Pith. sign in

REVIEW 5 major objections 5 minor 29 references

Generalized Coordination of Partially Cooperative Urban Traffic

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A single roadside planner can improve urban traffic flow at intersections, roundabouts, and road narrowings even when only 40 percent of vehicles are connected and cooperative.

desk verdict A solid engineering extension of an intersection-only planner to roundabouts and road narrowing, with honest but statistically thin simulation evidence that the 40%-CAV efficiency gains are real. read the letter →

arxiv 2505.20879 v1 pith:THXCGGEX submitted 2025-05-27 cs.RO cs.MA

Max Bastian Mertens , Michael Buchholz This is my paper
classification cs.ROcs.MA
keywords cooperativemaneuverplanningconnectedautomatedvehiclesautomaticintersectionmanagementmixedtrafficV2Xcommunicationroundaboutcoordinationspace-timeconstraintsurbanefficiency
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

This paper argues that a roadside coordinator can make urban traffic more efficient even when most vehicles are ordinary human-driven cars. The planner does not steer human drivers; it only decides the order in which cooperative connected vehicles cross each shared zone, then sends each one a time window to enter and leave. The authors test this at five intersections, seven roundabouts, and a road narrowing, using simulations built on real drone-observed traffic. They report that with 40 percent of vehicles cooperating, average waiting time falls by about 13 percent and throughput rises by about 3 percent, with no increase in critical close encounters. If the simulated human-driver model is faithful, this means an infrastructure-based coordinator could deliver measurable congestion relief at a minority-penetration stage of connected automated vehicles.

What carries the argument

The load-bearing object is the priority set: a collection of ordered pairs of conflicting connected automated vehicles, where the first vehicle is given right of way over the second. The planner generates candidate priority sets by adding, removing, or reversing pairs relative to the previous best solution, predicts the next 12 seconds of traffic for each candidate using learned gap-acceptance and acceleration models, and scores every candidate with a time-loss metric that integrates each vehicle's relative velocity loss over the horizon. The winning set is then converted into space-time constraints, so that vehicles from different directions never occupy a shared conflict zone simultaneously, and communicated to the connected vehicles for execution.

What would settle it

Run the same planner against real or high-fidelity closed-loop tests at one four-leg intersection and one roundabout, with 40 percent connected vehicles and the rest driven by human drivers who are not coordinated, then measure mean waiting time, throughput, and the rate of post-encroachment times below 1 second; if waiting time reduction is well below 13 percent or critical encounters increase, the simulation's driver model was too optimistic.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes that a centralized, priority-based maneuver planner generalizes across a class of intersection-like urban scenes—priority intersections, right-before-left intersections, roundabouts, and a road narrowing—and works in mixed traffic at any ratio of human-driven to connected automated vehicles. The planner represents a maneuver as a set of priority pairs between conflicting connected vehicles, predicts the scene 12 seconds ahead under each candidate set using learned behavior models for human drivers, and selects the set that minimizes a time-loss metric. The chosen priorities are converted to longitudinal space-time constraints, a time window to enter and a time window to leave each conflict zone, and transmitted to the connected vehicles. In simulation, the optimization planner reduces mean waiting time to 87 percent of the non-cooperative baseline at 40 percent connected vehicles, raises throughput by about 3 percent, and leaves the share of critical post-encroachment times (under 1 second) unchanged or slightly lower; at roundabouts it reports throughput gains up to 30.1 percent and waiting-time reductions by a factor of five. A fast heuristic that orders connected vehicles by a learned urgency score stays under 50 milliseconds and nearly matches the optimization results, which the paper presents as evidence that real-time operation at high traffic load is feasible.

Load-bearing premise

The measured efficiency and safety gains are obtained entirely inside a simulated traffic world whose human-driver behavior model was trained on real intersections and roundabouts; if those simulated human drivers are more predictable or more yielding than real ones, the reported gains would not transfer to the field.

Editorial extensions

If this is right

  • At 40 percent connected vehicles, average waiting time drops to 87 percent of the non-cooperative baseline and throughput rises to 103 percent, while the share of critical encounters does not increase.
  • At roundabouts, the planner reports the largest gains: throughput up to 30.1 percent higher and waiting time reduced by a factor of five relative to no coordination.
  • A heuristic based on a learned urgency score runs in under 50 milliseconds and stays close to the full optimization, supporting real-time use in high-load scenes.
  • Even simpler baselines—FIFO ordering and merely telling connected vehicles which other connected vehicles are non-conflicting—produce measurable efficiency gains, implying that route knowledge alone explains part of the benefit.
  • At a road narrowing, reordering changes which lane waits rather than total flow, cutting mean waiting time about 10 percent while slightly lowering throughput, showing that the planner deliberately trades off performance metrics.

