Steady delivery against the milestone schedule — now moving from proving the approach to scaling data collection.
Stages 1 and 2 are complete; we are now inside Stage 3 — designing the AI system.
Progress Reports are submitted on a fixed quarterly cadence. Two are in; the third is in hand.
Prove the whole system end-to-end with a single driver first — surfacing every hurdle while the stakes are low — then scale to the volume of data the model needs.
One driver, ~5–10 hours over a month. Get the hardware, capture, exports and workflow right end-to-end.
Scale to multiple drivers and 50–100 hours — the model-training dataset, with Phase 1's lessons built in.
A single driver ran real trips across late April to mid-May. The objective wasn't training data — it was to confirm we can capture, export and time-sync every stream the AI team specified, from four independent sources.
IntelliTrac OBD — GPS, speed, heading, ignition, odometer
Garmin Fenix 6 — heart rate, HRV, sleep & recovery
Driver-facing video for blink / yawn / head pose
Sensor Logger — 100 Hz accel / gyro, GPS, barometer
Journey & position data export cleanly as XLSX / KML / GPX, time-stamped row by row.
1 Hz heart rate + GPS on every trip; HRV and sleep pull cleanly from Connect.
OBD (AEST) and Garmin (UTC) align to the second once converted — confirmed across trips.
100 Hz IMU, GPS & baro with zero dropouts over 38 min — validated by two independent tests.
All four sources joined onto one trip clock — a single, model-ready record.
Trip Summary & Trip Data templates built and handed to the model team.
This is the real value of logistics testing — finding the problems now, with one driver, not later with ten.
The telematics vendor's export had no accel/gyro. FIX → adopted Sensor Logger for phone-side 100 Hz motion.
No mechanism to log driver-rated sleepiness (KSS). FIX → added a 10-second start/end KSS prompt to the protocol.
Overnight sleep/HRV missing on some nights. FIX → wearable compliance built into driver onboarding.
Wrong lens, portrait framing, huge re-encoded transfers. FIX → standardised lens, orientation & original-quality upload.
OBD logged a previous device owner's name. FIX → corrected identity & device labelling in the portal.
~5% battery / 38 min and ~200 MB per trip. FIX → battery & cloud-sync plan for full-shift recording.
Phase 1's fixes become Phase 2's foundations. Now being set up with Murdoch University:
Onboard drivers, capture data & KSS prompts in one place
Spanning every data area — lens, orientation, wearable compliance locked in
Ready to bring additional drivers on at pace
From 5–10 hours to a target of 50–100 hours of driving data — the bulk of the model-training dataset.
A close research partnership underpins the credibility and ethics of the work.
Reviewing and strengthening the pilot methodology across iterations.
Guiding data approval and the ethical foundations of collection.
Co-designing the participant app and data framework behind Phase 2.
Our Current State survey scored AI readiness across five constructs (1–5 scale). Every construct sits below the neutral midpoint of 3.0 — the barrier is trust and surveillance, not awareness. That's exactly why this project is built around drivers.
Building industry awareness and recruiting participants across LinkedIn. Live posts are marked Posted; the rest are scheduled to follow.
Industry fatigue survey goes live
Interactive AI & safety session
Detecting fatigue earlier
Call for pilot drivers
Building for drivers, not at them
Inside the pilot build
How driver data is protected
Industry challenge & call to collaborate
Featured to talk fatigue, human-centred design and the psychology of getting drivers home safely — bridging the gap between the lab and the cab.
Ran a session bringing the project's themes directly to an industry audience.
Submitted to present the work at upcoming industry and research conferences.
A comprehensive fatigue & AI-readiness survey informing the design.
The next step — how partners get involved in Drive S(AI)fe.