Reconstructing Sober Driving Baselines: Strategy for Evaluating Automated Driving Systems Beyond the Status Quo

Introduction: The status quo of driving risk in the US is skewed by a small minority of the population driving with alcohol intoxication. This study evaluates this premise by quantifying the distinct crash rates of both impaired and unimpaired driver cohorts, shedding light on the true safety burden of alcohol on the status quo. To evaluate Automated Driving Systems (ADS) against a normative expectation of how humans ought to drive, localized, expectations-based benchmarks are required.

Method: This study utilized exposure reconstruction across the 50 most populous U.S. urban areas, disaggregating vehicle miles traveled (VMT) and crash rates temporally (day versus night, weekday versus weekend) using INRIX telematics. By pairing U.S. Fatality Analysis Reporting System (FARS) crash outcomes with established, continuous relative risk curves, this study back-calculated alcohol impaired exposure, which was then used to estimate an unimpaired baseline rate. The reconstructed exposure was then carried downstream to scale county-level police-reported crash rates in active SAE Level 4 deployment counties.

Results: When compared to the all drivers (the status quo), Removing alcohol-involved drivers (the sober cohort; BAC = 0.00) reduced the average fatal crash involvement rate by 23% (from 1.42 to 1.10 incidents per 100 million miles), varying geographically from 134% in Minneapolis to 312% in Houston. For these sober drivers, nighttime fatal crash risk remained 3.1 to 3.9 times higher than daytime risk, which demonstrates that elevated nighttime risk is likely heavily influenced by other known risk factors such as reduced visibility, fatigue, speeding, and responding to alcohol-impaired road users. For lower-severity police-reported crashes, the sober baseline rate was 6.0% to 11.0% lower than the status quo, reflecting structurally lower relative risk multipliers than in fatal crashes.

Conclusions: Isolating and removing alcohol impairment established a localized, expectations-based safety benchmark. Using flat national averages introduces extreme spatial bias, underestimating local safety burdens in high-exposure markets. These findings demonstrate that while comparing against the status quo is essential for measuring the net safety impact of replacing active drivers, evaluating rule-compliant ADS fleets against a sober aggregate provides an expectations-based target to actively raise the roadway safety bar.

Practical Applications: Municipal planners can cheaply monitor localized impairment exposure using fatal crash proportions and continuous relative risk models. ADS developers can use these sober human baselines to establish geographically aligned, expectations-based safety targets.