pith. sign in

arxiv: 2604.10806 · v1 · submitted 2026-04-12 · 💻 cs.HC

Adaptive Bounded-Rationality Modeling of Early-Stage Takeover in Shared-Control Driving

Jian Sun , Xiyan Jiang , Xiaocong Zhao , Jie Wang , Peng Hang , Zirui Li This is my paper

Pith reviewed 2026-05-10 15:23 UTC · model grok-4.3

classification 💻 cs.HC
keywords bounded rationalitydriver modelingshared-control drivingtakeover predictionparticle filteringeye-trackingreinforcement learninghuman-vehicle interaction
0
0 comments X

The pith

An adaptive bounded-rationality model predicts hazardous driver takeovers in shared-control vehicles with higher coverage and longer lead times than static baselines.

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

This paper develops an interpretable driver model that incorporates bounded rationality by embedding cognitive constraints into reinforcement learning and then adapts latent cognitive parameters in real time from observed driver actions using particle filtering. The goal is to forecast control quality in the critical first seconds after handover from automation to human, when rapid fluctuations in cognitive state make purely rational models unreliable. In a vehicle-in-the-loop experiment with 41 participants, the adaptive version anticipated unsafe takeovers earlier and aligned its parameter estimates with simultaneous eye-tracking measures of risk perception. A sympathetic reader would care because these first seconds determine whether safety fallback systems can intervene before steering or pedal errors become accidents.

Core claim

The adaptive bounded-rationality model, which adapts latent cognitive parameters online from driver actions via particle filtering, anticipates hazardous takeovers with higher coverage and longer lead times than non-adaptive baselines and demonstrates strong alignment between the inferred parameters and real-time eye-tracking metrics, confirming that it captures genuine fluctuations in driver risk perception.

What carries the argument

The online particle-filter adaptation of latent cognitive parameters inside a bounded-rationality reinforcement-learning driver model that encodes cognitive constraints to generate predictions of control quality from observed actions.

If this is right

  • Safety fallback systems can receive earlier warnings of impending unsafe steering or pedal inputs during handover.
  • Inferred parameters provide a real-time window into fluctuations in driver risk perception.
  • Cognitively grounded assistance becomes possible because the model distinguishes between different cognitive states rather than treating all drivers identically.
  • The same adaptation mechanism can be applied to other shared-control scenarios where cognitive load changes rapidly.

Where Pith is reading between the lines

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

  • Extending the particle-filter adaptation to include additional sensor streams such as steering torque or heart-rate variability could further tighten the link between model parameters and physiology.
  • The approach suggests that vehicle safety architectures could maintain a running estimate of each driver's current bounded-rationality state rather than relying on fixed models.
  • If the parameter-to-behavior mapping holds across different vehicle interfaces, the same framework might apply to handover scenarios in aviation or remote robot operation.

Load-bearing premise

That latent cognitive parameters can be reliably inferred and adapted in real time from driver actions using particle filtering, and that these parameters correspond to actual cognitive states that affect control quality.

What would settle it

A new study in which the model's inferred cognitive parameters show no statistically significant correlation with simultaneous eye-tracking metrics or in which the adaptive model fails to outperform non-adaptive baselines on coverage and lead-time for hazardous takeover prediction.

Figures

Figures reproduced from arXiv: 2604.10806 by Jian Sun, Jie Wang, Peng Hang, Xiaocong Zhao, Xiyan Jiang, Zirui Li.

