Multidimensional Assessment of Takeover Performance in Conditionally Automated Driving

Bani Anvari; Jan Luca K\"astle; J.W.C. van Lint; Kexin Liang; Simeon C. Calvert

REVIEW 3 major objections 5 minor 13 references

Multidimensional Assessment of Takeover Performance in Conditionally Automated Driving

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Situational awareness and spare capacity play distinct roles in automated-driving takeovers: awareness governs fast reflexive reactions, spare capacity governs takeover quality.

desk verdict A useful ten-metric map of SA versus SC in takeover performance, but the SC-quality link leans on gaze features measured inside the outcome window, so the causal framing needs to be reined in or defended. read the letter →

arxiv 2507.22252 v2 pith:L3EWBAD2 submitted 2025-07-29 cs.HC

Kexin Liang , Jan Luca Kästle , Bani Anvari , Simeon C. Calvert , J.W.C. van Lint This is my paper

classification cs.HC

keywords situationalawarenesssparecapacitytakeoverperformanceconditionallyautomateddrivingreactiontimesqualitysimulatorXGBoost

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Conditionally automated cars must hand control back to the driver in emergencies, and how well that handover goes depends on more than the driver's age or experience. This paper argues that two cognitive states play distinct roles: situational awareness (how well the driver perceives and understands the traffic situation) and spare capacity (how much capability remains after task demands are subtracted). Using a driving simulator with 57 participants and 466 takeover events, it claims that awareness primarily governs fast, reflexive reactions such as pressing the mode-switch button and looking back at the road, while spare capacity governs the quality of the takeover, including subjective comfort and the smoothness of steering, acceleration, and braking. The practical point is that a single readiness indicator is insufficient; systems that monitor and support drivers should track both states, because each predicts a different failure mode.

What carries the argument

The machinery is a nested predictor comparison using gradient-boosted decision trees (XGBoost) with Shapley-value (SHAP) interpretation, a method that attributes each prediction to individual input features with direction and magnitude. Four model families predict each of ten takeover metrics: driver characteristics alone (DC), DC plus SA, DC plus SC, and DC plus SA plus SC; improvements in prediction error show which construct carries explanatory power for which outcome. SA is operationalized by self-report scales and by pre-request gaze statistics (fixation counts, fixations on road, dashboard, and mirrors, and fixation and saccade durations), while SC is operationalized by self-reported task capability and demand and by gaze interactions with road, mirrors, and HMI during the request-to-lane-change window. The model comparison, Bonferroni-adjusted significance tests on input features, and SHAP direction and magnitude are what let the paper attribute reflexive reactions to SA and takeover quality to SC.

What would settle it

Recompute spare-capacity predictors using only gaze data from the 30 seconds before the takeover request rather than from the request-to-lane-change window; if the prediction improvements for steering, acceleration, and deceleration disappear, the claimed causal role of spare capacity is largely overlap between predictor and outcome. A complementary check is to vary spare capacity experimentally with a secondary task while holding pre-request gaze fixed and observe whether takeover quality still changes.

Watch

Extended reading notes

Core claim

The paper's central claim is that situational awareness (SA) and spare capacity (SC) are not interchangeable readiness signals: they explain complementary components of takeover performance. SA, measured by self-reported arousal, spare attention, and gaze patterns before the takeover request, predicts the time to reflexive actions such as pressing the takeover button ($t_{button}$) and redirecting gaze to the road ($t_{road}$); higher SA means faster reactions. SC, measured by perceived task capability and demand plus gaze engagement with mirrors, HMI, and road between the request and the lane change, predicts reflective response time ($ToT$) and takeover quality: lower SC is associated with longer conscious takeover time, lower satisfaction, higher perceived risk, and more abrupt steering, acceleration, and deceleration. For most quality metrics, adding SC to the driver-characteristics baseline improves prediction substantially, while SA adds little once SC is included; for reflexive reaction times, the pattern reverses. The paper concludes that the two constructs should be monitored and supported separately when designing human-automation handovers.

Load-bearing premise

The load-bearing assumption is that spare capacity can be measured from how the driver's eyes move during the takeover itself, even though that gaze behavior is recorded in the same time window as the performance being predicted.

