Traffic capacity constraints from level 3 control transitions

Abstract

With the increasing integration of conditionally automated Level 3 systems into real-world traffic, concerns about their impact on traffic efficiency and capacity have emerged. When such systems reach their operational limits, mandatory control transitions could disrupt traffic flow and reduce overall capacity. This study employs large-scale simulations and numerical experiments to analyze these effects and quantify potential capacity constraints. The results of the two-lane highway scenario show an experimental capacity reduction of up to in an almost fully automated but unmanaged traffic mix, corresponding to a loss of about . Control transition-related effects become increasingly pronounced at a Level 3 penetration rate between and . Estimated capacity reductions suggest that the maxima in time headway increments during the transition phase contribute most to these effects.

1 Introduction

As manufacturers begin to introduce Level 3 automated driving systems to the market, the potential impact of such systems on overall traffic flow and capacity needs to be investigated. A key challenge arises from the fact that Level 3 systems require human drivers to take over control when reaching system limits, leading to so-called transitions of control (ToC), which may disrupt traffic flow and reduce road capacity. Despite regulatory advancements concerning Level 3 systems (R157 by UNECE (2023)), the macroscopic impact of such procedural ToC effects on traffic conditions remains insufficiently explored. This raises the general question of how Level 3 control transitions in conditionally automated vehicles (AVs) affect traffic capacity and, more specifically, what characteristics of procedural ToC-induced time headway increments in vehicle strings contribute to this effect. To investigate this, we conduct a large-scale simulation-based analysis and complement it with simplified numerical experiments to estimate macroscopic capacity impacts. Our study also explores the underlying mechanisms of the transition phase in greater detail. Existing research on potential capacity gains from AVs, as exemplified by Friedrich (2016) and Park et al. (2021), has primarily focused on higher automation levels (4–5) under optimistic assumptions of short time headways, e.g., s, in contrast to observed headways in manually driven vehicles of at least in freeway traffic, depending on vehicle speed, as shown by Wagner (2012). Our previous work in Alms et al. (2022) and Alms and Wagner (2024) touched on ToC-related capacity effects but lacked a comprehensive quantification of resulting capacity losses. This study addresses these gaps by (i) adopting a macroscopic perspective using realistic, R157-compliant time headways of s, and (ii) introducing an exploratory estimation approach that explicitly accounts for Level 3 disengagements in road capacity assessment.

The rest of the paper is organized as follows: Section 2 introduces the conceptual aspects of ToCs in Level 3 automated systems. In Section 3, we present a highway scenario calibration based on real-world detector data. Section 4 details our methodology for investigating ToC-related capacity effects in a simulation study, while Section 5 presents and discusses our results, comparing simulated and estimated capacity reductions. Lastly, Section 6 offers our perspective on the interpretation and limitations of this study.

2 Transitions of control in level 3 automated driving

The six levels of driving automation, defined by SAE International (2021), not only classify automated driving functions and capabilities but also specify the human driver’s role in terms of engagement and responsibility, as illustrated in Figure 1. Conditional automated driving (Level 3, highlighted with a purple frame in Figure 1) represents a fundamental shift toward automated vehicle operation within defined Operational Design Domains (ODD), specified in British Standards Institution (2020), allowing human drivers to disengage from the primary driving task. However, if the Level 3 system requires the driver to resume control, a takeover request (ToR) is issued, initiating a critical transfer of authority: these procedures are referred to as transitions of control (ToC, plural: ToCs). Detailed insights into various aspects of ToCs are available through a comprehensive literature review on takeovers in automated driving (McDonald et al., 2019). Further studies examine the intricacies of modeling human factors, such as situational awareness and task demand (Van Lint and Calvert, 2018; Calvert and van Arem, 2020), or reduced driver performance (Wang et al., 2025b), during ToCs.

FIGURE 1

The current regulations R157 from UNECE (2023) specify technical requirements for the certification of Level 3 Automated Lane Keeping Systems (ALKS) and set the time range to before a failed transition escalates to a minimum risk maneuver (MRM), which is critical for the process of control transitions. Within the context of the EC project (TransAID, 2021; Lücken et al., 2019; Mintsis et al., 2019) introduced a novel ToC model, which is fully parametrizable to align with these later-established UNECE specifications and for which a detailed description of the model’s implementation is provided. The operationalization of ToCs is further specified in the ongoing EC project Hi-Drive (Bolovinou et al., 2023; Sauvaget et al., 2023) and demonstrated in Schulte-Tigges et al. (2023).

