pith. sign in

arxiv: 2607.00855 · v1 · pith:4AFOUEKPnew · submitted 2026-07-01 · 📡 eess.SY · cs.SY

Investigating Driver Behavior in Complex Traffic Situations While Driving Partially Automated Vehicles

Lukas K\"oning , Nata\v{s}a Mili\v{c}i\'c , Klaus Bogenberger This is my paper

Pith reviewed 2026-07-02 07:33 UTC · model grok-4.3

classification 📡 eess.SY cs.SY
keywords driver behaviortraffic complexitypartially automated vehiclesgaze patternseye trackingbraking behaviorreal-world drivingdefensive adaptation
0
0 comments X

The pith

Statistical trends link rising traffic complexity to defensive changes in speed, braking, and gaze behavior in partially automated driving.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper examines how drivers behave in real urban traffic as complexity increases while using partially automated vehicles. It uses expert labels to rate complexity and then checks 16 metrics from vehicle data, eye tracking, and fixations. Small but consistent shifts appear: drivers deviate more from speed limits, brake more often but less intensely, spread their gaze more, and reduce guiding fixations. These patterns suggest drivers adapt defensively and change how they perceive the road. The work points to specific metrics that could help vehicles adjust their automation based on detected complexity.

Core claim

Analysis of real-world data from 20 drivers shows that as expert-labeled traffic complexity increases, drivers exhibit increased deviation from speed limits, higher brake rates with lower intensity, wider horizontal gaze dispersion and entropy, and decreased guiding fixation rates, reflecting defensive adaptation and shifts in perceptual processing.

What carries the argument

Correlation of 16 driver behavior metrics (vehicle speed deviation, brake rate and intensity, gaze dispersion, entropy, guiding fixation rate) with expert-labeled subjective traffic complexity levels.

If this is right

  • Specific metrics like driven speed, brake rate, gaze yaw entropy, and guiding fixation rate can serve as indicators for detecting traffic complexity.
  • These indicators could be integrated into complexity-adaptive partially automated vehicles to enhance safety and predictability.
  • Real-world validation supports the use of gaze metrics and entropy in complexity assessment.
  • The findings provide a foundation for combined complexity scores in vehicle systems.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Future work could test whether these same metrics predict driver workload or takeover readiness in varied road environments.
  • The entropy-based gaze measures might apply to other human-machine systems where visual attention signals environmental demand.
  • If expert labels prove consistent across sites, the metrics could support real-time vehicle adaptation without new labeling each time.

Load-bearing premise

Expert labeling of traffic complexity accurately reflects the subjective experience that drives behavioral changes in the drivers studied.

What would settle it

A replication study with a different set of experts labeling the same drives that finds no significant correlations between the labeled complexity and the 16 behavior metrics would falsify the claim.

Figures

Figures reproduced from arXiv: 2607.00855 by Klaus Bogenberger, Lukas K\"oning, Nata\v{s}a Mili\v{c}i\'c.

Figure 1
Figure 1. Figure 1: Route driven with the two ADAS setups in the real-vehicle study. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Labeling tool for expert labeling. The labeling results from all experts were preprocessed to exclude segments with vehicle speeds below 1 km/h, since the analysis focuses on driver behavior while driving. Furthermore, a ±2-second tolerance window was applied to account for variations in label-change timestamps and expert reaction times [18], [19]. The individual labels were combined using the Dawid-Skene … view at source ↗
read the original abstract

Traffic complexity critically influences driver task demands in partially automated vehicles, yet subjective perception and its behavioral indicators remain underexplored in real-world settings. This paper analyzes driver behavior - vehicle interaction, glance patterns, and guiding fixation - across varying levels of subjective traffic complexity, using real-world data from 20 drivers in real urban traffic. Traffic complexity was determined by expert labeling and served as ground truth for vehicle data. Statistical analysis of 16 driver behavior metrics revealed small but significant trends with increasing complexity: deviation from speed limit increased, brake rate increased while braking intensity decreased, horizontal gaze dispersion and entropy widened, and guiding fixation rate decreased, indicating defensive adaptation and perceptual shifts. Contributions include real-world validation of gaze metrics and guiding fixation under subjective complexity, novel insights from gaze and guiding fixation entropy metrics, and the identification of promising indicators~(driven speed, brake rate, gaze yaw entropy, guiding fixation rate) for complexity-adaptive partially automated vehicles. While based on a limited urban sample and expert-labeled subjective complexity, the findings provide a foundation for combined complexity scores and their integration into complexity-adaptive, partially automated vehicles, boosting human-like automation and enhancing safety and predictability in the traffic system.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript reports an empirical study of driver behavior in partially automated vehicles using real-world data from 20 drivers in urban traffic. Traffic complexity levels are assigned by expert labeling and treated as ground truth. Statistical analysis of 16 metrics shows small but significant trends with rising complexity: increased deviation from speed limit, increased brake rate with decreased intensity, widened horizontal gaze dispersion and entropy, and decreased guiding fixation rate. These are interpreted as defensive adaptation and perceptual shifts. The work identifies driven speed, brake rate, gaze yaw entropy, and guiding fixation rate as promising indicators for complexity-adaptive automation and notes the limited urban sample and expert labeling as limitations.