Editorial extensions

If this is right

System designers should not treat situational awareness as a proxy for overall takeover readiness; high SA predicts fast reflexive reactions but does not by itself ensure smooth, comfortable control.
Takeover quality models gain the most from spare-capacity indicators, especially self-reported task capability; monitoring driver capability and demand could flag drivers likely to brake or steer abruptly.
Reaction-time metrics should be split into reflexive and conscious phases, since the two respond to different cognitive states; combining them can obscure the awareness effect.
Driver characteristics alone explain perceived time sufficiency and minimum time-to-collision, so stable traits should remain part of any personalized takeover-support design.
When both fast reactions and high-quality control matter, neither SA nor SC alone is sufficient; both need to be measured and supported.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A direct test of the causal claim would be to recompute spare-capacity predictors from gaze data collected only before the takeover request; if the quality predictions lose their edge, part of the reported SC effect is an artifact of the predictor sharing its time window with the outcome.
The reflexive/reflective split offers a way to reconcile contradictory earlier findings: studies measuring button presses or hand-on-wheel time will tend to credit situational awareness, while studies measuring full takeover time or trajectory smoothness will tend to credit workload or spare capacity.
A practical extension is an adaptive alert system that watches gaze for low awareness to speed up reactions and watches task demand and capability to soften the manual takeover; the paper provides evidence for the two channels but does not implement such a system.
Because subjective spare-capacity ratings dominate the quality models, an independent behavioral measure of spare capacity, such as secondary-task accuracy, would clarify whether the effect comes from actual available capacity or from drivers' confidence.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Worth a look. The genuinely new thing is the systematic ten-metric comparison: same experiment, same participants, SA and SC measured side by side, with the reflexive/reflective split (button press and road reorientation vs takeover time) and the quality dimensions separated. The XGBoost results are internally consistent across Random Forest and LightGBM, and the SHAP analyses give a readable picture of which features matter for each metric. That is a real empirical contribution to the automated-driving human factors literature, and the authors give credit to prior SA and TCI work.

The soft spots are concentrated in what the authors claim from the models. First, the SC objective gaze features in Table 31 are computed in the window between the takeover request and lane change — the same window that defines most of the outcome metrics. Predictors measured during the maneuver cannot establish that a pre-existing resource 'influences' the maneuver. The t_road duplication is the sharpest case: it appears as an SC predictor in Table 1 and as a reaction-time outcome in Table 2. The paper never says it was excluded. Table 5's negligible SC improvement for t_road suggests it probably was, but the omission is the kind of thing a referee should force into the open. More broadly, features like mirror-check counts and durations are concurrent behaviors, not pre-existing spare capacity. The paper needs a sensitivity analysis without those in-window features, and a statement about which variables were excluded from their own outcome models.

Second, the 10-fold cross-validation splits trials, not participants. With 57 drivers and 466 takeovers, the same driver appears in train and test. That can inflate all the RMSE improvements, and it does not disappear because three tree models agree — they share the same leakage. Participant-level CV is needed.

Third, the causal language is too strong. 'SA enables,' 'SC influences,' 'SC determines' — the data are predictive correlations. The authors do temper some conclusions with 'correlating' in the abstract, but the Discussion and Conclusion revert to causal phrasing. And for the subjective outcomes, self-reported SC and self-reported risk/satisfaction come from the same post-takeover questionnaire, so common method variance is a live concern.

None of this kills the empirical contribution. The distinction between awareness-driven reflexive responses and capacity-related quality is plausible and worth testing. But as written, the central claim is stronger than the analysis supports. This paper should go to peer review with a request for major revisions: participant-level CV, explicit feature-exclusion checks, an in-window feature sensitivity analysis, and softened causal language. The right reader is someone working on adaptive takeover alerts or cognitive-state monitoring in automated driving; that reader will find the ten-metric comparison useful even while treating the causal framing with caution.

