Comparability of Automated Vehicle Crash Databases

Noah Goodall
2023-08-02
Abstract:Advanced driving assistance systems are available on many late-model vehicles, and automated driving systems are testing on public roads. Regulators and developers continue to assess the safety of these vehicles by comparing automated vehicle crash rates to baseline, human-driven crash rates. While there are several widely-cited automated vehicle and conventional vehicle crash databases, these databases have different underlying assumptions and inclusion criteria. Crash rates among databases may be directly comparable only with significant filtering and normalization, if at all. This paper reviews current automated vehicle and baseline human-driven crash databases and evaluates their comparability. Recommendations are presented to improve their comparability, both in terms of normalization and contextualization, as well as additional data fields that can be incorporated into existing databases. These findings may assist researchers, regulators, and automated vehicle developers attempting to evaluate the safety of driving automation systems.
Computers and Society
What problem does this paper attempt to address?