Significance. If the central correlations hold under validated complexity measures, the study supplies real-world evidence on behavioral and gaze-based indicators of traffic complexity in partially automated driving, including novel entropy metrics. Explicit credit is due for the real-world dataset, the identification of four specific candidate indicators, and the attempt to link metrics to adaptive vehicle design. The small sample and unvalidated labeling, however, constrain the strength of claims about subjective complexity and generalizability.

major comments (2)
  1. [Abstract and Methods (complexity labeling)] Abstract and Methods (complexity labeling): Expert labeling is presented as ground truth for subjective traffic complexity with no reported inter-rater reliability statistics, correlation with driver self-reports, or other validation. This is load-bearing for the central claim that observed metric trends constitute 'defensive adaptation' to subjective complexity; divergence between expert labels and driver perception would invalidate that interpretation.
  2. [Results (statistical trends)] Results (statistical trends): The claim of 'small but significant trends' across 16 metrics from 20 drivers lacks reported effect sizes, confidence intervals, per-level sample sizes, or correction for multiple comparisons. Without these, it is difficult to assess whether the trends are robust enough to support the proposed indicators for adaptive automation.
minor comments (2)
  1. [Abstract] Abstract: The 16 metrics are referenced but not enumerated; adding an explicit list or pointer to a table would improve readability.
  2. [Discussion] Discussion: The acknowledged limitation of the urban sample could be expanded with a brief statement on how findings might differ in highway or rural settings.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thorough review and constructive comments. We address each of the major comments below.

read point-by-point responses

  1. Referee: [Abstract and Methods (complexity labeling)] Abstract and Methods (complexity labeling): Expert labeling is presented as ground truth for subjective traffic complexity with no reported inter-rater reliability statistics, correlation with driver self-reports, or other validation. This is load-bearing for the central claim that observed metric trends constitute 'defensive adaptation' to subjective complexity; divergence between expert labels and driver perception would invalidate that interpretation.

    Authors: We agree that validation of the expert labeling is important. The revised manuscript will include inter-rater reliability statistics for the expert labels. Since driver self-reports were not part of the data collection, we cannot provide correlations with driver perception. We will strengthen the discussion of this as a limitation and its potential impact on the interpretation of defensive adaptation. revision: partial

  2. Referee: [Results (statistical trends)] Results (statistical trends): The claim of 'small but significant trends' across 16 metrics from 20 drivers lacks reported effect sizes, confidence intervals, per-level sample sizes, or correction for multiple comparisons. Without these, it is difficult to assess whether the trends are robust enough to support the proposed indicators for adaptive automation.

    Authors: We concur that additional statistical details are needed to better evaluate the robustness of the findings. In the revision, we will report effect sizes, confidence intervals, per-level sample sizes, and apply a multiple comparison correction with adjusted p-values. revision: yes

Circularity Check

0 steps flagged

Empirical observational study with no derivation or modeling chain

full rationale

The paper conducts statistical analysis of 16 driver behavior metrics against expert-labeled traffic complexity levels in real-world data from 20 drivers. No equations, fitted parameters presented as predictions, self-citations as load-bearing premises, or self-definitional constructs appear in the provided text or abstract. Complexity labeling is treated as an input assumption rather than derived from the metrics themselves, and the reported trends are direct statistical outputs without reduction to the inputs by construction. This is a standard empirical study whose central claims rest on data collection and analysis rather than any circular derivation.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The analysis depends on the validity of expert-labeled complexity as ground truth and the assumption that the observed statistical trends reflect true behavioral adaptations rather than noise or confounding factors in the small sample.

axioms (1)
  • domain assumption Expert labeling of traffic complexity accurately represents subjective driver perception of complexity.
    Used as ground truth for correlating with vehicle data and behavior metrics.

pith-pipeline@v0.9.1-grok · 5753 in / 1288 out tokens · 29576 ms · 2026-07-02T07:33:58.899993+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