Referee Report

3 major / 5 minor

Summary. The paper reports a driving-simulator study with 57 participants and 466 takeover events, in which XGBoost models (validated with Random Forest and LightGBM) are used to predict ten takeover-performance metrics from three predictor groups: basic Driver Characteristics (DC), Situational Awareness (SA), and Spare Capacity (SC). SA is measured by self-report and by gaze metrics collected in the 30 s before the takeover request; SC is measured by self-reported task capability/demand and by gaze metrics collected between the takeover request and the completion of the lane change. The reported results are that SA is the dominant predictor of fast reflexive responses (button-press time and road-reorientation time), while SC dominates prediction of takeover time, subjective risk/satisfaction, and objective quality metrics such as maximum steering, acceleration, and deceleration. The paper concludes that SA and SC play distinct and complementary roles in takeover performance.

Significance. If the findings hold, they would provide a useful multidimensional decomposition of takeover performance and a practical distinction between two cognitive constructs that are often conflated. The study has notable strengths: a nontrivial experimental dataset, explicit comparison of three tree-based models in Table 23, use of repeated cross-validation, Bonferroni-adjusted significance testing, SHAP-based interpretation, and discussion of external literature. The consistency of the RMSE patterns across XGBoost, Random Forest, and LightGBM is a genuine internal-validity plus. However, the central claim that SC causally shapes takeover quality is currently threatened by a predictor--outcome temporal overlap that the manuscript does not address, so the significance is conditional on a sensitivity analysis that has not yet been reported.

major comments (3)

[Section 2.3.1, Tables 2 and 31] The SC gaze metrics in Table 31 are computed for the period between the initiation of takeover requests and the completion of lane changes, which is exactly the interval over which the outcome metrics in Table 2 are defined. Most concretely, t_road appears both as a reaction-time outcome in Table 2 ('time to first visually fixate on the road following a TOR') and as an SC predictor in Table 31 ('the interval between the initiation of a takeover request and the moment the driver first establishes visual fixation on the forward road'). The paper never states that each outcome was excluded from its own predictor set, and models such as XGBToT_dc+sc (Table 7) and XGBOQsteer_dc+sc (Table 17) draw SC features from this same window. A feature generated by the very maneuver being predicted cannot establish that a pre-existing resource 'influences' the maneuver; at best it shows that in-maneuver gaze behavior covaries with in-maneuver control outcomes. The asymmetry with SA, whose gaze metrics are measured in the 30 s before the TOR (Table 30), makes the DC+SC versus DC+SA comparison temporally unbalanced. I request a sensitivity analysis that removes all same-window gaze features, or at minimum excludes each outcome and its operational equivalents from the predictor set, and a clear statement of what was done.
[Section 2.3.2] The modeling section states that 10-fold cross-validation repeated 100 times is applied to the 466 takeovers, but it does not state that the folds were grouped by participant. With 57 participants contributing multiple takeovers each, random splits place takeovers from the same driver in both training and test folds, so the model can exploit stable individual-specific patterns (including self-report style and gaze tendencies) that would not transfer to a new driver. This can bias the reported RMSE improvements and the incremental comparisons that underpin the paper's conclusions. Please use participant-level or leave-one-participant-out cross-validation, or explicitly justify why ungrouped folds are appropriate for the generalization claim being made.
[Section 5 and Section 4] The conclusions use causal language: 'SA predominantly facilitates faster takeover responses' and 'SC more substantially influences takeover quality, with lower SC levels generally correlating with poorer subjective comfort ratings and more abrupt control operations.' The data are observational, the SA and SC measures are partly concurrent with the outcomes, and the analysis is predictive rather than experimental. The conclusions should be reframed in terms of association and predictive contribution, unless the authors can support temporal precedence by restricting SC predictors to measurements taken before the takeover request. The limitations paragraph in Section 4 lists three limitations but does not mention this predictor--outcome overlap, which is arguably the most consequential threat to the central claim.

minor comments (5)