Figure 1
Figure 1. Figure 1: Research framework. (a) Collect multimodal data centered on the takeover moment for model development; (b) [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Schematic cognition-to-control coupling frame [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Reinforcement learning framework with embedded [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Takeover experiments on a high-fidelity vehicle-in-the-loop driving simulator. [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Experimental settings of the unexpected highway [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Complete experimental procedure for each participant. [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Alignment performance between parameter [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Comparison of the match rate across different sce [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Alignment performance between parameter [PITH_FULL_IMAGE:figures/full_fig_p015_9.png] view at source ↗
Figure 11
Figure 11. Figure 11: Root mean squared error (RMSE) of position and [PITH_FULL_IMAGE:figures/full_fig_p016_11.png] view at source ↗
Figure 10
Figure 10. Figure 10: Comparison of the match rate across different [PITH_FULL_IMAGE:figures/full_fig_p016_10.png] view at source ↗
Figure 12
Figure 12. Figure 12: Cognition-aware prediction model evaluation during a takeover episode. (A) Spatial trajectory showing the driver [PITH_FULL_IMAGE:figures/full_fig_p018_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Temporal evolution of inferred cognitive parameters during the takeover episode. The cognitive parameters are [PITH_FULL_IMAGE:figures/full_fig_p019_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Driver control inputs throughout the takeover [PITH_FULL_IMAGE:figures/full_fig_p019_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: This figure presents the relationship between abnormal segments of the perceptual noise parameters, [PITH_FULL_IMAGE:figures/full_fig_p020_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: This figure presents the relationship between abnormal segments of the looming aversion weighting coefficient, [PITH_FULL_IMAGE:figures/full_fig_p021_16.png] view at source ↗
read the original abstract

Human drivers' control quality in the first seconds after a handover is critical to shared-driving safety; potentially unsafe steering or pedal inputs therefore require detection and correction by the automated vehicle's safety-fallback system. Yet performance in this window is vulnerable because cognitive states fluctuate rapidly, causing purely rationality-driven, cognition-unaware models to miss early control dynamics. We present an interpretable driver model grounded in bounded rationality with online adaptation that predicts early-stage control quality. We encode boundedness by embedding cognitive constraints in reinforcement learning and adapt latent cognitive parameters in real time via particle filtering from observations of driver actions. In a vehicle-in-the-loop study (n=41), we evaluated predictive performance and physiological validity. The adaptive model not only anticipated hazardous takeovers with higher coverage and longer lead times than non-adaptive baselines but also demonstrated strong alignment between inferred cognitive parameters and real-time eye-tracking metrics. These results confirm that the model captures genuine fluctuations in driver risk perception, enabling timely and cognitively grounded assistance.

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 paper proposes an interpretable driver model for predicting early-stage control quality after handover in shared-control driving. It grounds the model in bounded rationality by embedding cognitive constraints within reinforcement learning and adapts latent cognitive parameters online via particle filtering from observed driver actions. A vehicle-in-the-loop study with n=41 participants shows the adaptive model anticipates hazardous takeovers with higher coverage and longer lead times than non-adaptive baselines while exhibiting alignment between inferred parameters and real-time eye-tracking metrics.

Significance. If the results hold, the work advances safety in autonomous driving by incorporating fluctuating cognitive states into predictive models for timely interventions. The integration of bounded-rationality RL, particle-filter adaptation, performance metrics, and independent physiological validation via eye-tracking provides both predictive utility and interpretability, with potential for cognitively grounded assistance systems.

major comments (2)
  1. [Methods (particle filtering and adaptation)] Methods (particle filtering and adaptation): The inference of latent cognitive parameters from driver actions via particle filtering, followed by use of those parameters to predict control quality, requires explicit clarification on independence to address potential circularity. The manuscript should specify whether predictions rely on held-out data, forward prediction, or predictive likelihoods separate from the fitting observations, especially since the central claim of superior anticipation rests on this pipeline.
  2. [Results (n=41 study evaluation)] Results (n=41 study evaluation): The abstract states higher coverage and longer lead times than baselines, but the manuscript must report the specific quantitative metrics, statistical tests (e.g., p-values, effect sizes), and baseline definitions to substantiate the superiority claim. Without these, the empirical support for the adaptive model's advantage remains difficult to evaluate fully.
minor comments (2)
  1. [Abstract] Abstract: Quantify the 'strong alignment' with eye-tracking metrics (e.g., report correlation coefficients or R² values) rather than describing it qualitatively, and move supporting details to the main text.
  2. Ensure consistent definition of acronyms on first use throughout (e.g., RL, AV).

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive review and the recommendation for minor revision. The comments help strengthen the clarity of our methodological pipeline and the substantiation of empirical claims. We respond point by point below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: Methods (particle filtering and adaptation): The inference of latent cognitive parameters from driver actions via particle filtering, followed by use of those parameters to predict control quality, requires explicit clarification on independence to address potential circularity. The manuscript should specify whether predictions rely on held-out data, forward prediction, or predictive likelihoods separate from the fitting observations, especially since the central claim of superior anticipation rests on this pipeline.

