Towards A Comprehensive Evaluation of Driving Impairment and Assessment Technologies

Abstract

Impaired driving is a persistent threat to traffic safety, with alcohol and cannabis frequently involved in motor vehicle crashes. This dissertation examines how alcohol, cannabis, and their combination influence driving behavior and performance in real-world settings, and for alcohol, in a controlled driving environment. These investigations focus on vehicle control, behavioral adaptations, and, in the case of alcohol, the potential value of physiological signals as early indicators of impairment. The first study analyzed over two years of naturalistic driving data from 41 participants. Trip-level substance use was self-reported and, when available, confirmed using breath or oral fluid testing. Alcohol-positive trips mostly occurred between 6:00 PM and 1:00 AM and on the weekends and showed a statistically significant reduction in highway mileage compared to baseline (-13%, p = 0.0475). Cannabis-positive trips followed a temporal distribution similar to baseline, apart from elevated activity on Fridays, and showed reduced highway mileage and a modest 3.2% reduction in speeding (F = 4.81, p = 0.0371) along with slight degradation in lateral control. Polysubstance trips occurred predominantly on weekends and exhibited the lower overall speeding proportions. Kinematic event rates increased at lower severity thresholds during impaired trips, particularly at higher speeds, suggesting either subtle destabilization or compensatory behavior. While high-severity safety-critical event rates were comparable between cannabis-positive and baseline trips, these findings challenge assumptions that cannabis-positive drivers may engage in safer driving behavior. In general, this dataset provides detailed insight into how alcohol- and cannabis-related impairment may manifest in routine driving behavior. The second study evaluated alcohol's effects on physiology and driving performance, and the feasibility of using wearable sensors to monitor impaired driving. Five participants completed standardized drives under both sober and alcohol-impaired conditions (BrAC = 0.08%) while instrumented with ECG, respiration, and EEG sensors. Alcohol consumption produced consistent changes in heart rate (+13.5 bpm, p = 0.0267), heart rate variability (−150.3 ms, p = 0.0165), respiration patterns (reduced RVT, increased variability), and EEG signals (increased frontal alpha and theta power, reduced peak alpha frequency). ECG and respiration sensors performed reliably, while EEG data quality varied and required extensive processing. Behavioral changes on the road were consistent but subtle, with only lab-based reaction time tests reaching statistical significance (+22 ms, p = 0.0132). Participants showed poor accuracy in estimating their own intoxication (20% error) and expressed ambivalence toward driving under hypothetical impaired scenarios, consistent with longstanding evidence that drivers often lack accurate insight into their own impairment and risk. Overall, this study demonstrated that wearable physiological sensors can reliably capture alcohol-induced changes in heart, respiratory, and brain activity during real-world driving, even when observable effects on driving performance are subtle or inconsistent. This dissertation advances impaired driving research by integrating large-scale naturalistic observation with controlled experimental testing. It clarifies how alcohol and cannabis influence real-world driving behavior, demonstrates that physiological signals can reveal alcohol impairment even when driving effects are subtle, and underscores the limitations of driver self-assessment. These findings support the development of intelligent monitoring systems that use objective, physiological data to improve impaired driving detection and prevention.