Figure 2 illustrates the basic mechanisms of the ToC model for successful and failed control transitions implemented in the microscopic traffic simulation SUMO (Alvarez Lopez et al., 2018). After a ToR, the AV enters a preparatory phase characterized by headway enlargement and disabled lane changing. Automated driving continues for the limited lead time, after which either the driver resumes control in time (successful transition), or, if not, the AV initiates an MRM (failed transition). For failed transitions, the AV initiates a phase of constant deceleration and may come to a full stop if the human driver does not respond. Although such events are rare, they can have a high impact and are the subject of extensive safety investigations based on disengagement reports (e.g., (Ward, 2024; Kohanpour et al., 2025)). However, this aspect is not the focus of the present work. In the case of a successful transition, the driver state model accounts for a phase of reduced human driving performance, with recent studies gaining further insights into both post-ToC durations (Wang et al., 2025a) and potential negative impacts on traffic stability (Wang et al., 2024).

FIGURE 2

FIGURE 3

These numerical experiments are highly simplified due to identical vehicle parametrizations, yet they effectively illustrate the isolated ToC effects discussed. Given the cumulative deceleration patterns observed, we expect that ToC-induced disturbances may lead to noticeable reductions in traffic capacity. To examine whether these effects also manifest under more realistic traffic flow conditions, we calibrate a SUMO simulation scenario to detector data in Section 3.

3 Calibrating SUMO for a highway traffic scenario

To analyze the impact of ToCs on traffic capacity, we use real-world detector data from a German highway west of Berlin as a reference for SUMO calibration. The following sections detail the dataset and simulation setup.

3.1 AVUS detector data

Figure 4a shows a section of the Bundesautobahn A115, referred to as AVUS, which was occasionally used as a motor racing track in the past and is a highly frequented highway with up to 80.000 vehicles per day. We hypothesize a potential ODD zone for Level 3 automated driving in the inbound segment of the road (cf.Figure 4a, panel (a)), which is a two-lane highway with speed limits of starting at an interchange section and increasing to up to the next traffic exit, that is located nearly downstream. The inbound traffic data on the two-lane section come from a detector at an underpass (cf.Figure 4a, panel (c)). After this point, the road has a slight slope for a few hundred meters, but the exact gradient could not be verified. A ramp merges onto the main edge about downstream, from where the speed limit increases to .

FIGURE 4

Figure 4b displays speed—flow relations for several years of the AVUS between 2015 and 2022, as scatterplots based on data from Digitale Plattform Stadtverkehr Berlin (2024). These data are originally tagged as hourly flows with corresponding average speeds per hour, but we suspect that this is not accurate. While the number of vehicles is accumulated over a full hour, the high variations in speeds at lower flow rates suggest that these data points from the detector database might actually represent speed averages over intervals of 1 minute or less. We were unable to verify this suspicion directly with the publisher of the data, but we argue that the actual speed value recorded in the database is likely the last entry of a full hour — possibly for efficiency and memory-saving reasons in data processing — rather than the average speed over the entire hour. This ultimately results in a notably wider distribution of speed values at lower flow rates than expected for true hourly data. For reference, we also added the model fit developed by Van Aerde (1995) to each plot.

Table 1 lists the yearly maximum flows , the and percentiles as suggested by Brilon and Geistefeldt (2010), and the deterministic capacity derived from the van Aerde model, as well as the corresponding shares for heavy good vehicles (HGVs) extracted from the raw detector data between 2016 and 2024. Additionally, data provided by the BASt (2025) from a detector downstream of the AVUS at “Eichkamp” are also included in the table for comparison (cf.Figure 4a, panel (b)). Note that 2024 shows an oddly high HGV share, which we consider very unlikely and attribute to recent technical changes in sensor-based detection and data processing by the provider. The report from BASt (2021) stated a nationwide HGV share of in 2021 on Germany’s highways.

3.2 Simulation setup for calibration

To investigate the impacts of ToCs in mixed-autonomy traffic, we compose a traffic mix of four different vehicle types: automated passenger vehicles (AVs), manual passenger vehicles (MVs), light goods vehicles (LGVs), and heavy goods vehicles (HGVs). The most relevant parameters for a heterogeneous traffic behavior in this AVUS highway scenario are visualized in Figure 5. Instead of utilizing SUMO’s default parameters, vehicle type specific distributions were deployed. Table 2 presents the full parametrization scheme for all vehicle types.