[Figure 4 caption] The caption reads 'significant Spare Capacity (S A) factors' but should be 'Spare Capacity (SC) factors.'
[Table 17] The input label for the combined model is 'OQDC + SA + SC', which appears to be a typo for 'DC + SA + SC'.
[Table 1 vs. Table 31] Table 1 lists SC objective variables as t_road, t_HMI, t_mirror, PCT_road, PCT_HMI, PCT_mirror, NO_road, NO_HMI, NO_mirror, AVG_road, AVG_HMI, and AVG_mirror, whereas Table 31 defines a DUR_road/DUR_HMI/DUR_mirror family and does not define any PCT variables; the notation should be harmonized.
[Section 3.2.2] The sentence 'A summary plot derived from SHAP values is illustrated in Figure 5' appears to refer to Figure 6, since Figure 5 is the time-sufficiency plot; please correct the cross-reference.
[Section 2.3.2] The term 'internal validity' is used to describe agreement between XGBoost, Random Forest, and LightGBM, but that agreement is better described as algorithmic robustness or predictive consistency; internal validity also requires attention to leakage and construct validity, which are the concerns raised above.

Circularity Check

2 steps flagged · score 6.0 of 10

The SC advantage over SA is partly self-inflicted: t_road is defined both as an SC predictor and as a reaction-time outcome, and all SC gaze predictors are measured in the same TOR-to-lane-change window as the outcomes, so several 'SC predicts takeover quality' results are tautological or temporally overlapping as written.

self definitional [Section 2.3.1, Tables 1 and 2; Section 3.1.2, Table 5]
"objective measurements t_road, t_HMI, t_mirror, PCT_road, PCT_HMI, PCT_mirror, NO_road, NO_HMI, NO_mirror, AVG_road, AVG_HMI, AVG_mirror (Table 1); t_road s time to first visually fixate on the road following a TOR (Table 2); t_road s the interval between the initiation of a takeover request and the moment ... the driver first establishes visual fixation on the forward road (Table 31); XGBTroad_dc+sc inputs DC+SC (Table 5)."

By the paper's own tables, t_road is simultaneously the outcome metric 'road reorientation time' and an SC objective predictor. The model XGBTroad_dc+sc is defined with inputs DC+SC, so its predictor set includes the target variable itself. If t_road were actually used, the reported RMSE reduction would be an identity artifact; the paper does not state that the outcome was excluded from its own predictor set. Any conclusion about which factors explain t_road therefore rests on an undefined exclusion rule.
other [Section 2.3.1, Table 31; Section 3.3]
"Definitions of Spare Capacity (SC)-related visual metrics for the period between the initiation of takeover requests and the completion of lane changes (Table 31); four key metrics spanning the period from the initiation of the takeover request to the lane change (Section 3.3)."

The SC predictors and the outcome metrics are computed over the identical TOR-to-lane-change interval, so the predictors are simultaneous outputs of the same maneuver rather than antecedent spare-capacity states. The central comparison DC+SC vs DC+SA is temporally imbalanced because SA gaze metrics are measured in the 30 s before the TOR (Table 30). The reported SC advantage for objQ_steer/acc/dec and ToT may therefore be an artifact of predictor-outcome interval overlap; no sensitivity check excluding in-maneuver gaze features is reported.

full rationale

The paper is not globally circular: the XGBoost framework, cross-validation, RF/LightGBM robustness checks, and external comparisons are genuine. The main circularity is contained in the predictor definitions. As written, Table 1 lists t_road as an SC objective predictor while Table 2 defines t_road as a reaction-time outcome, and Section 2.3.2 defines the DC+SC models as using the SC set for every metric; no exclusion of the target variable from its own predictor set is stated. Consequently, the t_road model with SC would be able to copy the target. The reported RMSE for XGBTroad_dc+sc (0.6889) suggests the implementation may have silently dropped the target, but the manuscript never says this, so the derivation as written is tautological for that metric. Independently, all twelve SC gaze predictors are measured in the TOR-to-lane-change window that also defines the reaction-time and objective-quality outcomes, while SA gaze predictors are measured in the 30 s before the TOR; the conclusion that SC dominates takeover quality is therefore not a clean test of a pre-existing spare-capacity construct. These are specific, quotable overlaps rather than a vague 'feels circular' objection. Self-citations (e.g., Liang et al., 2024) are used for questionnaire provenance and background, not as a load-bearing uniqueness argument, so they do not add to the score.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The paper introduces no new physical or theoretical entities; SA, SC, and subjective quality are operationalized from existing instruments. The main burden sits in measurement assumptions and unstated modeling choices: gaze metrics as spare capacity, trial-level independence, and the causal interpretation of predictive feature attributions. No fitted constants or invented mediators are required.