Reading between the lines

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

  • If the 40-percent result transfers to real roads, an infrastructure-based coordinator could be rolled out before full automation, but the paper does not test how human drivers respond when they know a coordinator is managing the scene.
  • Because the planner only constrains longitudinal timing windows, the same priority-set machinery could advise human drivers through a connected dashboard or route guidance, not just automated vehicles; the paper only evaluates automated vehicles.
  • The learned urgency heuristic is trained on a random mix of scenario types, so an implicit assumption is that deployment resembles that mix; a testable extension would train on only one scenario class and measure transfer to the others.
  • Roundabouts show the largest gains, which suggests that deployment pilots might prioritize roundabouts over signalized intersections, since they are common, unsignalized, and already flow efficiently without lights.
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.

Referee Report

5 major / 5 minor

Summary. The paper proposes a centralized cooperative maneuver planning framework for urban traffic scenes with mixed human-driven and connected automated vehicles. The approach generalizes a prior intersection-only method to a class of 'intersection-like' scenarios including roundabouts and a road narrowing. Two priority-set generation methods are presented: an optimization-based method (OPT) and a learned heuristic (HEUR), with FIFO and no-coordination baselines. The framework is evaluated in the DeepSIL simulation environment across 13 real-world-derived scenarios. The authors report efficiency gains starting at 40% CAV penetration, such as a 13% reduction in average waiting time and a 3% throughput increase, and claim no negative safety impact based on post-encroachment time analysis.

Significance. If the reported results are valid, the paper makes a useful contribution by extending cooperative maneuver planning beyond isolated fully-automated intersections to partially cooperative mixed traffic at multiple urban conflict geometries, and by providing a real-time-capable heuristic. The evaluation is extensive in scope, using real-world map data, a V2X protocol, and a CAV trajectory planning stack, and the authors are careful to separate the optimization and heuristic methods. However, the central quantitative claims currently rest on a self-built simulation loop without external validation or inferential statistics, and the safety claim is based on very sparse critical-event counts. These issues are fixable, but they are load-bearing for the paper's headline conclusions.