FIGURE 5

3.3 Calibration including ramp flow

In the first step, we ran simulations with increasing demand using MVs only. To better capture the full spectrum of the fundamental diagram in SUMO, we introduced additional vehicle flow on the incoming ramp. This creates a merging scenario, leading to traffic breakdown upstream of the main edge’s detector position. Figure 6 shows the speed–flow relations as scatterplots for (i) real-world detector data from 2024, (ii) SUMO’s default parametrization, and (iii) the aggregated main edge data from the final calibration. The graphic also color-codes the demand intensities from the on-ramp and highlights the maximum of the AVUS detector data 2024. For reference of the expected average speeds defined by the german Highway Capacity Manual—referred to as HBS—in (FGSV, 2015, Part A, Figure A3-10) for a two-lane highway (slope , speed limit ), a black solid line was added.

FIGURE 6

The phenomenon described in point 4 is, in part, a limitation of SUMO’s current modeling of cooperative lane-changing behavior between neighboring lanes under traffic breakdown conditions. Correspondingly, Figure 7 compares the lane-specific calibrated flows in SUMO with real AVUS data from 2024. We clearly identify the disparate speed levels between the lanes in SUMO (bottom panel), whereas the real-world data (top panel) indicate similar speed–flow relations on both lanes. Rummel (2017) indirectly revealed this issue in his investigation but was unable to unequivocally identify the lane-specific breakdowns as the underlying cause of SUMO’s oversaturation compared to the HBS predictions, nor did the report by Geistefeldt et al. (2017), which ultimately disregarded SUMO in its analysis for this very reason. While this limitation prevents a full replication of the real-world dynamics, we proceed with the calibration of the scenario as a basis for our analysis and will address this shortcoming in our future work.

FIGURE 7

3.4 Refining calibration by incorporating HGV share

In a second step, based on the parametrization scheme plausibilised for MVs in the ramp scenario (cf.Table 2), we conducted simulations with different shares of HGVs, LGVs, and MVs, but without any ramp flow. As a result, we can no longer reproduce the entire fundamental diagram for this highway scenario, since SUMO’s flow does not naturally lead to a traffic breakdown as observed in real-world highway traffic. The reason we need to disregard the unstable part of the fundamental diagram at this point is technical: SUMO does not maintain precise LGV/HGV shares for vehicle insertions when approaching maximum flow. Instead, the share of LGVs and HGVs declines to zero until SUMO can only insert MVs when the traffic breakdown at capacity is expected. This behavior stems partly from the parametrization of LGVs and HGVs, such as their larger vehicle lengths and time headways.

FIGURE 8

Even though these capacities are still about larger than the detector numbers in Table 1, we consider this an adequate calibration, particularly compared to SUMO’s default, since the real-world detector flow data are overall lower compared to the HBS range, which we identified to be between 3,600 and . Other local factors, such as road curvature, slope, shoulder lane width, underpass length, or surface conditions, which might impact the local capacity and could explain the rather low detector-based flows, are unknown to us. While more detailed microscopic calibration using high-resolution trajectory data (e.g., as in Schrader (2024) or Liu et al. (2024)) would be desirable, such data were not available for this study.

4 Methodology to quantify capacity effects of ToCs

To determine ToC-related capacity impacts, we conduct a simulation study with an increasing AV penetration rate and measure the corresponding maximum flows . Additionally, we estimate the anticipated ToC-induced capacity reduction and later compare these estimates with the measured results from the simulation study.

4.1 Simulation experiment

For cases 2–4, we additionally run simulations with -distributions for MVs around and . In case 3, we deploy the heuristic algorithm developed by Lücken et al. (2019). The control algorithm basically emulates a V2X-based traffic management scheme by dispatching ToRs to AVs in a coordinated manner to mitigate the accumulation effect of consecutive ToCs. For case 4, AVs are only inserted into the simulation on the rightmost lane, overtaking is disabled, and their speed is limited to .

To measure the capacity per AV share as precisely as possible, we run the AVUS scenario with 12 seeds per traffic mix, deploying a binary search as illustrated in Algorithm 1. To ensure we obtain the correct maximum flow, the results of each run must be checked against the actual traffic share versus the expected share due to SUMO’s insertion mechanism, as described in Section 3.4. The binary search continues increasing the demand as long as valid traffic shares are observed, until the maximum flow per simulation run is reached.