free parameters (3)

XGBoost hyperparameters = not reported
Tree depth, learning rate, regularization, and number of estimators are chosen by hand or by tuning, but the values are not given. The reported RMSE improvements and feature importances depend on these choices.
Behavioral metric thresholds = steering >2 degrees, pedal >10%, fixation velocity <900 px/s, 7 s takeover lead time
Hand-set cutoffs define reaction times and gaze features. Small changes in these thresholds could shift which takeovers are counted as fast or which fixations are counted as meaningful.
Cross-validation folds and repeats = 10-fold, 100 repeats
The repeated 10-fold scheme is a modeling choice. It does not group by participant, so same-driver trials can appear in both training and test folds, potentially inflating predictive performance.

assumptions (5)

domain assumption Questionnaire-based SA, SC, and subjective quality measure distinct psychological constructs rather than shared response biases.
Invoked in Section 2.2, where SA, SC, and subjective takeover quality are all collected via self-report. If common method variance is large, the strong SC-subjective quality links could be inflated.
domain assumption Gaze behavior during the takeover maneuver reflects spare capacity rather than strategic or reactive behavior of the maneuver itself.
Invoked in Table 31, where SC-related gaze metrics are measured from takeover request to lane change, the same period that defines the outcomes. This assumption is load-bearing and is not tested.
domain assumption Driving simulator behavior generalizes to real conditionally automated driving.
Acknowledged as a limitation in Section 4. The central design recommendations depend on this generalization, but no real-world validation is provided.
domain assumption Cross-validation folds treat takeover trials as independent even when produced by the same driver.
Invoked in Section 2.3.2 with 10-fold cross-validation on 466 takeovers from 57 drivers. Without participant-level splitting, driver-specific patterns can leak across folds.
domain assumption Tree-model feature importance and SHAP values can be interpreted as causal effects of SA and SC on takeover performance.
The abstract and conclusions use causal language such as 'enables' and 'influences', but the analysis is predictive. This assumption is not defended in the paper.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Multidimensional Assessment of Takeover Performance in Conditionally Automated Driving." pith.science (2026). https://pith.science/paper/L3EWBAD2

@misc{pith2026250722252,
  author       = {Pith},
  title        = {Pith review of: Multidimensional Assessment of Takeover Performance in Conditionally Automated Driving},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/L3EWBAD2}},
  note         = {Machine review of arXiv:2507.22252}
}

read the original abstract

When automated driving systems encounter complex situations beyond their operational capabilities, they issue takeover requests, prompting drivers to resume vehicle control and return to the driving loop as a critical safety backup. However, this control transition places significant demands on drivers, requiring them to promptly respond to takeover requests while executing high-quality interventions. To ensure safe and comfortable control transitions, it is essential to develop a deep understanding of the key factors influencing various takeover performance aspects. This study evaluates drivers' takeover performance across three dimensions: response efficiency, user experience, and driving safety - using a driving simulator experiment. EXtreme Gradient Boosting (XGBoost) models are used to investigate the contributions of two critical factors, i.e., Situational Awareness (SA) and Spare Capacity (SC), in predicting various takeover performance metrics by comparing the predictive results to the baseline models that rely solely on basic Driver Characteristics (DC). The results reveal that (i) higher SA enables drivers to respond to takeover requests more quickly, particularly for reflexive responses; and (ii) SC shows a greater overall impact on takeover quality than SA, where higher SC generally leads to enhanced subjective rating scores and objective execution trajectories. These findings highlight the distinct yet complementary roles of SA and SC in shaping performance components, offering valuable insights for optimizing human-vehicle interactions and enhancing automated driving system design.

Figures

Figures reproduced from arXiv: 2507.22252 by the authors.