26 extracted references · 26 canonical work pages

  1. [1]

    SAE J3016: Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles,

    SAE International, “SAE J3016: Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles,” p. 41, 2021

    work page 2021

  2. [2]

    Mental workload and driving,

    J. Paxion, E. Galy, and C. Berthelon, “Mental workload and driving,” Frontiers in Psychology, vol. 5, p. 1–11, 2014

    work page 2014

  3. [3]

    Analysis of Freeway Safety Influencing Factors on Driving Workload and Performance Based on the Gray Correlation Method,

    L. Xie, C. Wu, M. Duan, and N. Lyu, “Analysis of Freeway Safety Influencing Factors on Driving Workload and Performance Based on the Gray Correlation Method,”Journal of Advanced Transportation, vol. 2021, p. 1–11, 2021

    work page 2021

  4. [4]

    Driver’s Cognitive Workload and Driving Performance under Traffic Sign Information Exposure in Complex Environments: A Case Study of the Highways in China,

    N. Lyu, L. Xie, C. Wu, Q. Fu, and C. Deng, “Driver’s Cognitive Workload and Driving Performance under Traffic Sign Information Exposure in Complex Environments: A Case Study of the Highways in China,”International Journal of Environmental Research and Public Health, vol. 14, p. 203, 2017

    work page 2017

  5. [5]

    Familiarity and Com- plexity during a Takeover in Highly Automated Driving,

    M. S. L. Scharfe-Scherf and N. Russwinkel, “Familiarity and Com- plexity during a Takeover in Highly Automated Driving,”International Journal of Intelligent Transportation Systems Research, vol. 19, p. 525–538, 2021

    work page 2021

  6. [6]

    Research on the Quantitative Evaluation of the Traffic Environment Complexity for Unmanned Vehicles in Urban Roads,

    S. Yang, L. Gao, Y . Zhao, and X. Li, “Research on the Quantitative Evaluation of the Traffic Environment Complexity for Unmanned Vehicles in Urban Roads,”IEEE Access, vol. 9, p. 23139–23152, 2021

    work page 2021

  7. [7]

    Analyzing Gaze During Driving: Should Eye Tracking Be Used to Design Automotive Lighting Functions?

    K. Kunst, D. Hoffmann, A. Erkan, K. Lazarova, and T. Q. Khanh, “Analyzing Gaze During Driving: Should Eye Tracking Be Used to Design Automotive Lighting Functions?”J Eye Mov Res, vol. 18, p. 13, 2025

    work page 2025

  8. [8]

    Gaze- based indicators of driver cognitive distraction: Effects of different traf- fic conditions and adaptive cruise control use,

    A. Halin, A. Deli `ege, C. Devue, and M. Van Droogenbroeck, “Gaze- based indicators of driver cognitive distraction: Effects of different traf- fic conditions and adaptive cruise control use,” inAdjunct Proceedings of the 17th International Conference on Automotive User Interfaces and Interactive Vehicular Applications, 2025, pp. 151–156

    work page 2025

  9. [9]

    The characteristics of driver lane-changing behaviour in congested road environments,

    W. Wang and G. Cheng, “The characteristics of driver lane-changing behaviour in congested road environments,”Transportation Safety and Environment, vol. 6, pp. 1–15, 2023

    work page 2023

  10. [10]

    Gaze entropy metrics for mental workload estimation are heterogenous during hands- off level 2 automation,

    C. M. Goodridge, R. C. Gonc ¸alves, A. Arabian, A. Horrobin, A. Sol- ernou, Y . T. Lee, Y . M. Lee, R. Madigan, and N. Merat, “Gaze entropy metrics for mental workload estimation are heterogenous during hands- off level 2 automation,”Accident Analysis & Prevention, vol. 202, pp. 1–13, 2024

    work page 2024

  11. [11]

    Gaze Strategies in Driving–An Ecological Approach,

    O. Lappi, “Gaze Strategies in Driving–An Ecological Approach,” Frontiers in Psychology, vol. 13, pp. 1–15, 2022

    work page 2022

  12. [12]

    Drivers use active gaze to monitor waypoints during auto- mated driving,

    C. Mole, J. Pekkanen, W. E. A. Sheppard, G. Markkula, and R. M. Wilkie, “Drivers use active gaze to monitor waypoints during auto- mated driving,”Scientific Reports, vol. 11, pp. 1–18, 2021

    work page 2021

  13. [13]

    Dynamic scan paths investigations under manual and highly automated driving,

    J. Navarro, O. Lappi, F. Osiurak, E. Hernout, C. Gabaude, and E. Reynaud, “Dynamic scan paths investigations under manual and highly automated driving,”Scientific Reports, vol. 11, pp. 1–17, 2021

    work page 2021

  14. [14]