Simulations run with a warm-up phase to populate the scenario and then record data for a full hour of simulated time (SUMO version 1.22 from Alvarez Lopez et al. (2025)). A detector near the end of the ODD zone records speed and flow to identify potential traffic breakdowns and measures the maximum flow. Figure 9 exemplarily shows spatiotemporal heatmaps of the ODD zone for speed and flow. Figures 9A,B, result from the same demand level and AV share—only the seed values, which determine the randomization process of vehicle insertions, differ.

FIGURE 9

4.2 Estimating capacity reduction

Considering the minGap in SUMO as , individual vehicle lengths , type-specific , and varying vehicle shares , the theoretical lane capacity at speed is given by:

For the SUMO default minGap of , a speed , and the respective vehicle lengths and -values from Table 2, we compute lane capacities across all traffic mixes. Assuming a constant time headway for each vehicle type, without considering ToCs, the resulting capacities per mix are shown in Figure 10a. The results demonstrate that as the -values for MVs increase ( to ), while keeping fixed values for AVs , LGVs , and HGVs , the decline in maximum capacity across traffic mixes becomes less pronounced. If MVs had the same -value as AVs—in this case, — the lane capacities would remain stable, regardless of the increasing AV share.

FIGURE 10

To account for ToC effects in such estimations, we repeat the simplified numerical experiment with the 32-vehicle platoon described in Section 2, this time varying the AV–MV share in 10%-intervals between the two vehicle types. The top panel in Figure 11a shows the time headway profiles for a 100% AV share, corresponding to the acceleration profile discussed in Figure 3b. The increasing headways for later-following AVs are clearly identifiable. In the bottom panel (Figure 11b), which depicts a 50–50 share, the headway increase is far less pronounced compared to the top panel.

FIGURE 11

With these estimator-based -values, the original capacity Equation 1 is adjusted by replacing the fixed headway term in the denominator with , where is the empirically derived time headway for AVs as a function of their share , while other vehicle types retain fixed values. The estimators max and mean provide these AV-specific headways based on the numerical experiments. Figure 10b presents the results of the calculations that account for ToCs by utilizing these estimators. Both trends exhibit a notable decline in estimated lane capacity compared to the ToC-ignorant estimation depicted in Figure 10a.

5 Results and discussion

FIGURE 12

Overall, the results in Figure 12 show that in the unmanaged scenario, maximum flow declines significantly with increasing AV share. At AV85, the max flow is approximately lower than in the managed case—indicating that ToC management measures could help alleviate, but not fully prevent, capacity losses.

FIGURE 13

In summation, the capacity reductions observed in the simulation might align only unsatisfactorily with theoretical estimates, as deviations occur in the mid-range AV shares due to the simplified assumptions of the estimators. This partial mismatch is also reflected in the RMSE and R² values, for which no established benchmarks exist in this context. Therefore, our assessment of estimator performance focuses on relative differences and qualitative trends within the observed results. However, the estimator max performs best in comparison to the simulation results, substantiating our suspicion that the maxima in time headway increments dominate ToC-related capacity effects. Nevertheless, the overall findings from Figures 12, 13 highlight the potential ToC effects on capacity reductions across various scenarios and parameter dependencies, in line with the stated expectations.

6 Conclusion

To investigate ToC-related capacity reductions, we conducted comprehensive simulation experiments with a calibrated two-lane highway scenario, as well as numerical experiments to estimate the large-scale impact of time headway increments during consecutive control transitions. Our main findings can be summarized as follows: (i) capacity reductions of up to , corresponding to approximately loss, were observed in shares with near-full Level 3 automation but no traffic management coordination; (ii) ToC effects became notably impactful starting from a Level 3 share of to ; (iii) a coordination of ToCs could mitigate losses by roughly or ; (iv) binding Level 3 operation to the rightmost lane resulted in the most severe reduction, with up to or loss; and (v) maxima in time headway increments during ToCs emerge as the dominant factor contributing to these capacity effects.