**Figure 1.** Instrumentation in the study. The experiment comprises nine takeover scenarios, varying in traffic densities and Non-Driving Related Tasks (NDRTs). Traffic densities are manipulated by introducing 0, 10, or 20 vehicles per kilometre, while NDRTs are induced using 𝑛-back tasks 1 where 𝑛 = 0, 1, 2. These scenarios are arranged using a Latin Square design to ensure balanced exposure (Calvert, Taale, Snelder and Hoogend… view at source ↗

**Figure 2.** Summary plot of significant Driver Characteristics (𝐷𝐶) and Situational Awareness (𝑆𝐴) factors influencing the button press time (𝑡 𝑏𝑢𝑡𝑡𝑜𝑛) [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

**Figure 3.** Summary plot of significant Driver Characteristics (𝐷𝐶) and Situational Awareness (𝑆𝐴) factors influencing the road reorientation time (𝑡 𝑟𝑜𝑎𝑑 ) [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

Figures from the paper (8 more)

**Figure 4.** Figure 4: Summary plot of significant Spare Capacity (𝑆𝐴) factors influencing drivers’ takeover time (𝑇𝑜𝑇 ). 3.1.4. Summary Two main patterns in drivers’ reaction times following takeover requests are identified: (i) Drivers’ immediate responses - including mode switch activatio…

**Figure 5.** Figure 5: Summary plot of significant Driver Characteristics (𝐷𝐶) factors influencing drivers’ perceived time sufficiency (𝑠𝑢𝑏𝑗𝑄𝑠𝑢𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑐𝑦) [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]

**Figure 6.** Figure 6: Summary plot of significant Driver Characteristics (𝐷𝐶) and Spare Capacity (𝑆𝐶) factors influencing drivers’ perceived risk (𝑠𝑢𝑏𝑗𝑄𝑟𝑖𝑠𝑘) [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]

**Figure 7.** Figure 7: Summary plot of significant Driver Characteristics (𝐷𝐶) and Spare Capacity (𝑆𝐶)- factors influencing drivers’ perceived performance satisfaction (𝑠𝑢𝑏𝑗𝑄𝑠𝑎𝑡𝑖𝑠𝑓𝑎𝑐𝑡𝑖𝑜𝑛). 3.2.4. Summary Three main patterns in drivers’ subjective takeover experience are identified: (i) Perce…

**Figure 8.** Figure 8: Summary plot of significant Driver Characteristics (𝐷𝐶) factors influencing minimum time to collision (𝑜𝑏𝑗𝑄𝑡𝑡𝑐 ). 3.3.2. Maximum steering wheel angle [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]

**Figure 9.** Figure 9: Summary plot of significant Spare Capacity (𝑆𝐶) factors influencing maximum steering wheel angle (𝑜𝑏𝑗𝑄𝑠𝑡𝑒𝑒𝑟). 3.3.3. Maximum acceleration [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]

**Figure 10.** Figure 10: Summary plot of significant Driver Characteristics (𝐷𝐶) and Spare Capacity (𝑆𝐶) factors influencing maximum acceleration (𝑜𝑏𝑗𝑄𝑎𝑐𝑐 ). K. Liang et al.: Preprint submitted to Elsevier Page 17 of 23 [PITH_FULL_IMAGE:figures/full_fig_p017_10.png]

**Figure 11.** Figure 11: Summary plot of significant Spare Capacity (𝑆𝐶) factors influencing maximum deceleration (𝑜𝑏𝑗𝑄𝑑𝑒𝑐 ) [PITH_FULL_IMAGE:figures/full_fig_p019_11.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

13 extracted references · 12 canonical work pages

[2]

take over!

