Predicting Driver Takeover Time in Conditionally Automated Driving

Ayoub, Jackie; Du, Na; Yang, X. Jessie; Zhou, Feng · 2021 · arXiv

archive: archived pipeline: cataloged verified

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

Summary

This study addresses the critical safety challenge of predicting driver takeover time during transitions from conditionally automated driving (SAE Level 3) to manual control. While previous research has identified individual factors influencing takeover time—such as takeover lead time, non-driving tasks, and alert modalities—there is a lack of computational models that simultaneously consider these variables to estimate exact takeover durations. Accurate prediction is essential for designing adaptive in-vehicle alert systems that provide timely situation awareness. To fill this gap, the authors developed an explainable machine learning framework using eXtreme Gradient Boosting (XGBoost) for prediction and SHapley Additive exPlanations (SHAP) for interpreting model outputs. The researchers utilized a dataset derived from a meta-analysis of 129 previous studies, comprising 519 takeover events. The dataset included 18 predictor variables categorized into driver characteristics (e.g., age, cognitive load), vehicle system attributes (e.g., automation level, simulator fidelity), takeover request (TOR) modalities, non-driving tasks (NDTs), and scenario urgency. The XGBoost model was trained using a 10-fold cross-validation strategy repeated 100 times to optimize hyperparameters and handle missing values. Model performance was evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Adjusted R², and correlation coefficients. SHAP was employed to rank variable importance, analyze main and interaction effects, and provide local explanations for individual predictions. The results demonstrated that the XGBoost model outperformed linear regression, linear SVM, fine tree, and random forest models, achieving an RMSE of 0.806 seconds and an Adjusted R² of 0.573. Feature selection identified seven critical predictors: urgency, time budget to collision/boundaries, age, hand occupation during NDTs, visual TOR presence, simulator fidelity, and interaction with other road users. SHAP analysis revealed that urgency was the most significant factor, with high urgency reducing takeover time by approximately 1 second. The study also uncovered complex interaction effects; for instance, visual TORs increased takeover time in low-urgency scenarios but decreased it in high-urgency scenarios. Additionally, age showed a non-linear relationship with takeover time, increasing until age 45 and decreasing thereafter. The significance of this work lies in its provision of a high-performance, interpretable model for predicting takeover time, which can inform the design of personalized, adaptive alert systems in automated vehicles. By identifying specific variable interactions and main effects, the study offers actionable insights for human-automation interaction design. For example, understanding how TOR modality interacts with urgency can help systems select optimal alert types based on scenario criticality. The findings also highlight the need for further research into older drivers’ behaviors, as the majority of existing data involves participants under 45. Overall, the integration of XGBoost and SHAP establishes a robust baseline for future predictive modeling in conditional automation safety.

Key finding

An XGBoost+SHAP model trained on a 129-study takeover-time meta-analysis identifies seven critical predictors and produces both accurate and interpretable takeover-time predictions for SAE Level 3.

Methodology

modeling

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. Discovered via discover_arxiv on 2026-05-03 (5 acquisition events logged).

Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.

Topics

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

• takeover transitions
• automation
• temporal
• situational awareness
• automation surprise

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: behavioral performance data
• Theoretical Contribution: conceptual framework, computational model