Investigating Robotaxi Crash Severity Using Geographical Random Forest (2505.06762v1)

Abstract: This paper quantitatively investigates the crash severity of Autonomous Vehicles (AVs) with spatially localized machine learning and macroscopic measures of the urban built environment. We address spatial heterogeneity and spatial autocorrelation, while focusing on land use patterns and human behavior. Our Geographical Random Forest (GRF) model, accompanied with a crash severity risk map of San Francisco, presents three findings that are useful for commercial operations of AVs and robotaxis. First, spatially localized machine learning performed better than regular machine learning, when predicting AV crash severity. Bias-variance tradeoff was evident as we adjust the localization weight hyperparameter. Second, land use was the most important built environment measure, compared to intersections, building footprints, public transit stops, and Points Of Interests (POIs). Third, it was predicted that city center areas with greater diversity and commercial activities were more likely to result in low-severity AV crashes, than residential neighborhoods. Residential land use may be associated with higher severity due to human behavior and less restrictive environment. This paper recommends to explicitly consider geographic locations, and to design safety measures specific to residential neighborhoods, when robotaxi operators train their AV systems.

Investigating Robotaxi Crash Severity Using Geographical Random Forest

This paper explores the crash severity of Autonomous Vehicles (AVs) utilizing a Geographical Random Forest (GRF) model, focusing on spatially localized machine learning techniques combined with macroscopic measures of the urban built environment, particularly in San Francisco. The paper emphasizes spatial heterogeneity and spatial autocorrelation, key factors often overlooked in existing literature concerning AV crash analyses.

Technical Approach and Key Findings

The authors employ a GRF model to dissect the factors influencing AV crash severity. Compared to conventional machine learning models, GRF demonstrates superior predictive capabilities by accounting for spatial patterns within data, manifesting spatial heterogeneity and autocorrelation. Specifically, a balanced localization approach—50% global and 50% local weighting within the GRF model—yielded an optimal predictive accuracy of 78%, indicating how geographical factors profoundly affect AV crash outcomes.

Feature importance analysis conducted as part of the paper indicated that land use emerged as the strongest predictor of AV crash severity outcomes, overshadowing other built environment metrics such as intersections and Points Of Interests (POIs). The paper reveals residential areas are associated with higher severity crashes, whereas diverse commercial zones correlate with lower severity incidents.

Discussion and Implications

The paper concludes with pragmatic recommendations for AV operations, particularly robotaxi firms, emphasizing the necessity of considering geographic location during AV system training processes. This allows operators to embrace spatial variations, tailoring safety protocols specific to neighborhood characteristics like high-severity residential areas. Implementing heightened safety measures in residential neighborhoods could mitigate potentially severe crash consequences. Additionally, it proposes the strategic adaptation of motion planning algorithms according to spatial heterogeneity to enhance AV system reliability.

Future Directions

While this research primarily focuses on the urban landscape of San Francisco, it opens avenues for broader cross-city analyses, which could validate these findings by examining AV crash severity across diverse urban environments. Moreover, integrating crash frequency with severity analytics could refine AV safety assessments, contributing to a comprehensive understanding of AV operational risks. Considering the burgeoning presence of AVs on urban roads, ongoing research investigating the interplay between AV technology and urban dynamics remains critical in propelling the transportation industry's evolution toward enhanced safety and efficiency.

In summary, this paper provides a nuanced exploration of AV crash severity implications within a spatially aware analytical framework, advocating for localized machine learning methodologies and emphasizing targeted safety training measures. The detailed findings offer substantial insight into how urban planning and advanced technological integrations can jointly facilitate safer autonomous vehicular operations in increasingly complex urban environments.