing video in workspaces. |
| QR Codes / Visual Markers | < 0.5 m (when in view) | Very Low (print markers) |
• Extremely cheap and robust for known checkpoints. • No radio infrastructure; easy to implement. |
• Requires line-of-sight and active scanning. • Not suitable for continuous free-roam tracking. |
| IMU Dead-Reckoning / PDR | Meter-level; drifts over time | Low (built into phones) |
• Good for short-term tracking between fixes. • Works offline and on-device. |
• Accumulated drift requires periodic correction from RF/vision fixes. |
| Hybrid (RF + IMU + Vision) | 0.1–3 m (depends on sensors) | Medium (mix of sensors) |
• Balances strengths of each method; robust and practical. • Can achieve room/desk level accuracy with sensor fusion. |
• More complex software (sensor fusion, calibration). • Higher integration & maintenance effort. |
Conclusion
With commodity hardware, the most practical approaches are:
- Wi-Fi fingerprinting
- BLE beacons
- Camera-based markers (QR/ArUco)
- IMU sensor fusion
PART THREE
A New Approach
After extensive research into existing technologies, we decided to build a custom solution using wearable BLE beacons and Raspberry Pi receivers to track face-to-face interactions in an office environment.
A NEW APPROACH
How it works
THE PROTOTYPE
Wearable Beacons
01 Voluntary Participation: participants can easily opt in or out by wearing or removing the beacon.
02 Low setup overhead: no need for phones or apps; the beacon works independently once powered.
03 Consistent signal source each wearable broadcasts at a steady rate, giving more reliable readings than phones with variable Bluetooth behaviour.
04 any receiver (Raspberry Pi, phone, laptop with Bluetooth) can detect them, so you’re not locked into a single device ecosystem.
05 Battery efficient: BLE beacons often last weeks to months on a coin cell, reducing charging/maintenance needs.
06 Real-time-tracking: location updates can be frequent (e.g., every second) without draining user devices.
In addition to the above, I hypothesised the beacons, when worn properly could be used to determine Field of view. A very important concept when calculating Interactions
Wearable Bluetooth Beacon - A.K.A 'The Sensicorn'
A participant posing with a beacon
THE PROTOTYPE
Receivers
In development render of the Receiever and its enclosure
Assembled receiver station
.jpg)
A progress shot taken while 3D printing parts for the Receiever station hardware enclosure
EXPERIMENT 1
Determining Field of View
Aim
To determine whether beacon signal strength can be used to infer the facing direction of a participant, thereby distinguishing face-to-face orientation in indoor positioning systems.
Hypothesis
If a participant is facing directly toward a receiver base station, then the beacon’s signal strength received at that station will be at its maximum. Conversely, when the participant faces away, the participant’s body will obstruct the transmission path, resulting in a lower signal strength.
Methodology
Participants: A single participant was recruited for this pilot experiment.
Apparatus and Materials
- One wearable beacon transmitter.
- Four receiver base stations capable of measuring received signal strength (RSSI).
- Indoor room with minimal interference and controlled spacing.
- Data logging system for recording signal strength values.
Data Analysis
- Recorded RSSI values were compiled for each 15-degree increment.
- A polar plot was generated to visualize the variation in signal strength as a function of participant orientation relative to each base station.
- Peaks in RSSI values were interpreted as moments when the participant was directly
Procedure
- The participant wore a wearable beacon device fixed securely to the chest to ensure consistent orientation relative to body position.
- The participant was positioned at the center of the room.
- Four receiver base stations were placed one meter away from the participant, positioned at the cardinal points around the participant (0°, 90°, 180°, and 270°).
- The participant rotated in 15-degree increments every 10 seconds, pausing at each angle to allow signal data to stabilize.
- At each orientation, RSSI values were recorded simultaneously from all four base stations.
Apparatus
- 20 Minute experiment
- 4 x Raspberrie pies stationed at 1, 2 3 and 4 meters apart.
