Crash Pattern Heterogeneity in Automated Vehicles Across Varying Levels of Automation

Usman, Sheikh Muhammad · 2026 · Crossref

DOI: 10.21203/rs.3.rs-8340165/v1

archive: archived pipeline: cataloged

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

**Research Question and Motivation** This study investigates crash pattern heterogeneity in Automated Vehicles (AVs) across varying levels of automation, specifically comparing Advanced Driver Assistance Systems (ADAS, Levels 1–2) and Automated Driving Systems (ADS, Levels 3–5). Motivated by the limited empirical evidence on real-world safety performance and the distinct Operational Design Domains (ODDs) in which these technologies operate, the research aims to determine how roadway, environmental, and crash-related factors influence four specific crash types: rear-end, sideswipe, frontal, and angle crashes. The study addresses a gap in the literature by providing a comparative statistical analysis using national-level data, rather than relying on limited state-specific datasets like those from California. **Methods and Data** The analysis utilizes crash reports from the National Highway Traffic Safety Administration (NHTSA) Standing General Order, covering July 2021 to June 2022. After removing duplicates and excluding records with missing key variables, the final dataset comprised 332 ADAS crashes and 147 ADS crashes. The study employed Random Parameters Multinomial Logit models, estimated via maximum simulated likelihood, to account for unobserved heterogeneity. In the ADAS model, the dry road indicator was specified as a random parameter for frontal crashes, while in the ADS model, the large vehicle collision indicator was a random parameter for rear-end crashes. The models were refined using a stepwise process, retaining variables significant at the 90% confidence level to balance statistical rigor with the limited sample size, particularly for ADS. **Findings** Descriptive statistics reveal stark contrasts in operational contexts: 70% of ADAS crashes occurred on freeways/highways, whereas 51% of ADS crashes occurred at intersections. Consequently, ADAS-equipped vehicles were predominantly involved in frontal crashes (48.2%), often at high speeds (>40 mph) and involving fixed objects or animals. In contrast, ADS-equipped vehicles were more frequently involved in angle crashes (40.1%) and rear-end crashes (27.2%), typically at low speeds (zero mph pre-crash speed in 45.6% of cases) and in intersection environments. The random parameters logit models confirmed that ADAS vehicles are more associated with frontal and sideswipe crashes under high-speed and lane-changing conditions, while ADS vehicles are more commonly involved in rear-end and angle crashes in low-speed, intersection-heavy environments. **Significance** The findings highlight that crash patterns are not uniform across automation levels but are shaped by both system capabilities and deployment contexts. The study emphasizes that due to differing reporting thresholds (ADS crashes are reported regardless of severity, while ADAS crashes require specific criteria like injury or airbag deployment) and the lack of exposure measures (vehicle miles traveled), the results should not be interpreted as a direct comparison of overall crash risk. Instead, the analysis provides critical insights for transportation agencies and AV developers regarding the specific safety challenges associated with different automation levels, such as the need for improved ADS performance in complex urban intersection interactions and ADAS reliability in high-speed highway environments.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.8-27b-gittensor on 2026-08-23; verification: pending re-verification.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• naturalistic crash near crash
• crash typology
• adas effectiveness
• incidence prevalence
• motorcycle crash typology
• pre crash contributing factors

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Empirical Findings: crash risk outcomes
• Methodological Resource: dataset resource
• Theoretical Contribution: conceptual framework