major comments (5)
  1. [Section IV-D, Tables II-III] The headline claim of 'significant efficiency gains already at a cooperation rate of 40%' is not supported by any inferential statistics. No error bars, confidence intervals, or significance tests are reported for any metric in Figs. 4-7 or Tables II-III. The reported 0.87x waiting-time ratio (a 13% reduction) and 1.03x throughput at 40% CAVs are modest relative to the scenario-to-scenario variance in the base values (waiting-time bases range from 2.06 s to 7.80 s across scenario types), so the 40% threshold cannot currently be distinguished from scenario noise.
  2. [Section IV-B-1 and Section III-D-2] The efficiency and safety claims are evaluated entirely inside the DeepSIL simulation framework, with HDV behavior generated by the MLP_acc and MLP_gap models trained on inD and openDD data and fine-tuned via PPO and GAIL on the 13 scenarios. The reported prediction accuracy (84.4%/77.8% correct crossing order, 16.5/20.9 m position RMSE at 10 s, Fig. 3) shows substantial error, and there is no evidence that the HDV model remains realistic when human drivers face the unusual priority impositions created by coordinated CAVs. If simulated HDVs yield more readily than real drivers, the 0.87x waiting-time ratio at 40% CAVs (Table II) would be a simulation artifact. Please provide external validation (for example, real-world test drives as in [4]) or, at minimum, a sensitivity analysis over HDV model variants or parameters.
  3. [Section IV-A] The exclusion of approximately 1.1% of simulation runs due to software crashes or timeouts, described as 'mostly unrelated to maneuver planning,' is not accompanied by a breakdown by planning method, CAV percentage, or scenario type, nor by a sensitivity analysis. If excluded runs correlate with the planning method (for example, OPT exceeds the 200 ms cycle time in 13.3% of scenarios, Section IV-D-3) or with CAV percentage, the averaged metrics in Figs. 4-7 and Tables II-III are biased. Please report the excluded runs by condition and rerun the analysis with conservative bounds or imputation.
  4. [Section IV-D-2, Tables II-III] The safety claim 'without impacting traffic safety' rests on critical PET rates of only 0.1%-0.3% of encounters. These are extremely sparse counts, no confidence intervals are provided, and the 1 s critical PET threshold is taken from [29], which was developed for right-turn related crashes on intercity highways rather than for the full set of intersection, roundabout, and narrowing scenarios studied here. A sensitivity analysis over PET thresholds and a statistical comparison of critical-encounter rates across methods are needed before the no-safety-impact claim can be considered supported.
  5. [Fig. 7 and Section IV-A] The road-narrowing results, including the claim that HEUR and OPT reduce waiting time by approximately 10% by alternating priorities, are based on only 6 simulated scenarios. With no error bars or significance tests, this scenario type is too thinly sampled to support the paper's generalization claim across intersection-like scenarios. Please either increase the number of road-narrowing scenarios or clearly label these results as preliminary.
minor comments (5)
  1. [Abstract] The abstract contains a typo: 'planer' should be 'planner.'
  2. [Section III-A-3] The scenario list states 'five intersections with main and side roads, one intersection with right-before-left regulation, seven roundabouts, and a road narrowing,' which sums to 14, but the surrounding text says 13 scenarios. Section IV-A lists 13 scenarios (4 inD intersections, 7 openDD roundabouts, 1 additional intersection, 1 road narrowing). Please reconcile this inconsistency.
  3. [Section III-E-1, Eq. (8)] The notation O(P) is used in Eq. (8) before it is defined; please define the predicted crossing-order set before presenting the switching-cost expression.
  4. [Section IV-B-2] The vehicle reinsertion mechanism (removal at 20 m after the scene and reinsertion at 45 m before the first conflict zone at up to 30 km/h, with a maximum of 10 vehicles) is a strong modeling assumption that should be justified or tested, as it may inflate throughput and affect the comparison between planning methods.
  5. [Section IV-D-1] The text states that at roundabouts cooperation can 'reduce waiting time by a factor of five,' but Fig. 6's y-axis starts at 0.5x, so a fivefold reduction is not visible in the plotted range; please verify the reported factor against the data shown.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the planner's optimization objective is independent of the evaluation metrics, and the prediction models are trained on external datasets.

full rationale

The paper's derivation chain does not reduce to its own inputs. The planning objective L(Pk) in Eq. (2) is a velocity-loss integral, while the reported efficiency metrics (waiting time, throughput, stopped vehicles) and the criticality metric (PET) are defined independently in Section IV-C. The HDV prediction models MLP_acc and MLP_gap are trained on the external inD and openDD datasets and compared against the IDM baseline, providing out-of-sample evidence for their realism. The HEUR urgency MLP is trained on labels generated from the same objective L, but it is explicitly presented as a learned heuristic, not as an independent prediction; the central 40% CAV efficiency claim is also supported by the optimization method OPT, which does not rely on that training. Self-citations to DeepSIL [5], the CAV trajectory planner [6], and the authors' prior prediction work [3] supply simulation components, but none is invoked as a uniqueness theorem or as an unverified load-bearing premise. The exclusion of roughly 1.1% of runs and the sparse critical-PET statistics are external-validity and robustness concerns, not circularity.

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

The headline result is an empirical simulation claim. Its validity rests on the realism of the simulation framework (a domain assumption), on hand-chosen weights and thresholds that shape the trade-offs, and on standard vehicle-dynamics assumptions. No new physical entities are introduced; the heuristic MLP is a trained component, not an invented entity.

