Benchmarks for retrospective automated driving system crash rate analysis using police-reported crash data

Area of Science:

• Road safety
• Automated driving systems (ADS)
• Vehicle safety assessment

Background:

• Automated driving systems (ADS) are expanding, necessitating robust safety assessment methods.
• Current safety evaluations often compare ADS to human-driven vehicles, requiring accurate benchmark data.
• Existing benchmark methodologies face challenges in data comparability and potential biases.

Purpose of the Study:

• To address and extend existing literature on automated driving systems (ADS) safety impact.
• To generate human crash rates using police-reported data for areas with ADS deployment.
• To provide a repeatable and transparent methodology for ADS safety benchmark determination.

Main Methods:

• Leveraging publicly accessible, police-reported crash data.
• Applying underreporting correction methodologies for improved data comparability.
• Analyzing crash rate dependencies on geographic region, road type, and vehicle type.

Main Results:

• Identified essential steps for generating reliable ADS safety benchmarks.
• Demonstrated how failing to account for crash rate dependencies can bias ADS comparisons.
• Presented analyses within the context of existing ADS benchmark literature.

Conclusions:

• Creating accurate crash rate benchmarks for ADS evaluation presents significant challenges.
• Researchers must be cautious in selecting and generating benchmark data.
• This work offers analytical guidance to foster consensus on accurate ADS benchmark estimation.