Relevant Physiological Indicators for Assessing Workload in Conditionally Automated Driving, Through Three-Class Classification and Regression

Meteier, Quentin; De Salis, Emmanuel; Capallera, Marine; Widmer, Marino; Angelini, Leonardo; Abou Khaled, Omar; Sonderegger, Andreas; Mugellini, Elena · 2022 · Crossref

DOI: 10.3389/fcomp.2021.775282

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 continuously assessing driver mental workload (MWL) in conditionally automated driving scenarios. As vehicles advance toward higher automation levels, drivers may engage in non-driving-related tasks (NDRTs), potentially leading to cognitive overload that degrades takeover performance. The authors aim to identify relevant physiological indicators and develop machine learning models to classify MWL levels and regress subjective workload scores, thereby enabling non-intrusive, real-time monitoring of driver state. The experimental design involved 80 participants engaging in 1 hour of simulated conditionally automated driving. During the simulation, participants performed the N-back cognitive task 15 times for 90 seconds each, with task difficulty (low, medium, high) and modality (visual, auditory) manipulated. Physiological data, including electrodermal activity (EDA), electrocardiogram (ECG), and respiration (RESP), were collected using a Biopac system. The researchers employed a machine learning pipeline featuring repeated cross-validation and grid search across three algorithms. The primary objectives were to classify task difficulty into three classes, classify task modality, and regress participants’ post-task subjective workload ratings on a 0–20 scale. The study also analyzed the impact of sensor fusion and task performance on model accuracy. The results demonstrated that three distinct levels of mental load could be classified with an F1-score of 0.713 using skin conductance and respiration signals as inputs for a random forest classifier. The best regression model predicted subjective workload with a mean absolute error of 3.195 when utilizing all three physiological signals. Notably, model accuracy improved when participants performed well on the cognitive tasks, suggesting that task engagement is a critical factor for reliable physiological assessment. However, the classification of task modality (visual versus auditory) was unsuccessful, indicating that the measured physiological signals do not distinctly differentiate between sensory modalities. The study concludes that specific physiological indicators, particularly estimates of respiratory sinus arrhythmia, respiratory amplitude, and temporal indices of heart rate variability, are robust measures of mental workload in automated driving contexts. These findings suggest that continuous assessment of driver workload should prioritize these specific metrics. The research highlights the potential of machine learning to refine workload assessment beyond binary classifications, offering a more granular understanding of driver cognitive states. This approach supports the development of adaptive driver-vehicle interfaces that can ensure safety by detecting overload before it impacts takeover performance.

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-nvidia on 2026-08-10; verification: verified.

Topics

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

• workload measurement
• mental demand
• drowsiness detection algorithms
• stress driving
• neuro workload indices
• distraction detection algorithms

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: physiological data
• Theoretical Contribution: theory or model, computational model