- 1 Estimote Beacons
- 4 cycles
- Turning 15degree increments
PART FOUR
Results
The results showed peeks when the participant was facing the beacon, and gullies as they turn away. This meant when the signal is the strongest, and with a certian distance, we can infer the participants is facing the reciever.
Conclusion
The experiment demonstrated that wearable BLE beacons can provide directional cues based on signal strength
variations as a participant rotates. Peaks in RSSI values corresponded to orientations where the participant
faced directly toward a receiver, while lower values occurred when facing away.
This suggests that with
multiple receivers and appropriate calibration, it is feasible to infer facing direction and thereby
distinguish face-to-face interactions in indoor environments. Further studies with more participants and
varied settings are needed to validate and refine this approach.
EXPERIMENT 2
Full Office Roll Out
Aim
Evaluate whether a Bluetooth Low Energy (BLE) indoor positioning system (IPS) built with commodity hardware can reliably track employee movement across an entire office floor and generate accurate journey maps for analysis.
Hypothesis
Deploying BLE beacons with overlapping coverage across the floorplan, combined with participants wearing BLE tags, will produce location data with sufficient temporal and spatial accuracy to reconstruct meaningful journey maps (e.g., common paths, dwell areas, and handoffs between zones).
Methodology
Participants:
- 300 office staff (opt-in, consented).
- Each participant assigned a wearable BLE beacon/tag labeled with an anonymous ID.
Apparatus
- Commodity BLE beacons (mains or battery powered) installed on walls/ceilings.
- Receiver network: fixed gateways (or smartphones/Raspberry Pis) to scan BLE advertisements and forward RSSI + timestamps to a central server.
- Office floorplan mapped with zones/areas of interest (desks, meeting rooms, common areas).
- Beacon layout: grid/zone placement to ensure full coverage with ~20–30% overlap; placement density derived from vendor-stated radius and in-situ RSSI tests (aim for ≥-70 dBm at zone edges).
- Time sync across gateways (NTP).
- Backend: ingestion service, database, and processing pipeline (RSSI smoothing, multi-gateway fusion, zone inference).
- Visualization: journey-map tool to render paths, dwell heatmaps, and transitions.
EXPERIMENT 2
Full Office Roll Out
Data Processing
- Clean data (remove outliers, duplicate packets, clock drifts).
- Apply smoothing (e.g., exponential moving average) and multi-gateway fusion (e.g., strongest-signal or weighted centroid).
- Infer zone transitions and approximate paths; compute dwell time per zone and transition matrices.
- Generate individual and aggregate journey maps.
Evaluation Metrics
- Coverage: % of floor area with usable signal; % time per participant with valid location.
- Accuracy: Zone-level accuracy vs. ground truth spot checks; median transition detection delay.
- Stability: Packet loss rate, gateway uptime, tag battery life.
- Utility: Ability of journey maps to reveal high-traffic corridors and bottleneck zones (qualitative review with facilities/ops).
Ethics & Privacy
- Informed consent, anonymization, opt-out at any time.
- Aggregate reporting; no individual performance tracking.
- Data retention and access controls defined prior to rollout.
Expected Outcome
Zone-level accuracy sufficient to produce clear journey maps (heatmaps and Sankey-style transitions), enabling insights into space utilization, collaboration hotspots, and congestion areas. Potential limitations include multipath interference and transient RSSI drops near dense metal/glass; these are mitigated by overlap, calibration, and filtering.
PART FIVE
Analysis
Based on the results, we were able to determine each participants approximate path throughout the experiment.
.gif)
Detected signals for Participant ‘297’ for 1 minute.
Each participant’s journey through the office was carefully analysed by estimating their location every minute, creating a series of positioning markers throughout the day. The office floor was then divided into a grid, and a shortest-path algorithm was applied to approximate the most likely route each participant followed.
EXPERIMENT 2
Results
Some of the participants paths as a progress shot during analysing the data
Conclusion
The BLE-based indoor positioning system successfully tracked employee movement across the office floor, generating accurate journey maps that revealed common paths, dwell areas, and zone transitions.
