Stage 4 · AI system development under way

Smarter fatigue management. Safer roads.Better decisions.Earlier warnings.

We're building a human-centred, ethical AI system that detects and forecasts driver fatigue in heavy vehicles, so drivers get support early and fleet managers can act before incidents occur.

Get involved See how it works

Delivered by Opposite with Murdoch UniversityFunded under the NHVR Heavy Vehicle Safety InitiativeDec 2025 – Dec 2027

01 The challenge

Fatigue remains one of the biggest safety risks in heavy-vehicle transport. Most tools still react once it has already set in.

Reactive alerts

Most systems warn once fatigue has already set in, instead of helping to prevent it.

Low driver trust

Monitoring that feels intrusive gets resisted and worked around, which weakens safety.

Out of reach for smaller fleets

Many tools assume management systems that small and medium operators don't have.

Insight that goes nowhere

AI outputs rarely turn into practical decisions about rosters, routes and workload.

This project addresses those gaps by combining behavioural science, ethical AI and co-design with industry.

02 How it works

From raw signals to a clear, explainable warning.

Four kinds of evidence, collected together

A driver-facing camera, a wrist-worn wearable, motion and GPS sensing, and operational context such as shift times and breaks.

No single measure is treated as proof on its own
Each source keeps its own timestamps

One clock, fixed windows, honest gaps

Every stream is aligned to a single shared clock and cut into fixed analysis windows.

Each window carries quality flags
Missing data is recorded as missing, not filled in

An estimate with its reasons

The system returns a fatigue class, a 0–100 score and a confidence value, with the reasons behind it.

A separate forecast looks five to ten minutes ahead
The driver gets time to plan a safe stop

SIGNAL PIPELINE · CONCEPT30 S WINDOWS

Current estimateScore 38 / 100

Early fatigue · class 1

Blinks getting longer over the last six minutes
3 h 40 m of driving since the last break
Heart-rate variability below this driver's usual range

Forecast · next 5–10 minModerate fatigue likely. Plan a stop at the next safe parking bay.

Raw streams · not yet alignedExample data

03 The system

Warnings stay in the cab. Analysis happens after the trip.

01 / 03 · In the cab

Sensors that travel with the driver

A driver-facing camera, a wrist-worn wearable and motion and GPS sensing, combined with shift and break information.

Vehicle agnostic: nothing is wired into the truck
Built for fleets of any size

02 / 03 · On the device

Live alerts that don't need a connection

A reactive system detects current fatigue and alerts the driver straight away. A predictive system forecasts fatigue five to ten minutes ahead and suggests a safe action.

Both run locally with no internet connection
The driver sees confidence, not false certainty

03 / 03 · After the trip

The Fatigue Risk Advisor

Derived trip summaries and fatigue events sync to the cloud after the trip. Drivers and fleet managers review trends, reports and advice on shifts, rosters and workload.

Raw video is not uploaded
Kept separate from the live warning path

5–10min

Advance warning window for the predictive system

6sources

Kinds of evidence combined in every estimate

4levels

Fatigue classes, from alert to severe

0wires

Connections to vehicle wiring. The prototype runs on its own battery

04 Research design

Evidence before claims.

The method is designed with Murdoch University so that every result can be traced back to how it was measured and labelled.

Subjective sleepiness

A 1–5 self-rating before and after each trip. An anchor for labelling, never entered while driving.

Driver behaviour

Driver-facing video, reviewed for eye closure, blinking, yawning and head pose.

Physiological response

Heart rate and pulse intervals, compared with each driver's own baseline.

Motion and location

Motion sensing and GPS for trip progress, elapsed time and movement context.

Operational context

Shifts, breaks, sleep history, time of day and workload, from records or approved systems.

Data quality

Checks in every window for missing signals, glare, face occlusion and sensor faults.

Four levels of fatigue, and how sure we are

Each analysis window gets a fatigue class, a continuous score and a confidence value. Uncertain labels aren't treated as equally reliable training examples.

C Fatigue classF Score 0–100Q Label confidence

Read the research design

CLASS 0Alert

Normal eye behaviour, no clear signs.

CLASS 1Early fatigue

Slightly longer blinks, less facial activity.

CLASS 2Moderate fatigue

Repeated long blinks, yawning.

CLASS 3Severe fatigue

Long eye closure, head nodding. May be unsafe to continue.

