To analyze driving behavior of individuals under the influence of alcohol while driving in a real car. Based on the in-vehicle variables, the investigators aim at establishing algorithms capable of discriminating sober and drunk driving using machine learning.
Driving under the influence of alcohol (or "drunk driving") is one of the most significant causes of traffic accidents. Alcohol consumption impairs neurocognitive and psychomotor function and has been shown to be associated with an increased risk of driving accidents. However, autonomous driving (level 4 or 5) is likely to be broadly available only at a substantially later time point than previously thought due to increasing concerns of safety associated with this technology. Therefore, solutions bridging the upcoming time period by more rapidly and directly addressing the problem of drunk driving associated traffic incidents are urgently needed. On the supposition that driving behavior differs significantly between sober state and drunk state, the investigators assume that different driving patterns of people under alcohol influence compared to sober states can be used to generate drunk driving detection models using machine learning algorithms. In this study, driving for data collection is initially performed at a sober baseline state (no alcohol) and then after alcohol administration (with a target of 0.15 mg/l and 0.35 mg/l breath alcohol concentration).
Study Type
INTERVENTIONAL
Allocation
RANDOMIZED
Purpose
OTHER
Masking
SINGLE
Enrollment
55
Participants will drive in three different states (sober, drunk above and below the legal limit) on a designated circuit with a real car on a test track accompanied by a driving instructor. After the initial sober driving session, participants are administered pre-mixed alcoholic beverages (e.g., vodka orange). Participants are expected to achieve a target breath alcohol concentration of 0.35 mg/l (legal limit in Switzerland is 0.25 mg/l breath alcohol concentration) before the second driving session starts. Finally, the third driving session starts when the participants' breath alcohol concentration drops to 0.15 mg/l. Participants will be blinded to their alcohol levels during the study. Measurements: Heart rate, respiration rate, blood oxygen saturation, skin conductance, skin temperature, accelerometer, eye movement, radar, facial expression, audio recording, vehicle data, in-cabin gas concentration
Participants will drive three times at the same intervals as the treatment group on a designated circuit with a real car on a test track accompanied by a driving instructor. After the initial driving session, participants receive placebo beverages (e.g., orange juice with vodka flavor). Participants are fully blinded. Measurements: Heart rate, respiration rate, blood oxygen saturation, skin conductance, skin temperature, accelerometer, eye movement, radar, facial expression, audio recording, vehicle data, in-cabin gas concentration
Institut für Rechtsmedizin
Bern, Switzerland
Diagnostic accuracy of the drunk driving warning system (DRIVE) to detect states of alcohol influence while driving quantified as the Area Under the Receiver Operator Characteristics Curve (AUROC)
The machine learning model is developed and evaluated based on in-vehicle data generated in different states of alcohol intoxication. Detection performance of alcohol influence is quantified as AUROC.
Time frame: 480 minutes
Diagnostic accuracy of the drunk driving warning system using physiological data to detect states of alcohol influence quantified as the Area Under the Receiver Operator Characteristics Curve (AUROC)
The machine learning model is developed and evaluated based on physiological wearable data recorded in different states of alcohol intoxication. Detection performance of alcohol influence is quantified as AUROC.
Time frame: 480 minutes
Diagnostic accuracy of the drunk driving warning system using eye-tracking data to detect states of alcohol influence quantified as the AUROC
The machine learning model is developed and evaluated based on eye-tracking data recorded in different states of alcohol intoxication. Detection performance of alcohol influence is quantified as AUROC.
Time frame: 480 minutes
Diagnostic accuracy of the drunk driving warning system using controller area network data of the study car to detect states of alcohol influence quantified as the AUROC
The machine learning model is developed and evaluated based on controller area network data of the study car recorded in different states of alcohol intoxication. Detection performance of alcohol influence is quantified as AUROC.
Time frame: 480 minutes
Diagnostic accuracy of the drunk driving warning system using audio data to detect states of alcohol influence quantified as the AUROC
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.
The machine learning model is developed and evaluated based on audio data recorded in different states of alcohol intoxication. Detection performance of alcohol influence is quantified as AUROC.
