Predicting Driver Errors during Automated Vehicle Takeovers

Alambeigi, Hananeh; McDonald, Anthony D.; Manser, Michael; Shipp, Eva M.; Lenneman, John; Pulver, Elizabeth; Christensen, Scott · 2023 · Transportation Research Record Journal of the Transportation Research Board

DOI: 10.1177/03611981231159122

archive: archived pipeline: cataloged verified

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

Summary

This study addresses the critical safety challenge of predicting driver errors during the transition of control from automated vehicles to human drivers. While existing literature has explored modeling driver behavior during takeovers, there is limited understanding of how to accurately predict takeover performance errors and the optimal data structures, such as window sizes and feature inclusion, for such predictions. The authors aimed to fill this gap by using supervised machine learning to classify driver errors based on granular driving performance, physiological, and glance data collected prior to a takeover event. The research utilized data from a driving simulator experiment involving 64 participants who engaged in automated driving scenarios with SAE Level 2 automation. The study focused on braking response failures, where the automated system failed to respond to a decelerating lead vehicle, necessitating a manual takeover. Data were collected at high frequencies, including driving performance metrics (e.g., speed, acceleration, steering angle), physiological indicators (heart rate, breathing rate, electrodermal activity), and eye-tracking data (glance behavior). After preprocessing and filtering, the dataset comprised 122 complete drives, with 22 labeled as errors and 100 as non-errors. Errors were defined by failures to execute necessary subtasks, such as checking mirrors before lane changes or braking. The researchers extracted 42 features for various time windows ranging from 3 to 300 seconds before the takeover time. Three machine learning algorithms—decision tree, random forest, and support vector machine with a radial basis kernel—were trained and evaluated using five-fold grouped cross-validation. The results demonstrated that the random forest algorithm achieved the best predictive performance, with an area under the receiver operating characteristic curve (AUC) of 0.72 when trained on a 3-second window preceding the takeover. This performance was statistically significantly better than random guessing. Longer window sizes, particularly those exceeding 30 seconds, yielded lower AUC values, with some performing no better than random guessing. The study also identified the 10 most important predictors for error classification, highlighting the significance of specific physiological and glance features in the immediate moments before a takeover. The findings suggest that short-term data windows, specifically 3 seconds prior to a takeover, are most effective for predicting driver errors using machine learning models. This insight is significant for the development of automated vehicle systems, as it indicates that real-time monitoring of driver state using physiological and visual attention metrics can enhance the safety of the takeover process. By integrating these predictive algorithms into highly automated systems, manufacturers could potentially provide timely assistive technologies or alerts to mitigate driver errors, thereby improving overall safety during the transition of control.

Key finding

A random forest algorithm trained on a 3-second window of pre-takeover data achieved the highest predictive performance for classifying driver errors, with an area under the receiver operating characteristic curve of 0.72.

Methodology

simulator

Sample size: 64

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 author_sweep_intake on 2026-05-28.

StageOutcomeToolModelPromptAttemptsCompleted
discover success author_sweep 2 2026-05-28
archive success canonical_url 1 2026-06-06
extract success cached 3 2026-06-10
clean success clean 1 2026-06-04
chunk success chunk 1 2026-06-04
embed success embed Qwen/Qwen3-Embedding-8B 1 2026-06-04
enrich success 1 2026-05-28
promote success 1 2026-06-04
summarize success llm qwen3.6-27b-prismaquant summ-v5 2 2026-06-10
tag success vector_similarity 15 2026-06-11
verify success 2 2026-06-10

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.

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).