    Authors: We agree that an explicit statement on independence is warranted to eliminate any ambiguity regarding circularity. In the current pipeline, the particle filter recursively updates the posterior over latent cognitive parameters using only driver actions observed up to the present time step. Predictions of control quality are then generated via forward rollouts that draw from this posterior and simulate future trajectories without access to any subsequent observations. Performance metrics are computed on temporally held-out segments of each trial that were never used for the filtering updates leading to that prediction. We will insert a new subsection in Methods (with accompanying pseudocode) that details this forward-prediction structure, the separation between filtering observations and evaluation data, and the use of predictive likelihoods for the anticipation task. revision: yes

  2. Referee: Results (n=41 study evaluation): The abstract states higher coverage and longer lead times than baselines, but the manuscript must report the specific quantitative metrics, statistical tests (e.g., p-values, effect sizes), and baseline definitions to substantiate the superiority claim. Without these, the empirical support for the adaptive model's advantage remains difficult to evaluate fully.

    Authors: We accept that the abstract's qualitative phrasing should be supported by explicit quantitative reporting for complete evaluation. The Results section already contains the requested details: coverage and lead-time values for the adaptive model versus each baseline, together with the corresponding statistical tests, p-values, and effect sizes. The baselines are defined as (i) a non-adaptive bounded-rationality RL model with fixed parameters and (ii) a simple threshold-based predictor on steering/pedal variance. We will revise the manuscript to (a) state these numbers, tests, and definitions clearly in the main text with a summary table, (b) cross-reference them from the abstract, and (c) ensure the abstract itself includes the key quantitative contrasts if space permits under the venue constraints. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper's core pipeline—embedding cognitive constraints in RL, adapting latent parameters online via particle filtering on driver actions, and evaluating predictive coverage of hazardous takeovers against non-adaptive baselines—remains self-contained. Performance metrics are computed on observable takeover events and lead times, while parameter validity is cross-checked against independent eye-tracking data. No load-bearing step reduces by construction to its own inputs; the adaptation step uses action observations to infer states but the reported predictions concern subsequent control quality and physiological alignment, which are externally measurable and not tautological with the fit. The derivation therefore rests on empirical falsifiability rather than self-definition or renamed fits.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Assessment limited to abstract; no exhaustive list possible without full text. The model relies on latent cognitive parameters adapted from actions.

free parameters (1)
  • latent cognitive parameters
    Adapted online via particle filtering from driver actions; specific values or count not detailed in abstract.
axioms (1)
  • domain assumption Bounded rationality can be encoded by embedding cognitive constraints in reinforcement learning for driver decision making.
    Invoked to ground the model in the abstract description.

pith-pipeline@v0.9.0 · 5481 in / 1376 out tokens · 99622 ms · 2026-05-10T15:23:02.601409+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

60 extracted references · 60 canonical work pages · 1 internal anchor

  1. [1]

    Anderson and Pavel Zahorik

    Paul W. Anderson and Pavel Zahorik. 2014. Auditory/visual distance estimation: accuracy and variability.Frontiers in Psychology5 (2014). doi:10.3389/fpsyg.2014. 01097

    work page doi:10.3389/fpsyg.2014 2014

  2. [2]

    Boggs, Ramin Arvin, and Asad J

    Alexandra M. Boggs, Ramin Arvin, and Asad J. Khattak. 2020. Exploring the who, what, when, where, and why of automated vehicle disengagements.Accident Analysis & Prevention136 (2020), 105406. doi:10.1016/j.aap.2019.105406

    work page doi:10.1016/j.aap.2019.105406 2020

  3. [3]

    Mahdi Bonyani, Mina Rahmanian, Simindokht Jahangard, and Mahdi Rezaei. 2023. DIPNet: Driver intention prediction for a safe takeover transition in autonomous vehicles.IET Intelligent Transport Systems17, 9 (2023), 1769–1783. doi:10.1049/ itr2.12370

    work page 2023

  4. [4]

    Haolin Chen, Xiaohua Zhao, Chen Chen, Zhenlong Li, Haijian Li, and Qiuhong Wang. 2025. A systematic review on test performance of the driver takeover process in automated driving.Accident Analysis & Prevention215 (2025), 108012. doi:10.1016/j.aap.2025.108012

    work page doi:10.1016/j.aap.2025.108012 2025

  5. [5]

