Protocol for the REVELIO test-track pilot study: a randomised, controlled, single-centre trial in healthy recreational cannabis users investigating real-time in-vehicle detection of cannabis-impaired driving

The REVELIO protocol outlines a randomized, controlled pilot study on a closed test track designed to evaluate the feasibility of a multimodal in-vehicle system for detecting cannabis-impaired driving in healthy recreational users by correlating vehicle, driver, and biological data following controlled THC administration.

The Gist

Imagine you are trying to teach a car's "brain" how to tell if a driver is having a "bad day" behind the wheel—not because they are tired or distracted, but specifically because they have recently used cannabis.

The REVELIO study is a carefully planned experiment designed to build the "textbook" the car's brain needs to learn this skill. Here is how the study works, broken down into simple concepts:

The Big Problem: The "Black Box" of Cannabis Driving

We know that alcohol makes driving dangerous, and we have simple breathalyzers to catch it. But cannabis is trickier. It affects people differently, and unlike alcohol, there isn't a simple "one-size-fits-all" test that tells you if someone is currently too impaired to drive just by looking at their blood or breath.

Currently, police can only check for cannabis after an accident or during a stop, looking for the presence of the drug, not necessarily how well the person is actually driving at that moment. The REVELIO team wants to change that by creating a system that watches the car and the driver in real-time to spot impairment as it happens.

The Experiment: A "Driving School" for AI

Think of this study as a high-tech driving school, but instead of teaching humans how to drive, they are teaching a computer how to spot a "drugged" driver.

1. The Students (The Participants)
They are recruiting 45 healthy adults who already use cannabis recreationally (like someone who enjoys a glass of wine occasionally). They are split into two groups:

• The "Test" Group (33 people): These participants will smoke a specific, measured amount of cannabis (like a precise dose of medicine) right before they start driving.
• The "Control" Group (12 people): These participants go through the exact same day but do not smoke anything. They serve as the "baseline" or the "clean" example to compare against.

2. The Classroom (The Test Track)
To keep everyone safe, no one is driving on real public roads. They are on a closed, private test track.

• The Car: They are driving a standard 7-seater van, but it has a "co-pilot" seat with a certified driving instructor sitting right next to the driver. If the driver starts swerving or panicking, the instructor can hit the brakes instantly.
• The Schedule: Everyone starts with a "sober" drive to see how they normally drive. Then, the "Test" group smokes their joint. Afterward, everyone drives three more times over the next six hours. This lets the researchers see how driving changes as the cannabis wears off, hour by hour.

3. The Sensors (The "Eyes" and "Ears")
This is the most high-tech part. The car and the drivers are covered in sensors, acting like a giant data collection net:

• The Car's Nervous System (CAN Data): The computer records exactly how the driver touches the steering wheel, how hard they press the gas or brake, and how they stay in their lane.
• The Driver's Face (Cameras): Cameras watch the driver's eyes and head movements to see if they are looking around or zoning out.
• The Body (Wearables): The drivers wear smartwatches that track heart rate and breathing.
• The Chemistry Lab: Throughout the day, researchers take tiny samples of blood, saliva, and breath to measure exactly how much THC (the active part of cannabis) is in their system at every moment.

The Goal: Teaching the Computer

The researchers aren't just collecting data; they are feeding it into Machine Learning (a type of computer program that learns by example).

Imagine you are teaching a child to identify a "red" apple. You show them many red apples and many green apples. Eventually, the child learns the pattern.

• The Red Apples: The data from the drivers who smoked cannabis.
• The Green Apples: The data from the sober drivers.

The computer's job is to look at the steering wheel movements, eye tracking, and heart rate, and say, "Ah, this pattern looks like the 'Red Apple' (impaired) group," or "This looks like the 'Green Apple' (sober) group."

Why This Matters (According to the Paper)

The paper states that the main goal is to see if this "computer teacher" can actually learn the difference between a sober driver and an impaired one using only the data from the car and the driver's body.

• It's a Pilot: This is a "test run." The researchers know this is just the first step. They are checking if the method works and if the data is good enough to build a real system later.
• Safety First: Because they are on a closed track with a safety instructor, they can test this without risking real traffic accidents.
• Future Vision: The ultimate hope (mentioned in the paper) is to eventually build a "Fit-to-Drive" system. This wouldn't just look for cannabis; it would look for any impairment (whether from alcohol, low blood sugar, or cannabis) and warn the car if the driver isn't safe to operate the vehicle.

What the Paper Does Not Say

It is important to stick to what the paper actually claims:

• This study does not have a working product ready to sell to car companies yet.
• It does not say that cannabis-impaired driving detection is currently possible in real-world traffic.
• It does not involve any legal consequences for the participants; it is purely a research experiment.
• The participants are not being tested for police enforcement; they are helping to build the data needed to understand the problem better.

In short, the REVELIO study is like building a detailed map of a dangerous mountain pass. They aren't driving the cars over the pass yet; they are gathering the data to see if a self-driving car's computer can eventually learn to navigate that pass safely on its own.

Technical Summary

Based on the provided protocol paper, here is a detailed technical summary of the REVELIO test-track pilot study.