“take over!” how long does it take to get the driver back into the loop?, in: Proceedings of the human factors and ergonomics society annual meeting, Sage Publications Sage CA: Los Angeles, CA. pp. 1938–1942. Guo, H., Zhang, Y., Cai, S., Chen, X.,

work page 1938
[5]

Table 30 Definitions of Situational Awareness (SA)-related visual metrics during the 30 seconds preceding takeover requests. Metrics Unit Description Reference 𝑛𝑟𝑟𝑜𝑎𝑑 – the number of fixations on the forward road Kästle (2024); Zhou, Yang and De Winter (2021) 𝑛𝑟𝑟𝑒𝑎𝑟𝑣𝑖𝑒𝑤 – the number of fixations on the rear-view mirror 𝑛𝑟𝑠𝑖𝑑𝑒𝑣𝑖𝑒𝑤 – the number of fixations...

work page 2024
[9]

Plos one 16, e0260953

Perceived safety and trust in sae level 2 partially automated cars: Results from an online questionnaire. Plos one 16, e0260953. Nordhoff,S.,Stapel,J.,He,X.,Gentner,A.,Happee,R.,2023. Dodriver’scharacteristics,systemperformance,perceivedsafety,andtrustinfluence how drivers use partial automation? a structural equation modelling analysis. Frontiers in Psychology

work page 2023
[29]

better safe than sorry

All data were collected using the questionnaire detailed in Appendix A. C. Metric definition This Appendix details the definitions of Situational Awareness (SA)-related visual metrics (Table 30), Spare Capacity (SC)-related visual metrics (Table 31), and drivers’ operational metrics (Table 32). K. Liang et al.:Preprint submitted to Elsevier Page 23 of 23 ...

work page 2023
[1988]

Elsevier

Development of nasa-tlx (task load index): Results of empirical and theoretical research, in: Advances in psychology. Elsevier. volume 52, pp. 139–183. Hu,L.,Cai,H.,Huang,J.,Cao,D.,Zhang,X.,2023. Thechallengesofdrivingmodeswitchinginautomatedvehicles:Areview. IEEETransactions on Vehicular Technology . Hydén, C.,

work page 2023
[1995]

Personality and individual differences 19, 307–318

Driving experience, personality, and skill and safety-motive dimensions in drivers’ self-assessments. Personality and individual differences 19, 307–318. Li,Q.,Wang,Z.,Wang,W.,Zeng,C.,Wu,C.,Li,G.,Heh,J.S.,Cheng,B.,2023.Ahuman-centeredcomprehensivemeasureoftake-overperformance based on multiple objective metrics. IEEE Transactions on Intelligent Transporta...

work page 2023
[1996]

Traffic safety work with video processing 37, 3–14

Traffic conflicts technique: state-of-the-art. Traffic safety work with video processing 37, 3–14. Jia,L.,Huang,C.,Du,N.,2024. Drivers’situationalawarenessofsurroundingvehiclesduringtakeovers:Evidencefromadrivingsimulatorstudy. Transportation Research Part F: Traffic Psychology and Behaviour 106, 340–355. Kästle, J.L.,

work page 2024
[2010]

Accident Analysis & Prevention 42, 2125–2131

Parental and offspring assessment of driving capability under the influence of drugs or alcohol: Gender and inter-generational differences. Accident Analysis & Prevention 42, 2125–2131. Sharma,P.K.,Chakraborty,P.,2024. Areviewofdrivergazeestimationandapplicationingazebehaviorunderstanding. EngineeringApplications of Artificial Intelligence 133, 108117. Ta...

work page 2024

Show all 13 references

[2013]

Springer, pp

Retrieving human control after situations of automated driving: How to measure situation awareness, in: Advanced Microsystems for Automotive Applications 2013: Smart Systems for Safe and Green Vehicles. Springer, pp. 43–53. Calvert, S., Johnsen, S., George, A.,

2013
[2020]

arXiv preprint arXiv:2009.00508

A high-level description and performance evaluation of pupil invisible. arXiv preprint arXiv:2009.00508 . Van Lint, J., Calvert, S.C.,

2009 arXiv
[2021]

Journal of advanced transportation 2021, 8632685

Effects of level 3 automated vehicle drivers’ fatigue on their take-over behaviour: A literature review. Journal of advanced transportation 2021, 8632685. Hart, S.G., Staveland, L.E.,

2021
[2023]

Accident Analysis & Prevention 187, 107068

Supervising the self-driving car: Situation awareness and fatigue during highly automated driving. Accident Analysis & Prevention 187, 107068. K. Liang et al.:Preprint submitted to Elsevier Page 22 of 23 Multidimensional Assessment of Takeover Performance Miller,D.,Sun,A.,John...