    Lijie Chen, Daofei Li, Tao Wang, Jun Chen, and Quan Yuan. 2025. Driver Takeover Performance Prediction Based on LSTM-BiLSTM-ATTENTION Model.Systems 13, 1 (2025), 46. doi:10.3390/systems13010046

    work page doi:10.3390/systems13010046 2025

  6. [6]

    Lihong Chen, Xiangyong Yuan, Qian Xu, Ying Wang, and Yi Jiang. 2016. Sublimi- nal Impending Collision Increases Perceived Object Size and Enhances Pupillary Light Reflex.Frontiers in Psychology7 (2016). doi:10.3389/fpsyg.2016.01897

    work page doi:10.3389/fpsyg.2016.01897 2016

  7. [7]

    Credidio, Elisângela N

    Heitor F. Credidio, Elisângela N. Teixeira, Saulo D. S. Reis, André A. Moreira, and José S. Andrade Jr. 2012. Statistical patterns of visual search for hidden objects. Scientific Reports2, 1 (2012), 920. doi:10.1038/srep00920

    work page doi:10.1038/srep00920 2012

  8. [8]

    P. R. DeLucia. 2008. Critical Roles for Distance, Task, and Motion in Space Perception: Initial Conceptual Framework and Practical Implications.Human Factors: The Journal of the Human Factors and Ergonomics Society50, 5 (2008), 811–820. doi:10.1518/001872008X312297

    work page doi:10.1518/001872008x312297 2008

  9. [9]

    Pulver, Dawn M

    Na Du, Feng Zhou, Elizabeth M. Pulver, Dawn M. Tilbury, Lionel P. Robert, Anuj K. Pradhan, and X. Jessie Yang. 2020. Predicting driver takeover performance in conditionally automated driving.Accident Analysis & Prevention148 (2020), 105748. doi:10.1016/j.aap.2020.105748

    work page doi:10.1016/j.aap.2020.105748 2020

  10. [10]

    Engstr¨ om, R

    Johan Engström, Ran Wei, Anthony D. McDonald, Alfredo Garcia, Matthew O’Kelly, and Leif Johnson. 2024. Resolving uncertainty on the fly: modeling adaptive driving behavior as active inference.Frontiers in Neurorobotics18 (2024), 1341750. doi:10.3389/fnbot.2024.1341750

    work page doi:10.3389/fnbot.2024.1341750 2024

  11. [11]

    Aldo Faisal, Luc P

    A. Aldo Faisal, Luc P. J. Selen, and Daniel M. Wolpert. 2008. Noise in the nervous system.Nature Reviews Neuroscience9, 4 (2008), 292–303. doi:10.1038/nrn2258

    work page doi:10.1038/nrn2258 2008

  12. [12]

    Niels Galley. 1989. Saccadic eye movement velocity as an indicator of (de) activation: A review and some speculations.Journal of Psychophysiology3, 3 (1989), 229–244

    work page 1989

  13. [13]

    Christian Gold, Riender Happee, and Klaus Bengler. 2018. Modeling take-over performance in level 3 conditionally automated vehicles.Accident Analysis & Prevention116 (2018), 3–13. doi:10.1016/j.aap.2017.11.009

    work page doi:10.1016/j.aap.2017.11.009 2018

  14. [14]

    Goodridge, Rafael C

    Courtney M. Goodridge, Rafael C. Gonçalves, Amélie Reher, Jonny Kuo, Michael G. Lenné, and Natasha Merat. 2025. Assessing data imbalance cor- rection methods and gaze entropy for collision prediction.PLOS One20, 11 (2025), e0336777. doi:10.1371/journal.pone.0336777

    work page doi:10.1371/journal.pone.0336777 2025

  15. [15]

    F. J. Harris. 1978. On the use of windows for harmonic analysis with the discrete Fourier transform.Proc. IEEE66, 1 (1978), 51–83

    work page 1978

  16. [16]