05 The kit

A compact prototype that stays out of the way.

It sits on the dashboard, runs on its own battery and doesn't connect to the truck's wiring, so it works across makes and models.

Looks for visible signs of fatigue: eye closure, blink duration, yawning and head pose. Features are extracted on the device, and raw video isn't uploaded.
Heart rate and pulse intervals, read against each driver's own baseline, with signal-quality checks and automatic reconnection.
Acceleration, rotation and location give trip progress, elapsed time and movement context on a shared clock.
A compact, battery-powered computer on the dash runs the live detection and forecasting. It needs no vehicle wiring and no internet connection.
Shows only what matters while driving: the alert, the advance warning and a short safe-action prompt, by screen and/or voice.

Prototype concept · components finalised through testingBehaviour

No vehicle wiringNo internet needed for in-drive warningsRaw video not uploadedRole-based access to post-trip reports

06 Progress

Where we're up to.

The project runs from December 2025 to December 2027. The AI system design is complete and development is under way.

M1 · Dec 2025 Complete

Agreement signed

Project established under the NHVR Heavy Vehicle Safety Initiative.

M2 · Feb 2026 Complete

Planning

Governance, ethics and evaluation frameworks, and the engagement plan.

M3 · Apr 2026 Complete

Data collection begins

A logistics pilot tests the equipment, protocol and data workflow end to end.

M4 · Aug 2026 Complete

AI system design

Measurement framework, dataset structure, fatigue labels and a three-layer design.

M5 · Jan 2027 In progress

AI system development

Explainable detection and forecasting models, driver alerts and the Fatigue Risk Advisor.

M6 · Mid 2027 Next

Co-design and field testing

Dashboards co-designed with drivers and managers, then tested with partner fleets.

M7 · Late 2027 Planned

Second phase of testing

Alerts refined from feedback, and changes in fatigue risk measured.

M8 · Late 2027 Planned

Training and toolkits

Driver and manager training, and practical toolkits for smaller operators.

M9 · Dec 2027 Planned

Evaluation and sharing

Final evaluation, industry sharing of findings and the final report.

See full progress Milestone 1 summary

07 Ethics and trust

Built to support drivers.

Safety, privacy and trust are designed in from the start. The technology supports people's judgement and doesn't replace it.

01
Explainable decisions
Every alert and recommendation comes with clear reasons, so drivers and managers understand why.
02
Privacy and security safeguards
Strict access controls protect personal and operational data, with role-based access to reports.
03
Clear limits on monitoring
Defined limits on what is collected and how it's used, to protect driver dignity. The goal is support, not punishment.
04
Co-designed with drivers and managers
The people who'll use the system help design it, so it's practical and trusted.

Fatigue Risk Advisor

Recent

Tuesday night run

Weekly summary

Roster check

Why did I get an early warning at 3:42 am?

Trip summary checked · derived events only

Three things lined up in the ten minutes before the warning:

01Your blinks were getting longer over about six minutes.
02You'd been driving for 3 h 40 m since your last break.
03It was inside the early-morning window where your alertness usually drops.

Confidence Moderate

Ask a follow-up question…

Concept illustration · example data

08 Collaboration

Co-designed with industry, researchers and regulators.

Fleet operatorsDriversOppositeMurdoch UniversityNHVR Heavy Vehicle Safety InitiativeFatigue and behavioural science expertsTechnology providersSafety and regulatory bodies

Delivered by

Opposite

A consultancy specialising in human factors, organisational psychology and human-centred design. Opposite leads the project and industry engagement.

AI development

Murdoch University

The Murdoch University team leads the AI system design, model development and validation.

Funded by

National Heavy Vehicle Regulator

Funded under the NHVR's Heavy Vehicle Safety Initiative. Key insights and recommendations will be shared across the industry.

09 Get involved

Help shape the future of fatigue management.

We're looking for partners, participants and collaborators who want to help build ethical, human-centred fatigue management for real operations.

Partner with the project

Join as a research or industry partner and help shape how the system develops.

Take part with your fleet

Help us reach drivers for co-design sessions and real-world testing, which begins in January 2027.

Voluntary for every driver, never a condition of employment
No special routes or tasks
No cost to take part
Formal ethics and consent process

Access future toolkits

Get in touch

info@drivesafe.com.au

Help us build smarter fatigue management that keeps drivers safe and supports better decisions.