Real-time Trust Prediction in Conditionally Automated Driving Using Physiological Measures

Ayoub, Jackie; Avetisian, Lilit; Yang, X. Jessie; Zhou, Feng · 2022 · OpenAlex

DOI: 10.48550/arxiv.2212.00607

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 challenge of calibrating driver trust in conditionally automated vehicles (AVs) by developing a real-time prediction model based on physiological measures. Trust calibration is critical for safe human-AV interaction, yet existing methods like self-reported assessments are impractical for real-world application because they interrupt driving tasks. The authors aim to overcome this limitation by using non-intrusive physiological data—specifically galvanic skin response (GSR), heart rate (HR), and eye-tracking metrics—to dynamically predict trust levels during driving. The research was conducted using a desktop-based driving simulator configured to mimic SAE Level 3 automation. Fifty-nine university students participated in a between-subjects design with three conditions: a control condition (true alarms only), a false alarm condition, and a miss condition (where the system failed to detect hazards). Participants performed eight takeover requests (TORs) while engaging in a non-driving related task (Tetris). Physiological data was collected via eye-tracking headsets and wearable sensors, synchronized with self-reported trust ratings provided every 25 seconds. The authors extracted 17 features from the physiological data, including HR variability, GSR peaks, and gaze fixation metrics. They trained five machine learning models—Logistic Regression, Decision Tree, Naïve Bayes, K-Nearest Neighbors, and eXtreme Gradient Boosting (XGBoost)—to classify trust as high or low based on these features. The results demonstrated that the XGBoost model outperformed the other algorithms, achieving an F1-score of 89.1% and an accuracy of 81.6% in predicting real-time trust. Statistical analysis revealed that trust was significantly lower in the miss condition compared to the control and false alarm conditions. Feature importance analysis using SHAP (SHapley Additive exPlanations) identified seven key predictors: mean maximum heart rate, mean heart rate variability, number of fixations on the center screen, mean GSR, and three tablet-related gaze metrics. Notably, higher heart rates and fewer fixations on the central driving screen were associated with higher trust, while lower heart rate variability correlated with lower trust, likely indicating increased cognitive alertness following system failures. The significance of this work lies in its demonstration that physiological signals can effectively serve as proxies for real-time trust assessment without disrupting the driver. The findings provide a foundation for designing in-vehicle trust monitoring systems that can detect over-trust or under-trust dynamically. Such systems could facilitate trust calibration, thereby improving safety and acceptance of automated vehicles. The authors suggest that future iterations of this model could be personalized for individual drivers and validated in higher-fidelity driving environments to enhance practical applicability.

Key finding

XGBoost model achieved 89.1% F1-score for real-time trust prediction using physiological measures (GSR, HR, eye-tracking). System malfunctions, particularly false alarms and missed takeover requests, dramatically influenced participants' dynamic trust in automated vehicles.

Methodology

lab_experiment

Sample size: 74

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_normalize on 2026-05-27 (2 acquisition events logged).

StageOutcomeToolModelPromptAttemptsCompleted
discover success 1 2026-05-05
archive success canonical_url 2 2026-06-02
extract success cached 3 2026-06-07
clean success 1 2026-06-01
chunk success 1 2026-06-01
embed success 1 2026-06-02
enrich success openalex 5 2026-07-02
promote success 2 2026-06-06
summarize success llm qwen3.6-27b-prismaquant summ-v5 3 2026-06-07
tag success vector_similarity 16 2026-06-11
verify success 1 2026-05-08

Summary generated by qwen3.6-27b-prismaquant on 2026-06-07; 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).