Several relevant limitations should be acknowledged, as they may affect the applicability and interpretation of our findings. Recent research on data from Level 4 AVs reveals reduced time headways when MVs follow AVs (Jiao et al., 2024). Such effects, which might also apply to Level 3 systems, are not considered in this study. An additional aspect that has not yet been discussed is the impact of human response times for non-emergency ToCs. Throughout this investigation, the response time distribution was kept the same, at in all simulations, in line with our previous studies. More recent data from real-world tests presented by Pipkorn et al. (2023) indicate response times closer to 5 s, from which the authors infer that a lead time of , as specified in the R157, should be feasible for human drivers to take over in time. In our simulations, a few random sample reruns with these lower response times indicated approximately 10– higher capacities compared to the results presented here. A further limiting factor might be SUMO’s ACC model, which is parametrized for full string stability and deployed here as a proxy for Level 3 automated vehicles, although experimental studies and theoretical work have demonstrated string instabilities in ACC-equipped platoons, as we also discussed in Alms and Wagner (2024). All these limitations could potentially affect traffic capacity, though their precise contribution cannot be reliably quantified at this stage.

Future work should therefore include the development of a more accurate Level 3 model in SUMO, for example, an ACC-based ALKS system, as well as a systematic investigation of how model assumptions and human response variability together affect traffic capacity. Another important direction is to further examine the effects of MRMs, which are relevant for failed Level 3 transitions and Level 4 automation, on overall traffic, especially if they are not managed properly.

Lastly, we would like to reflect on the broader capacity implications of AVs. Our overall vehicle parametrization inherently results in slightly reduced theoretical capacities—even without ToCs—due to the implementation of lower time headways for MVs and higher ones for AVs, which contrasts with assumptions commonly made in other studies. While experimental research has demonstrated counterbalancing effects at high AV shares, which our own simulations also imply, this effect is diminished in the context of Level 3 systems. Unlike Level 4 or CACC-equipped vehicles, Level 3 automation, in its current form, does not typically support the low time headways often assumed to contribute to capacity gains. However, practical capacity impacts at relevant market penetration rates between 10% and 20% are likely still many years away, leaving room for further technical and regulatory development of Level 3 systems. Yet, in combination with the ToC-related capacity constraints demonstrated in this study, we take a more cautious view and do not share the seemingly widespread optimism regarding beneficial capacity effects of AVs in the mid-term.

References

• 1

  AlmsR.NoulisA.MintsisE.LückenL.WagnerP. (2022). Reinforcement learning-based traffic control: mitigating the adverse impacts of control transitions. IEEE Open J. Intelligent Transp. Syst.3, 187–198. 10.1109/OJITS.2022.3158688

  • CrossRef
  • Google Scholar

• 2

  AlmsR.WagnerP. (2024). Control transitions in level 3 automation: safety implications in mixed-autonomy traffic. Safety10, 1. 10.3390/safety10010001

  • CrossRef
  • Google Scholar

• 3

  Alvarez LopezP.BanseA.BarthauerM.BehrischM.CouéraudB.ErdmannJ.et al (2025). Simulation of urban mobility (SUMO). 10.5281/zenodo.14796685

  • CrossRef
  • Google Scholar

• 4

  Alvarez LopezP.BehrischM.Bieker-WalzL.ErdmannJ.FlötterödY.-P.HilbrichR.et al (2018). “Microscopic traffic simulation using SUMO,” in The 21st IEEE International Conference on Intelligent Transportation systems (IEEE), 2575–2582.

  • Google Scholar

• 5

  BASt (2021). Road Traffic Census Report 2021 - Ergebnisbericht der Straßenverkehrszählung 2021. Tech. rep. Fed. Highw. Res. Inst. (BASt).

  • Google Scholar

• 6

  BASt (2025). Straßenverkehrszählung – verkehrsdatenbank.

  • Google Scholar

• 7

  BolovinouA.AnagnostopoulouC.RoungasV.AmditisA.GonzálezR. B.CoelloL. T.et al (2023). HI-DRIVE Deliverable D3.1/Use cases definition and description. Tech. Rep. Eur. Comm.

  • Google Scholar

• 8

  BrilonW.GeistefeldtJ. (2010). Überprüfung der Bemessungswerte des HBS für Autobahnabschnitte außerhalb der Knotenpunkte, vol. 1033 of Forschung Straßenbau und Straßenverkehrstechnik. Bremerhaven, Germany: Wirtschaftsverl. NW Verl. für neue Wissenschaft.

  • Google Scholar

• 9

  British Standards Institution (2020). PAS 1883:2020 - operational design domain (ODD) taxonomy for automated driving systems (ADS) – specification. London, UK: British Standards Institution.