Time frame: 480 minutes
Diagnostic accuracy of the drunk driving warning system using radar sensor data to detect states of alcohol influence quantified as the AUROC
The machine learning model is developed and evaluated based on radar sensor data recorded in different states of alcohol intoxication. Detection performance of alcohol influence is quantified as AUROC.
Time frame: 480 minutes
Diagnostic accuracy of the drunk driving warning system using gas sensor data to detect states of alcohol influence quantified as the AUROC
The machine learning model is developed and evaluated based on gas sensor data recorded in different states of alcohol intoxication. Detection performance of alcohol influence is quantified as AUROC.
Time frame: 480 minutes
Change of steering over the alcohol intoxication trajectory
Steering is recorded based on the controller area network.
Time frame: 480 minutes
Change of steer torque over the alcohol intoxication trajectory
Steer torque is recorded based on the controller area network.
Time frame: 480 minutes
Change of steer speed over the alcohol intoxication trajectory
Steer speed is recorded based on the controller area network.
Time frame: 480 minutes
Change of velocity over the alcohol intoxication trajectory
Velocity is recorded based on the controller area network.
Time frame: 480 minutes
Change of acceleration over the alcohol intoxication trajectory
Acceleration is recorded based on the controller area network.
Time frame: 480 minutes
Change of braking over the alcohol intoxication trajectory
Braking is recorded based on the controller area network.
Time frame: 480 minutes
Change of swerving over the alcohol intoxication trajectory
Swerving is recorded based on the controller area network.
Time frame: 480 minutes
Change of spinning over the alcohol intoxication trajectory
Spinning is recorded based on the controller area network.
Time frame: 480 minutes
Change of gaze position over the alcohol intoxication trajectory
Gaze position is recorded using an eye-tracker device.
Time frame: 480 minutes
Change of gaze velocity over the alcohol intoxication trajectory
Gaze velocity is recorded using an eye-tracker device.
Time frame: 480 minutes
Change of gaze acceleration over the alcohol intoxication trajectory
Gaze acceleration is recorded using an eye-tracker device.
Time frame: 480 minutes
Change of gaze regions of interest over the alcohol intoxication trajectory
Gaze regions of interest (e.g., windshield, car dashboard, etc.) are recorded using an eye-tracker device.
Time frame: 480 minutes
Change of gaze events over the alcohol intoxication trajectory
Gaze events (e.g., fixations, saccades, etc.) are recorded using an eye-tracker device.
Time frame: 480 minutes
Change of head pose over the alcohol intoxication trajectory
Head pose (position/rotation) is recorded using an eye-tracker device.
Time frame: 480 minutes
Change of heart rate over the alcohol intoxication trajectory
Heart rate is recorded using a heart rate monitoring device and wearables.
Time frame: 480 minutes
Change of heart rate variability over the alcohol intoxication trajectory
Heart rate variability is recorded using a heart rate monitoring device and wearables.
Time frame: 480 minutes
Change of electrodermal activity over the alcohol intoxication trajectory
Electrodermal activity is recorded using wearables.
Time frame: 480 minutes
Change of wrist accelerometer measurements over the alcohol intoxication trajectory
Wrist accelerometer measurements are recorded using wearables.
Time frame: 480 minutes
Change of skin temperature over the alcohol intoxication trajectory
Skin temperature is recorded using wearables.
Time frame: 480 minutes
Self-assessment of driving performance over the alcohol intoxication trajectory
Participants rate their driving performance on a 7-point Likert Scale (lower value means poorer driving performance).
Time frame: 480 minutes
Self-estimation of alcohol concentrations over the alcohol intoxication trajectory
Participants estimate their blood alcohol concentration.
Time frame: 480 minutes
Number of driving mishaps over the alcohol intoxication trajectory
Any driving mishaps, accidents and interventions by the driving instructor will be documented.
Time frame: 480 minutes
Number of Adverse Events (AEs)
Adverse Events will be recorded at each study visit.
Time frame: 3 months, from screening to close out visit for each participant
Number of Serious Adverse Events (SAEs)
Serious Adverse Events will be recorded at each study visit.
Time frame: 3 months, from screening to close out visit for each participant.