    Mengxia Hu, Qianru Dong, Penghui Li, and Chunjiao Dong. 2023. Cognitive Load Affects Drivers’ Brake Reaction Time to Longitudinal Conflict through Gaze Concentration and Time Headway Compensation. InCICTP 2023. American Society of Civil Engineers, 263–273. doi:10.1061/9780784485040.026

    work page doi:10.1061/9780784485040.026 2023

  17. [17]

    Heye Huang, Zheng Li, Hao Cheng, Haoran Wang, Junkai Jiang, Xiaopeng Li, and Arkady Zgonnikov. 2025. Understanding driver cognition and decision-making behaviors in high-risk scenarios: A drift diffusion perspective.Accident Analysis & Prevention220 (2025), 108123. doi:10.1016/j.aap.2025.108123

    work page doi:10.1016/j.aap.2025.108123 2025

  18. [18]

    Janssen, Martin Baumann, and Antti Oulasvirta

    Christian P. Janssen, Martin Baumann, and Antti Oulasvirta. 2024. Computational models of cognition for human-automated vehicle interaction: State-of-the-art and future directions.International Journal of Human-Computer Studies185 (2024), 103230. doi:10.1016/j.ijhcs.2024.103230

    work page doi:10.1016/j.ijhcs.2024.103230 2024

  19. [19]

    C., POUGET, A

    David C. Knill and Alexandre Pouget. 2004. The Bayesian brain: the role of uncertainty in neural coding and computation.Trends in Neurosciences27, 12 (2004), 712–719. doi:10.1016/j.tins.2004.10.007

    work page doi:10.1016/j.tins.2004.10.007 2004

  20. [20]

    Dewen Kong, Xi Zhang, Miao Wang, Lishan Sun, Yan Xu, and Qingqing Wang

    work page

  21. [21]

    Ubiquitous Traffic Eyes: trajectory dataset focus on multiple traffic states and state transition on urban expressways,

    Modeling bounded rationality in discretionary lane-changing with the cumulative prospect theory.Transportation Letters17, 10 (2025), 1785–1800. doi:10.1080/19427867.2025.2495940

    work page doi:10.1080/19427867.2025.2495940 2025

  22. [22]

    Kosinski

    Robert J. Kosinski. 2008. A Literature Review on Reaction Time.Clemson Univ. 10, 1 (2008), 337–344

    work page 2008

  23. [23]

    Krzysztof Krejtz, Andrew Duchowski, Tomasz Szmidt, Izabela Krejtz, Fernando González Perilli, Ana Pires, Anna Vilaro, and Natalia Villalobos. 2015. Gaze Transition Entropy.ACM Transactions on Applied Perception13, 1 (2015), 1–20. doi:10.1145/2834121

    work page doi:10.1145/2834121 2015

  24. [24]

    Yan Kuang, Xiaobo Qu, Jinxian Weng, and Amir Etemad-Shahidi. 2015. How Does the Driver’s Perception Reaction Time Affect the Performances of Crash Surrogate Measures?PLOS ONE10, 9 (2015), e0138617. doi:10.1371/journal.pone. 0138617

    work page doi:10.1371/journal.pone 2015

  25. [25]

    Oh-Sang Kwon, Duje Tadin, and David C. Knill. 2015. Unifying account of visual motion and position perception.Proceedings of the National Academy of Sciences 112, 26 (2015), 8142–8147. doi:10.1073/pnas.1500361112

    work page doi:10.1073/pnas.1500361112 2015

  26. [26]

    Quanyi Li, Zhenghao Peng, Lan Feng, Qihang Zhang, Zhenghai Xue, and Bolei Zhou. 2022. MetaDrive: Composing Diverse Driving Scenarios for Generalizable Reinforcement Learning.IEEE Transactions on Pattern Analysis and Machine Intelligence(2022), 1–14. doi:10.1109/TPAMI.2022.3190471

    work page doi:10.1109/tpami.2022.3190471 2022

  27. [27]

    Qingkun Li, Zhenyuan Wang, Wenjun Wang, Chao Zeng, Changxu Wu, Guofa Li, Jia-Sheng Heh, and Bo Cheng. 2023. A Human-Centered Comprehensive Measure of Take-Over Performance Based on Multiple Objective Metrics.IEEE Transactions on Intelligent Transportation Systems24, 4 (2023), 4235–4250. doi:10. 1109/TITS.2022.3233623

    work page arXiv 2023

  28. [28]