  • Google Scholar

• 10

  CalvertS.van AremB. (2020). A generic multi-level framework for microscopic traffic simulation with automated vehicles in mixed traffic. Transp. Res. Part C Emerg. Technol.110, 291–311. 10.1016/j.trc.2019.11.019

  • CrossRef
  • Google Scholar

• 11

  Digitale Plattform Stadtverkehr Berlin (2024). Verkehrsdetektion berlin.

  • Google Scholar

• 12

  FGSV (2015). Handbuch für die Bemessung von Straßenverkehrsanlagen: HBS 2015. No. FGSV 299 B in FGSV W1 - Wissensdokumente (Cologne, Germany: FGSV-Verl). 2015 Edn.

  • Google Scholar

• 13

  FriedrichB. (2016). The effect of autonomous vehicles on traffic. Berlin, Heidelberg: Springer Berlin Heidelberg, 317–334. 10.1007/978-3-662-48847-8_16

  • CrossRef
  • Google Scholar

• 14

  GeistefeldtJ.GiulianiS.BuschF.SchendzielorzT.HaugA.VortischP.et al (2017). “HBS-conform simulation of freeway traffic flow,” in vol. 279 of Berichte der Bundesanstalt für Straßen- und Verkehrswesen, Reihe V: Verkehrstechnik (Germany: Federal Highway Research Institute).

  • Google Scholar

• 15

  JiaoY.LiG.CalvertS. C.van CranenburghS.van LintH. (2024). Beyond behavioural change: investigating alternative explanations for shorter time headways when human drivers follow automated vehicles. Transp. Res. Part C Emerg. Technol.164, 104673. 10.1016/j.trc.2024.104673

  • CrossRef
  • Google Scholar

• 16

  KohanpourE.DavoodiS. R.ShaabanK. (2025). Trends in autonomous vehicle performance: a comprehensive study of disengagements and mileage. Future Transp.5, 38. 10.3390/futuretransp5020038

  • CrossRef
  • Google Scholar

• 17

  LiuQ.GaoC.WangH.CaiY.ChenL.LvC. (2024). Learning from trajectories: how heterogeneous CACC platoons affect the traffic flow in highway merging area. IEEE Trans. Veh. Technol.73, 16212–16224. 10.1109/TVT.2024.3419143

  • CrossRef
  • Google Scholar

• 18

  LückenL.MintsisE.PorfyriK.AlmsR.FlötterödY.-P.KoutrasD. (2019). “From automated to manual - modeling control transitions with SUMO,”in SUMO user conference 2019. Editors WeberM.Bieker-WalzL.HilbrichR.BehrischM. (Stockport, United Kingdom: EPiC Series in Computing), 62, 124–144. 10.29007/sfgk

  • CrossRef
  • Google Scholar

• 19

  MaerivoetS.AkkermansL.CarlierK.FlötterödY.-P.LückenL.AlmsR.et al (2019). TransAID Deliverable 4.2 - preliminary simulation and assessment of enhanced traffic management measures. Tech. Rep. Eur. Comm.

  • Google Scholar

• 20

  McDonaldA. D.AlambeigiH.EngströmJ.MarkkulaG.VogelpohlT.DunneJ.et al (2019). Toward computational simulations of behavior during automated driving takeovers: a review of the empirical and modeling literature. Hum. Factors61, 642–688. 10.1177/0018720819829572

  • CrossRef
  • Google Scholar

• 21

  Mercedes-Benz Group (2024). Mercedes-Benz is approved for 95 km/h Level 3 autonomous driving in Germany. Tech. Rep. Mercedes Benz Group.

  • Google Scholar

• 22

  MintsisE.KoutrasD.PorfyriK.MitsakisE.LückenL.ErdmannJ.et al (2019). TransAID Deliverable 3.1 - modelling, simulation and assessment of vehicle automations and automated vehicles’ driver behaviour in mixed traffic. Tech. Rep. Eur. Comm.

