This study presents a comprehensive analysis of automated vehicle (AV) disengagements in California, leveraging traffic conditions and employing the XGBoost model and explainable AI techniques. The study investigates six distinct categories of disengagement initiation: driver-initiated, system-initiated, freeway-based, arterial-based, driverless-capable, and driverless-incapable. The findings reveal that predictors of each disengagement type differ significantly. Generally, the day of the week and vehicle miles of travel showed a great contribution to the prediction of overall AV disengagements. On the other hand, human-driven VMT was the main predictor for driver-initiated disengagement, while human-driven vehicle crashes were the highest-ranked factor for system-initiated, arterial roadways, and driverless-capable AV disengagement. For freeway-based disengagements, the year of operation was the major factor. The study contributes to the existing body of knowledge by providing a nuanced understanding of AV disengagements, emphasizing the need for targeted interventions and strategies. The research serves as a pioneering step toward unveiling hidden patterns in AV disengagements, showcasing the potential of explainable AI in unraveling complex AV disengagement scenarios, and providing a solid foundation for future advancements in AV technology and policy formulation. The findings point to further investigation of AV disengagements to understand the safe operation of AVs and better predict their maturity, thereby aiding policymakers in making informed decisions about AV testing and deployment.
Prediction of Daily Disengagements of Automated Vehicles Using Explainable Machine Learning Approach
International Conference on Transportation and Development 2024 ; 2024 ; Atlanta, Georgia
13.06.2024
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Operator vigilance using disengagements and headlight modifications
Europäisches Patentamt | 2024
|DataCite | 2023
|Estimating lane utilization for variable approach lanes using explainable machine learning
Taylor & Francis Verlag | 2023
|