1. Problem Statement

Driving under the influence of cannabis is a growing global safety challenge, linked to impaired cognitive and psychomotor performance and increased accident risks. Unlike alcohol, where functional impairment correlates relatively well with blood alcohol concentration (BAC), cannabis presents unique detection challenges:

• Pharmacokinetic Complexity: The relationship between biological THC concentrations (in blood, breath, or oral fluid) and functional driving impairment is inconsistent and highly variable between individuals.
• Lack of Real-Time Systems: Current detection methods rely on retrospective roadside testing or lab analysis of biological samples, which assess substance presence rather than real-time functional impairment.
• Gap in Technology: While in-vehicle systems exist for detecting drowsiness or alcohol impairment, no scalable, real-time, in-vehicle system specifically designed to detect cannabis-impaired driving using multimodal sensor data has been systematically developed or evaluated.

2. Methodology

The REVELIO study is a randomized, controlled, open-label, single-centre pilot trial designed to generate standardized multimodal data for training machine learning models.

• Study Design & Participants:

  • Population: 45 healthy recreational cannabis users (defined as using >1x/month, CUDIT-R score ≤7) with valid driving licenses.
  • Allocation: Participants are randomly assigned to an Intervention Group ($n=33$) or a Reference Control Group ($n=12$) in a 3:1 ratio.
  • Setting: A secured, closed test track in Thun, Switzerland, using a dual-pedal 2020 Volkswagen Touran with a certified driving instructor present.

• Intervention:

  • Intervention Group: Receives a single, controlled inhalative dose of cannabis (target: 0.67 mg THC per kg body weight) via a standardized joint (15–18% THC content).
  • Control Group: Undergoes identical procedures without cannabis exposure.
  • Timeline: All participants complete a sober baseline driving session, followed by three additional standardized 50-minute driving sessions at 1–2, 3–4, and 5–6 hours post-administration.

• Data Collection (Multimodal):
  The study captures synchronized data streams to distinguish sober from impaired states:

  1. Vehicle Data: Controller Area Network (CAN) signals (steering, braking, acceleration, speed, lane position).
  2. Driver Monitoring: Driver Monitoring Camera (DMC) data (eye gaze, head movement, blink rate).
  3. Physiological Signals: Wearable sensors (heart rate, HRV, oxygen saturation, respiration).
  4. Biological Samples: Capillary blood, oral fluid, and breath samples collected at multiple time points to measure THC and metabolites.
  5. Functional Assessments: Traffic-medical/psychological tests, pupillography, and self-reported sleepiness/intoxication levels.

• Statistical & Analytical Approach:

  • Primary Goal: Develop and evaluate machine learning (ML) models to classify driving states (sober vs. cannabis-impaired).
  • Models: A baseline Logistic Regression with LASSO will be compared against flexible classifiers (Random Forests, SVM, Gradient Boosting, Bayesian Networks) and deep learning approaches (CNNs, RNNs).
  • Validation: Strict participant-level data splits to prevent information leakage; nested cross-validation for hyperparameter tuning.
  • Metrics: Primary outcome is AUROC (Area Under the Receiver Operating Characteristic Curve), supported by accuracy, sensitivity, specificity, and precision.

3. Key Contributions

• Standardized Multimodal Dataset: Creation of a high-fidelity, time-synchronized dataset linking in-vehicle sensor data, driver behavior, physiological states, and biological THC concentrations under controlled conditions.
• Methodological Framework for Cannabis Detection: Establishing a protocol to move beyond "per se" (concentration-based) laws toward functional impairment detection using real-time vehicle data.
• Foundation for "Fit-to-Drive" Models: The study is designed to be pooled with prior datasets from the same group (alcohol and hypoglycemia impairment studies) to develop a cause-independent detection model that identifies functional impairment regardless of the underlying cause.
• Safety-First Pilot: Demonstrating a rigorous safety framework (closed track, dual-control vehicle, medical screening) for conducting controlled cannabis administration studies in a driving context.

4. Results

• Note: As this is a study protocol (published as a preprint), no empirical results or performance metrics (e.g., specific AUROC values) are reported yet.
• The paper outlines the planned analysis and the rationale for the sample size (based on prior hypoglycemia and alcohol studies suggesting $n=25$ is sufficient for initial model discrimination).
• The study is currently in the recruitment and data collection phase (anticipated completion: June 2026).

5. Significance

• Road Safety Innovation: If successful, this research could lead to the development of real-time in-vehicle warning systems that detect cannabis impairment before an accident occurs, offering a proactive safety measure superior to retrospective police testing.
• Policy Implications: Provides evidence to support a shift from rigid THC concentration thresholds (which may not correlate with impairment) to functional driving capability assessments.
• Scientific Advancement: Addresses the critical gap in understanding the temporal dynamics of cannabis impairment (early, intermediate, and late phases) and how these manifest in specific driving behaviors and physiological signals.
• Scalability: The methodology is designed to be adaptable for future large-scale, real-world validation studies, potentially paving the way for automated driving restrictions or intervention systems in commercial and private vehicles.

Ethical & Regulatory Context:
The study is approved by the Cantonal Ethics Committee Bern (BASEC ID: 2025-01590) and registered at ClinicalTrials.gov (NCT07401628). It adheres to ICH-GCP and Swiss Human Research Act regulations, ensuring participant safety through strict exclusion criteria, medical monitoring, and the absence of automated legal consequences for participants during the study.