All project evidence

First commercial program · Architecture to prototype

Fleet Safety AI

Can wearable driver signals improve warning lead time at a tolerable false-alert rate?

Jump to the test plan

Published research

Independent studies · Free-to-read full text

  1. Detecting driver fatigue using heart rate variability: A systematic review

    Lu et al. · Accident Analysis & Prevention · 2022 · 10.1016/j.aap.2022.106830

    Across 18 studies, fatigue-related heart-rate variability changes were inconsistent and classification performance varied. This supports testing context and individual differences explicitly.

  2. Deriving heart rate variability indices from cardiac monitoring—An indicator of driver sleepiness

    Buendia et al. · Traffic Injury Prevention · 2019 · 10.1080/15389588.2018.1548766

    On-road Holter ECG measures from 76 drivers varied with self-rated sleepiness. The observed association leaves warning lead time and performance in new fleets open.

  3. A Dataset on Takeover during Distracted L2 Automated Driving

    Hwang et al. · Scientific Data · 2025 · 10.1038/s41597-025-04781-8 · Alternate full text

    Fifty participants completed 500 simulator scenarios with physiological, ocular and takeover measures. This is a precedent for synchronized collection in simulation.

  4. Dynamic Vibrotactile Signals for Forward Collision Avoidance Warning Systems

    Meng et al. · Human Factors · 2015 · 10.1177/0018720814542651

    Selected dynamic tactile patterns prompted faster braking than comparison cues in simulated tasks. Fleet risk prediction and crash outcomes were outside the study scope.

Technical feasibility

Wearable ECG or optical pulse sensing and vehicle telemetry can be evaluated together in a simulator. The hard questions are usable signal during driving, performance on unseen drivers, and whether the wearable adds value over camera and vehicle data.

Proposed first test

Compare a camera-and-vehicle baseline against the same baseline plus wearable signals. Keep drivers separate across training and testing. Measure warning lead time, sensitivity, false alerts per driver-hour, calibration and signal availability.

Proposed pass criterion

Advance to field feasibility if median warning lead time improves at the same false-alert rate without reducing sensitivity on held-out drivers. Road outcomes require prospective validation.

These are research targets. This brief reports independent findings and a proposed test plan for SpectraCortex.