Estimating the out-of-the-loop phenomenon from visual strategies during highly automated driving

Schnebelen, Damien; Charron, Camilo; Mars, Franck · 2020 · Accident Analysis & Prevention

DOI: 10.1016/j.aap.2020.105776

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 investigates the "out-of-the-loop" (OOTL) phenomenon in highly automated driving, specifically examining whether spontaneous visual strategies can estimate a driver’s level of disengagement. The OOTL phenomenon, where drivers lack physical control and fail to monitor the driving scene, poses significant safety risks, particularly during prolonged automation periods that may induce mind wandering (MW). The research addresses two primary questions: whether specific gaze behaviors characterize OOTL states and whether these behaviors can predict a driver’s MW score, thereby enabling online estimation of supervision quality. The experiment utilized a fixed-base driving simulator with 12 participants who underwent an 18-minute automated driving session (SAE Level 3) without secondary tasks. Gaze data were recorded using an eye-tracker and analyzed across 13 areas of interest (AOIs), including road center, mirrors, dashboard, and human-machine interface screens. The researchers computed static indicators (percentage of time spent in each AOI) and dynamic indicators (transition probabilities between AOIs). To quantify disengagement, participants self-reported the proportion of time spent thinking about non-driving topics, creating a standardized "MW score." The study employed Partial Least Squares (PLS) regression to model the relationship between gaze behavior and MW scores. Training datasets were generated from varying time windows (1–15 minutes) preceding the final two minutes of driving, which served as the validation dataset. Results indicated that MW scores varied significantly among participants, with higher scores correlating with failure to handle a critical take-over request at the end of the trial. Drowsiness indicators like PERCLOS and blink rate showed no significant correlation with MW scores. The PLS regression analysis identified that a 10-minute integration window of gaze data provided the most accurate prediction of MW scores for the final two minutes, minimizing validation error. The optimal model retained 12 specific visual indicators. Key findings revealed that increased MW was associated with higher gaze transitions toward "other" areas (outside the driving scene) and the dashboard, as well as multiple gazes toward the area below the road center. Conversely, frequent transitions from the central mirror or road center to the left screen or left mirror were negatively correlated with MW, suggesting better monitoring. Notably, the traditional metric of percentage of time spent on the road center was not selected as a significant predictor in the final model. The study concludes that spontaneous gaze behavior, particularly dynamic transitions rather than static fixation durations, can effectively estimate the OOTL phenomenon and mind wandering during automated driving. By identifying specific visual strategies associated with disengagement, this approach offers a potential method for real-time driver monitoring systems. This allows for the assessment of a driver’s readiness to take over control before a critical event occurs, addressing a critical safety gap in highly automated vehicles where traditional distraction metrics may fail to capture cognitive disengagement.

Key finding

Spontaneous gaze behavior, specifically transitions away from the road center and increased fixation on peripheral or non-driving areas, significantly predicts self-reported mind-wandering and the out-of-the-loop state during automated driving.

Methodology

simulator

Sample size: 12

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

Topics

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

• situational awareness
• automation complacency bias
• eye movements scanning
• automation
• peripheral attention
• 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
• Methodological Resource: measurement protocol
• Theoretical Contribution: conceptual framework