  • Google Scholar

• 23

  ParkJ. E.ByunW.KimY.AhnH.ShinD. K. (2021). The impact of automated vehicles on traffic flow and road capacity on urban road networks. J. Adv. Transp.2021, 1–10. 10.1155/2021/8404951

  • CrossRef
  • Google Scholar

• 24

  PipkornL.TivestenE.FlannaganC.DozzaM. (2023). Driver response to take-over requests in real traffic. IEEE Trans. Human Mach. Syst.53, 823–833. 10.1109/THMS.2023.3304003

  • CrossRef
  • Google Scholar

• 25

  RummelJ. (2017). Replication of the hbs autobahn with sumo. Berichte aus dem DLR-Institut für Verkehrssystemtechnik, Berlin, Germany: Institute of Transportation Systems, German Aerospace Center171–178.

  • Google Scholar

• 26

  SAE International (2021). SAE international recommended practice: taxonomy and definitions for terms related to driving automation systems for on-road motor vehicles. SAE Int.10.4271/J3016_202104

  • CrossRef
  • Google Scholar

• 27

  SauvagetJ.-L.DakilM.GriffonT.AnagnostopoulouC.BolovinouA.SintonenH.et al (2023). HI-DRIVE deliverable D5.1/descriptions of “operations”. Tech. Rep. Eur. Comm.

  • Google Scholar

• 28

  SchraderM. (2024). Calibrating traffic microsimulation for optimization of intelligent transportation systems using roadside radar. Tuscaloosa, AL: The University of Alabama. Ph.d. thesis.

  • Google Scholar

• 29

  Schulte-TiggesJ.MatheisD.RekeM.WalterT.KasznerD. (2023). “Demonstrating a V2X enabled system for transition of control and minimum risk manoeuvre when leaving the operational design domain,” in HCI in mobility, transport, and automotive systems. Editor KrömkerH. (Cham: Springer Nature Switzerland), 200–210. 10.1007/978-3-031-35678-0_12

  • CrossRef
  • Google Scholar

• 30

  ShuttleworthJ. (2019). SAE updates J3016 automated-driving graphic. Mobil. Eng.

  • Google Scholar

• 31

  TransAID (2021). Transition areas for infrastructure-assisted driving (TransAID). H2020 Res. Proj. 723390, Eur. Comm.

  • Google Scholar

• 32

  UNECE (2023). Addendum 156 – UN regulation No. 157 - amendment 4 - uniform provisions concerning the approval of vehicles with regard to automated lane keeping systems. Tech. Rep.United Nations Economic commission for Europe.

  • Google Scholar

• 33

  Van AerdeM. (1995). “A single regime speed-flow-density relationship for freeways and arterials,” in 74th annual meeting of the transportation research board. Preprint paper no. 950802.

  • Google Scholar

• 34

  Van LintJ.CalvertS. C. (2018). A generic multi-level framework for microscopic traffic simulation—theory and an example case in modelling driver distraction. Transp. Res. Part B Methodol.117, 63–86. 10.1016/j.trb.2018.08.009

  • CrossRef
  • Google Scholar

• 35

  WagnerP. (2012). Analyzing fluctuations in car-following. Transp. Res. Part B Methodol.46, 1384–1392. 10.1016/j.trb.2012.06.007

  • CrossRef
  • Google Scholar

• 36

  WangC.RenW.XuC.ZhengN.PengC.TongH. (2024). Exploring the impact of conditionally automated driving vehicles transferring control to human drivers on the stability of heterogeneous traffic flow. IEEE Trans. Intel. Veh., 1–17. 10.1109/TIV.2024.3419789

  • CrossRef
  • Google Scholar

• 37

  WangC.XuC.PengC.TongH.RenW.andY. J. (2025a). Predicting the duration of reduced driver performance during the automated driving takeover process. J. Intel. Transp. Syst.29, 218–233. 10.1080/15472450.2024.2307029

  • CrossRef
  • Google Scholar

• 38

  WangC.XuC.ShaoY.ZhengN.PengC.TongH.et al (2025b). Identifying factors affecting driver takeover time and crash risk during the automated driving takeover process. J. Transp. Saf. Secur.0, 1–26. 10.1080/19439962.2025.2450695

  • CrossRef
  • Google Scholar

• 39

  WardT. (2024). “Areas of improvement for autonomous vehicles: a machine learning analysis of disengagement reports,” in 2024 International Conference on Electrics and Computer (INTCEC). ArXiv:2408.00051.

  • Google Scholar

• 40

  XiaoL.WangM.van AremB. (2017). Realistic car-following models for microscopic simulation of adaptive and cooperative adaptive cruise control vehicles. Transp. Res. Rec.2623, 1–9. 10.3141/2623-01

  • CrossRef
  • Google Scholar