    Zheng Li, Heye Huang, Hao Cheng, Junkai Jiang, Xiaopeng Li, and Arkady Zgonnikov. 2024. Human Decision-Making in High-Risk Driving Scenarios: A Cognitive Modeling Perspective. In2024 IEEE International Automated Vehicle Validation Conference (IA VVC). IEEE, 1–8. doi:10.1109/IAVVC63304.2024.10786483

    work page doi:10.1109/iavvc63304.2024.10786483 2024

  29. [29]

    Zhi Li, Yu-Jung Ko, Aini Putkonen, Shirin Feiz, Vikas Ashok, Iv Ramakrishnan, Antti Oulasvirta, and Xiaojun Bi. 2023. Modeling Touch-based Menu Selection Performance of Blind Users via Reinforcement Learning. InProceedings of the 2023 CHI Conference on Human Factors in Computing Systems. ACM, 1–18. doi:10. 1145/3544548.3580640

    work page arXiv 2023

  30. [30]

    Griffiths

    Falk Lieder and Thomas L. Griffiths. 2020. Resource-rational analysis: Under- standing human cognition as the optimal use of limited computational resources. Behavioral and Brain Sciences43 (2020), e1. doi:10.1017/S0140525X1900061X Adaptive Bounded-Rationality Modeling of Early-Stage Takeover in Shared-Control Driving CHI ’26, April 13–17, 2026, Barcelona, Spain

    work page doi:10.1017/s0140525x1900061x 2020

  31. [31]

    Gustav Markkula, Johan Engström, Johan Lodin, Jonas Bärgman, and Trent Victor

    work page

  32. [32]

    Markkula, J

    A farewell to brake reaction times? Kinematics-dependent brake response in naturalistic rear-end emergencies.Accident Analysis & Prevention95 (2016), 209–226. doi:10.1016/j.aap.2016.07.007

    work page doi:10.1016/j.aap.2016.07.007 2016

  33. [33]

    Gustav Markkula, Yi-Shin Lin, Aravinda Ramakrishnan Srinivasan, Jac Billington, Matteo Leonetti, Amir Hossein Kalantari, Yue Yang, Yee Mun Lee, Ruth Madigan, and Natasha Merat. 2023. Explaining human interactions on the road by large- scale integration of computational psychological theory.PNAS nexus2, 6 (2023), pgad163

    work page 2023

  34. [34]

    Callum Mole, Jami Pekkanen, William Sheppard, Tyron Louw, Richard Romano, Natasha Merat, Gustav Markkula, and Richard Wilkie. 2020. Predicting takeover response to silent automated vehicle failures.PLOS ONE15, 11 (2020), e0242825. doi:10.1371/journal.pone.0242825

    work page doi:10.1371/journal.pone.0242825 2020

  35. [35]

    Hee-Seung Moon, Seungwon Do, Wonjae Kim, Jiwon Seo, Minsuk Chang, and Byungjoo Lee. 2022. Speeding up Inference with User Simulators through Policy Modulation. InCHI Conference on Human Factors in Computing Systems. ACM, 1–21. doi:10.1145/3491102.3502023

    work page doi:10.1145/3491102.3502023 2022

  36. [36]

    2025.Third Amended Standing General Order 2021-01: Incident Reporting for Automated Driving Systems (ADS) and Level 2 Advanced Driver Assistance Systems (ADAS)

    National Highway Traffic Safety Administration. 2025.Third Amended Standing General Order 2021-01: Incident Reporting for Automated Driving Systems (ADS) and Level 2 Advanced Driver Assistance Systems (ADAS). Technical Report 2021-01 (Third Amended)

    work page 2025

  37. [37]

    Erfan Pakdamanian, Shili Sheng, Sonia Baee, Seongkook Heo, Sarit Kraus, and Lu Feng. 2021. DeepTake: Prediction of Driver Takeover Behavior using Multimodal Data. InProceedings of the 2021 CHI Conference on Human Factors in Computing Systems. ACM, 1–14. doi:10.1145/3411764.3445563

    work page doi:10.1145/3411764.3445563 2021

  38. [38]