    Driver engagement and fully shared longitudinal control - insights from real partially automated vehicles,

    L. K ¨oning, P. Santos, N. Milicic, and K. Bogenberger, “Driver engagement and fully shared longitudinal control - insights from real partially automated vehicles,” 2026, techRxiv, 10.36227/techrxiv.177126589.99792150/v1

    work page doi:10.36227/techrxiv.177126589.99792150/v1 2026

  15. [15]

    Improving Driver Engagement with Level 2 Automated Systems: The Impact of Fully Shared Longitudinal Control,

    J. Illgner, N. Milicic, B. Biebl, and M. Baumann, “Improving Driver Engagement with Level 2 Automated Systems: The Impact of Fully Shared Longitudinal Control,” inProceedings of the 16th International Conference on Automotive User Interfaces and Interactive Vehicular Applications, 2024, p. 43–52

    work page 2024

  16. [16]

    Continuous-Valued Annotations Aggregation for Heart Rate Detection,

    Y . Xie, J. Li, T. Zhu, and C. Liu, “Continuous-Valued Annotations Aggregation for Heart Rate Detection,”IEEE Access, vol. 7, p. 37664–37671, 2019

    work page 2019

  17. [17]

    No Rose for MLE: Inadmissibility of MLE for Evaluation Aggregation Under Levels of Expertise,

    C. Rastogi, I. Stelmakh, N. Shah, and S. Balakrishnan, “No Rose for MLE: Inadmissibility of MLE for Evaluation Aggregation Under Levels of Expertise,” in2022 IEEE International Symposium on Information Theory (ISIT), 2022, p. 3168–3173

    work page 2022

  18. [18]

    Analysis and Compensation of the Reaction Lag of Evaluators in Continuous Emotional Annotations,

    S. Mariooryad and C. Busso, “Analysis and Compensation of the Reaction Lag of Evaluators in Continuous Emotional Annotations,” in2013 Humaine Association Conference on Affective Computing and Intelligent Interaction, 2013, p. 85–90

    work page 2013

  19. [19]

    Correcting Time-Continuous Emotional Labels by Modeling the Reaction Lag of Evaluators,

    ——, “Correcting Time-Continuous Emotional Labels by Modeling the Reaction Lag of Evaluators,”IEEE Transactions on Affective Computing, vol. 6, p. 97–108, 2015

    work page 2015

  20. [20]

    Maximum Likelihood Estimation of Observer Error-Rates Using the EM Algorithm,

    A. P. Dawid and A. M. Skene, “Maximum Likelihood Estimation of Observer Error-Rates Using the EM Algorithm,”Journal of the Royal Statistical Society. Series C (Applied Statistics), vol. 28, p. 20–28, 1979

    work page 1979

  21. [21]

    A comprehensive guide to study the agreement and reliability of multi-observer ordinal data,

    S. Vanbelle, C. H. Engelhart, and E. Blix, “A comprehensive guide to study the agreement and reliability of multi-observer ordinal data,” BMC Med Res Methodol, vol. 24, p. 310, 2024

    work page 2024

  22. [22]

    The measurement of observer agreement for categorical data,

    J. R. Landis and G. G. Koch, “The measurement of observer agreement for categorical data,”Biometrics, vol. 33, p. 159–74, 1977

    work page 1977

  23. [23]

    Ley and T

    C. Ley and T. Verdebout,Modern Directional Statistics. Boca Raton, FL, USA: Taylor & Francis, 2017, vol. 59

    work page 2017

  24. [24]

    Cohen,Statistical Power Analysis for the Behavioral Sciences

    J. Cohen,Statistical Power Analysis for the Behavioral Sciences. New York, NY , USA: Routledge, 1988, vol. 2

    work page 1988

  25. [25]

    Validity of Mental Workload Measures in a Driving Simu- lation Environment,

    F. Galante, F. Bracco, C. Chiorri, L. Pariota, L. Biggero, and G. N. Bifulco, “Validity of Mental Workload Measures in a Driving Simu- lation Environment,”Journal of Advanced Transportation, vol. 2018, no. 1, p. 1–11, 2018

    work page 2018

  26. [26]

    What can entropy metrics tell us about the characteristics of ocular fixation trajectories?

    K. Melnyk, L. Friedman, and O. V . Komogortsev, “What can entropy metrics tell us about the characteristics of ocular fixation trajectories?” PLOS ONE, vol. 19, no. 1, p. 1–45, 2024

    work page 2024