2015
[2024]

Examining the role of driver perception in takeover time: Application of task-capability interface theory, in: 2024 IEEE Intelligent Vehicles Symposium (IV), IEEE. pp. 1–6. Liang,N.,Yang,J.,Yu,D.,Prakah-Asante,K.O.,Curry,R.,Blommer,M.,Swaminathan,R.,Pitts,B.J.,2021. Usingeye-t...

2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [2]

take over!

“take over!” how long does it take to get the driver back into the loop?, in: Proceedings of the human factors and ergonomics society annual meeting, Sage Publications Sage CA: Los Angeles, CA. pp. 1938–1942. Guo, H., Zhang, Y., Cai, S., Chen, X.,

work page 1938

[2] [5]

Table 30 Definitions of Situational Awareness (SA)-related visual metrics during the 30 seconds preceding takeover requests. Metrics Unit Description Reference 𝑛𝑟𝑟𝑜𝑎𝑑 – the number of fixations on the forward road Kästle (2024); Zhou, Yang and De Winter (2021) 𝑛𝑟𝑟𝑒𝑎𝑟𝑣𝑖𝑒𝑤 – the number of fixations on the rear-view mirror 𝑛𝑟𝑠𝑖𝑑𝑒𝑣𝑖𝑒𝑤 – the number of fixations...

work page 2024

[3] [9]

Plos one 16, e0260953

Perceived safety and trust in sae level 2 partially automated cars: Results from an online questionnaire. Plos one 16, e0260953. Nordhoff,S.,Stapel,J.,He,X.,Gentner,A.,Happee,R.,2023. Dodriver’scharacteristics,systemperformance,perceivedsafety,andtrustinfluence how drivers use partial automation? a structural equation modelling analysis. Frontiers in Psychology

work page 2023

[4] [29]

better safe than sorry

All data were collected using the questionnaire detailed in Appendix A. C. Metric definition This Appendix details the definitions of Situational Awareness (SA)-related visual metrics (Table 30), Spare Capacity (SC)-related visual metrics (Table 31), and drivers’ operational metrics (Table 32). K. Liang et al.:Preprint submitted to Elsevier Page 23 of 23 ...

work page 2023

[5] [1988]

Elsevier

Development of nasa-tlx (task load index): Results of empirical and theoretical research, in: Advances in psychology. Elsevier. volume 52, pp. 139–183. Hu,L.,Cai,H.,Huang,J.,Cao,D.,Zhang,X.,2023. Thechallengesofdrivingmodeswitchinginautomatedvehicles:Areview. IEEETransactions on Vehicular Technology . Hydén, C.,

work page 2023

[6] [1995]

Personality and individual differences 19, 307–318

Driving experience, personality, and skill and safety-motive dimensions in drivers’ self-assessments. Personality and individual differences 19, 307–318. Li,Q.,Wang,Z.,Wang,W.,Zeng,C.,Wu,C.,Li,G.,Heh,J.S.,Cheng,B.,2023.Ahuman-centeredcomprehensivemeasureoftake-overperformance based on multiple objective metrics. IEEE Transactions on Intelligent Transporta...

work page 2023

[7] [1996]

Traffic safety work with video processing 37, 3–14

Traffic conflicts technique: state-of-the-art. Traffic safety work with video processing 37, 3–14. Jia,L.,Huang,C.,Du,N.,2024. Drivers’situationalawarenessofsurroundingvehiclesduringtakeovers:Evidencefromadrivingsimulatorstudy. Transportation Research Part F: Traffic Psychology and Behaviour 106, 340–355. Kästle, J.L.,

work page 2024

[8] [2010]

Accident Analysis & Prevention 42, 2125–2131

Parental and offspring assessment of driving capability under the influence of drugs or alcohol: Gender and inter-generational differences. Accident Analysis & Prevention 42, 2125–2131. Sharma,P.K.,Chakraborty,P.,2024. Areviewofdrivergazeestimationandapplicationingazebehaviorunderstanding. EngineeringApplications of Artificial Intelligence 133, 108117. Ta...

work page 2024