    Petzschner, Stefan Glasauer, and Klaas E

    Frederike H. Petzschner, Stefan Glasauer, and Klaas E. Stephan. 2015. A Bayesian perspective on magnitude estimation.Trends in Cognitive Sciences19, 5 (2015), 285–293. doi:10.1016/j.tics.2015.03.002

    work page doi:10.1016/j.tics.2015.03.002 2015

  39. [39]

    Victor, Marco Dozza, and Emma Tivesten

    Linda Pipkorn, Trent W. Victor, Marco Dozza, and Emma Tivesten. 2021. Driver conflict response during supervised automation: Do hands on wheel matter? Transportation Research Part F: Traffic Psychology and Behaviour76 (2021), 14–25. doi:10.1016/j.trf.2020.10.001

    work page doi:10.1016/j.trf.2020.10.001 2021

  40. [40]

    Antonin Raffin, Ashley Hill, Adam Gleave, Anssi Kanervisto, Maximilian Ernestus, and Noah Dormann. 2021. Stable-Baselines3: Reliable Reinforcement Learning Implementations.Journal of Machine Learning Research22, 268 (2021), 1–8

    work page 2021

  41. [41]

    Rendon-Velez, P

    E. Rendon-Velez, P. M . Van Leeuwen, R. Happee, I. Horváth, W.F. Van Der Vegte, and J.C.F. De Winter. 2016. The effects of time pressure on driver performance and physiological activity: A driving simulator study.Transportation Research Part F: Traffic Psychology and Behaviour41 (2016), 150–169. doi:10.1016/j.trf.2016.06.013

    work page doi:10.1016/j.trf.2016.06.013 2016

  42. [42]

    Martin Schmidt-Daffy. 2014. Prospect balancing theory: Bounded rationality of drivers’ speed choice.Accident Analysis & Prevention63 (2014), 49–64. doi:10. 1016/j.aap.2013.10.028

    work page 2014

  43. [43]

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov

    work page

  44. [44]

    Proximal Policy Optimization Algorithms

    Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347

    work page internal anchor Pith review Pith/arXiv arXiv

  45. [45]

    Wilko Schwarting, Alyssa Pierson, Javier Alonso-Mora, Sertac Karaman, and Daniela Rus. 2019. Social behavior for autonomous vehicles.Proceedings of the National Academy of Sciences116, 50 (2019), 24972–24978. doi:10.1073/pnas. 1820676116

    work page doi:10.1073/pnas 2019

  46. [46]

    Sónia Soares, António Lobo, Sara Ferreira, Liliana Cunha, and António Couto

    work page

  47. [47]

    doi:10.1186/s12544-021-00505-2

    Takeover performance evaluation using driving simulation: a systematic review and meta-analysis.European Transport Research Review13, 1 (2021), 47. doi:10.1186/s12544-021-00505-2

    work page doi:10.1186/s12544-021-00505-2 2021

  48. [48]

    Madeline Sprajcer, Drew Dawson, Anastasi Kosmadopoulos, Edward J Sach, Meagan E Crowther, Charli Sargent, and Gregory D Roach. 2023. How Tired is Too Tired to Drive? A Systematic Review Assessing the Use of Prior Sleep Duration to Detect Driving Impairment.Nature and Science of Sleep15 (2023), 175–206. doi:10.2147/NSS.S392441

    work page doi:10.2147/nss.s392441 2023

  49. [49]

    Takaaki Teshima, Masahiro Niitsuma, and Hidekazu Nishimura. 2024. Determin- ing the onset of driver’s preparatory action for take-over in automated driving using multimodal data.Expert Systems with Applications246 (2024), 123153. doi:10.1016/j.eswa.2024.123153

    work page doi:10.1016/j.eswa.2024.123153 2024

  50. [50]

    Kai Tian, Gustav Markkula, Chongfeng Wei, Yee Mun Lee, Ruth Madigan, Natasha Merat, and Richard Romano. 2022. Explaining unsafe pedestrian road crossing behaviours using a Psychophysics-based gap acceptance model.Safety Science 154 (2022), 105837. doi:10.1016/j.ssci.2022.105837

    work page doi:10.1016/j.ssci.2022.105837 2022

  51. [51]