Overall, the deployment demonstrated that a commodity BLE IPS can provide valuable insights into workplace dynamics, supporting data-driven decisions for office design and operations. Future work will focus on refining accuracy, expanding participant numbers, and exploring real-time applications.
PART SIX
Determine when Face-to-face
Interactions occurred
There is no accepted baseline definition for what constitutes a face-to-face interaction; therefore, definitions need to be made concerning what a face-to-face interaction is or at least how we measure the probability that an interaction occurred.
.webp)
Algorithm outlining the variables to calculate Face-to-Face probability
Wa = The weighted probability value associated with angle.
Wd = The weight to modify probability based on distance.
m = Represents the value added to the probability if the scenario is within a meeting room.
G = (absolute obstructions e.g. walls) the weighted probability value associated with if there is an obstruction for a clear line of sight.
Wm = The weight to modify the influence this term has on the result.
PART SEVEN
Heatmap
The probability value represents a colour given to cell representing a 'state'. The higher the number of interactions in relation to the total number of interactions, the higher the percentage and the redder the colour.
Conclusion
We were able to visualise 'Hot zones' and 'Cold zones' within the data
PART EIGHT
The Office Now
We handed over the data to the primary architects on the project for redesigning the BVN office to see how they would interpret the results.
Shot taken from open floor plan design.
Window lit plant design area for mindful creative and peaceful output.
PART NINE
Next Steps
NEXT STEPS
Road to Open beta
Following the successful proof-of-concept, the next step is to refine the analysis process and develop software that can automate it. The goal is to integrate the software, hardware, and beacon designs into a scalable product that workplaces can lease when planning a redesign or seeking insights into how people move within their space.
Improve Accuracy
The accuracy of the position of the participants has a high error margin. The next step is to refine the signal gathering technique so that the position of each person is more accurate.
Training for results interpretation
Because this is a new technology, there isn’t yet a standard way to interpret the results. It may take time for people to get used to, and in some cases, require extra effort to demonstrate the value of the data—particularly to more traditional or older-generation architects.
Monitor Feedback
Provide feedback channels and actively review user critique about the product. Conduct more field testing if necessary.
Measure Impact
Evaluate key metrics such as increased productivity, collaboration, and worker wellbeing and happyness.
NEXT STEPS
Reflection
As Product Design Lead for this proof-of-concept project, I drove the end-to-end development for the prototype and its subsequent results for analysing BVN's floorplan.
Went Well
Formative user research was executed early and captured key insights from technology review
Progress over perfection This project focused on a key value which was progress over perfection. We didn't aim to have a perfect product built, we also didn't aim to have something that is super accurate as we knew we could improve these later.
Agile Methodologies: We adopted an agile approach, allowing us to iterate quickly and adapt to feedback. This flexibility was crucial given the experimental nature of the project.
Modular Design: We chose a modular design for the system and broke each step of the project into smaller components, so we could swap out with improved systems later. This was particularly important for the hardware side of the components we used as we knew changes were inevitable and the technology would get better over time.
To Improve
Data Volume: Given the experiment again, I think we should have started with a smaller number of participants for the office rollout. 300 participants prooved incredible time intensive to analyse and for a proof-of-concept, it really wasn't necessary.
Project budget: The large number of participants for the full office rollout meant we needed to provide beacons to everyone. Each beacon cost around $5 each, however many participants would loose their beacon or break them which meant our initial estimate for this part of the project, blow out of proportion
NEXT STEPS
Testimonials
 2.svg){kind=icon}
Amelia R., Lead Architect
"This project redefines how we think about interior spaces. For the first time, our designs are actively communicating with the people inside them — guiding, informing, and adapting in real time. It’s not just a building anymore, it’s an experience."
David L., Director of Design
"Integrating this level of spatial intelligence into our architecture is a leap forward. It’s as though the walls themselves have learned to listen and respond. What we’re doing here will set the precedent for every smart building we design going forward."
Sofia M., Principle Architect
"We’re only scratching the surface. Today it’s about helping people navigate and track movement; tomorrow it will inform sustainability, optimize energy use, and even personalize the environment for every individual. This is just the first chapter."