    Velichkovsky, Alexandra Rothert, Darius Miniotas, Sascha M

    Boris M. Velichkovsky, Alexandra Rothert, Darius Miniotas, Sascha M. Dornhofer, Markus Joos, and Sebastian Pannasch. 2003. Visual fixations as a rapid indicator of hazard perception.NATO Science Series Sub Series I: Life and Behavioural Sciences355 (2003), 313–322

    work page 2003

  52. [52]

    Victor, Emma Tivesten, Pär Gustavsson, Joel Johansson, Fredrik Sang- berg, and Mikael Ljung Aust

    Trent W. Victor, Emma Tivesten, Pär Gustavsson, Joel Johansson, Fredrik Sang- berg, and Mikael Ljung Aust. 2018. Automation Expectation Mismatch: Incorrect Prediction Despite Eyes on Threat and Hands on Wheel.Human Factors: The Journal of the Human Factors and Ergonomics Society60, 8 (2018), 1095–1116. doi:10.1177/0018720818788164

    work page doi:10.1177/0018720818788164 2018

  53. [53]

    Yuhui Wang, Hao He, and Xiaoyang Tan. 2020. Truly Proximal Policy Optimiza- tion. InProceedings of The 35th Uncertainty in Artificial Intelligence Conference, Ryan P. Adams and Vibhav Gogate (Eds.), Vol. 115. PMLR, 113–122

    work page 2020

  54. [54]

    Jokinen, Antti Oulasvirta, and Gustav Markkula

    Yueyang Wang, Aravinda Ramakrishnan Srinivasan, Jussi P.P. Jokinen, Antti Oulasvirta, and Gustav Markkula. 2025. Pedestrian crossing decisions can be explained by bounded optimal decision-making under noisy visual perception. Transportation Research Part C: Emerging Technologies171 (2025), 104963. doi:10. 1016/j.trc.2024.104963

    work page arXiv 2025

  55. [55]

    Yueyang Wang, Aravinda Ramakrishnan Srinivasan, Yee Mun Lee, and Gustav Markkula. 2025. Modeling Pedestrian Crossing Behavior: A Reinforcement Learning Approach With Sensory Motor Constraints.IEEE Transactions on Intelligent Transportation Systems26, 10 (2025), 16454–16465. doi:10.1109/TITS. 2025.3581693

    work page doi:10.1109/tits 2025

  56. [56]

    McDonald, Ranjana K

    Ran Wei, Anthony D. McDonald, Ranjana K. Mehta, and Alfredo Garcia. 2025. Active Inference Models of AV Takeovers: Relating Model Parameters to Trust, Sit- uation Awareness, and Fatigue.Human Factors: The Journal of the Human Factors and Ergonomics Society67, 6 (2025), 616–634. doi:10.1177/00187208241295932

    work page doi:10.1177/00187208241295932 2025

  57. [57]

    Christopher D. Wickens. 2008. Multiple Resources and Mental Workload.Human Factors: The Journal of the Human Factors and Ergonomics Society50, 3 (2008), 449–455. doi:10.1518/001872008X288394

    work page doi:10.1518/001872008x288394 2008

  58. [58]

    Wright, Ravi Agrawal, Siby Samuel, Yuhua Wang, Shlomo Zilberstein, and Donald L

    Timothy J. Wright, Ravi Agrawal, Siby Samuel, Yuhua Wang, Shlomo Zilberstein, and Donald L. Fisher. 2018. Effective cues for accelerating young drivers’ time to transfer control following a period of conditional automation.Accident Analysis & Prevention116 (2018), 14–20. doi:10.1016/j.aap.2017.10.005

    work page doi:10.1016/j.aap.2017.10.005 2018

  59. [59]

    Junwei Zhang, Zhenghao Zhang, Shuai Han, and Shuai Lü. 2022. Proximal policy optimization via enhanced exploration efficiency.Information Sciences609 (2022), 750–765. doi:10.1016/j.ins.2022.07.111

    work page doi:10.1016/j.ins.2022.07.111 2022

  60. [60]

    Jieyu Zhu, Yanli Ma, Yiran Zhang, Yaping Zhang, and Chen Lv. 2023. Takeover quality prediction based on driver physiological state of different cognitive tasks in conditionally automated driving.Advanced Engineering Informatics57 (2023), 102100. doi:10.1016/j.aei.2023.102100

    work page doi:10.1016/j.aei.2023.102100 2023