free parameters (5)
  • waiting time weighting coefficient = 1/10 s
    Eq. (9) sets w_i = 1 + t_slow,i/10s, a hand-chosen trade-off between preventing starvation and minimizing total time loss; it directly affects which priority set is selected.
  • switching cost coefficient = 1 s
    Eq. (8) penalizes each changed vehicle pair order by 1 s of equivalent time loss; hand-chosen to reduce maneuver oscillations.
  • critical PET threshold = 1 s
    Section IV-C-2 defines encounters with PET under 1 s as critical, following [29]; the safety claim depends on this threshold.
  • candidate priority set limit = 100
    Section III-C-1 caps the number of predicted scenarios at 100 to keep runtime near 200 ms; this makes OPT a local search, not a global optimizer.
  • simulation reinsertion parameters = 20 m after scene, 45 m before conflict zone, 30 km/h
    Section IV-B-2 reinserts vehicles to simulate dense traffic; these choices affect throughput and waiting-time measurements.
assumptions (5)
  • domain assumption Space-time constraints with non-overlapping intervals guarantee collision-free conflict zones
    Used in Section III-A-1 to translate priority orders into maneuvers; standard in AIM but relies on vehicles complying with constraints.
  • domain assumption Infrastructure perception provides a complete and accurate environment model over a 100 m diameter
    Section III-A-1 assumes this availability, referencing [1]; the planner's performance depends on accurate state inputs.
  • domain assumption HDV behavior is adequately captured by the DeepSIL simulation model and the prediction MLPs trained on inD and openDD
    Section IV-B-1 and III-D; if the HDV models are more predictable or more cautious than real drivers, the measured efficiency gains are optimistic.
  • domain assumption The most conflicting route assumption for HDVs is conservative yet still permits efficiency gains
    Section III-B states HDVs assume the most conflicting route; this is a safety-oriented worst-case assumption that could reduce attainable gains.
  • standard math Time-loss integral and vehicle dynamics integrability
    Eq. (2) assumes the velocity loss integral is well-defined over the prediction horizon; standard calculus.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Generalized Coordination of Partially Cooperative Urban Traffic." pith.science (2026). https://pith.science/paper/THXCGGEX

@misc{pith2026250520879,
  author       = {Pith},
  title        = {Pith review of: Generalized Coordination of Partially Cooperative Urban Traffic},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/THXCGGEX}},
  note         = {Machine review of arXiv:2505.20879}
}
read the original abstract

Vehicle-to-anything connectivity, especially for autonomous vehicles, promises to increase passenger comfort and safety of road traffic, for example, by sharing perception and driving intention. Cooperative maneuver planning uses connectivity to enhance traffic efficiency, which has, so far, been mainly considered for automated intersection management. In this article, we present a novel cooperative maneuver planning approach that is generalized to various situations found in urban traffic. Our framework handles challenging mixed traffic, that is, traffic comprising both cooperative connected vehicles and other vehicles at any distribution. Our solution is based on an optimization approach accompanied by an efficient heuristic method for high-load scenarios. We extensively evaluate the proposed planer in a distinctly realistic simulation framework and show significant efficiency gains already at a cooperation rate of 40%. Traffic throughput increases, while the average waiting time and the number of stopped vehicles are reduced, without impacting traffic safety.

Figures

Figures reproduced from arXiv: 2505.20879 by the authors.

Figure 1
Figure 1. Four of the 13 simulated real-world scenarios. They comprise five [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. System overview. Parts handled in this work are marked in red, [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Prediction accuracy per vehicle of our MLPacc vs. IDM on the inD and openDD datasets. Our gap acceptance model MLPgap was used for both. The driver model MLPacc is pre-trained using Proximal Policy Optimization [27] in a multi-agent closed-loop rein￾forcement learning environment on each of the 13 considered scenarios. The manually designed objective function rewards driving close to the speed limit and avoiding col… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Results of 135 simulations at four intersections with a main road. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Results of 43 simulated scenarios at one right-before-left intersection. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Results of 139 simulated scenarios at seven roundabouts. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Results of 6 simulated scenarios at a road narrowing. [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

29 extracted references · 27 canonical work pages

  1. [22]

    Optimal Control of Connected and Automated Vehicles at Roundabouts: An Investigation in a Mixed-Traffic Environment *,

    L. Zhao, A. Malikopoulos, and J. Rios-Torres, “Optimal Control of Connected and Automated Vehicles at Roundabouts: An Investigation in a Mixed-Traffic Environment *,”IFAC-PapersOnLine, vol. 51, no. 9, pp. 73–78, 2018

    work page 2018

  2. [4]

    Real-World Evaluation of two Cooperative Intersection Management Approaches,

    M. Klimke, M. B. Mertens, B. V ¨olz, and M. Buchholz, “Real-World Evaluation of two Cooperative Intersection Management Approaches,” 2025, arXiv:2403.16478

    work page arXiv 2025

  3. [29]

    Post encroachment time threshold identification for right-turn related crashes at unsignalized intersections on intercity highways under mixed traffic,

    M. Paul and I. Ghosh, “Post encroachment time threshold identification for right-turn related crashes at unsignalized intersections on intercity highways under mixed traffic,”Int. Journal of Injury Control and Safety Promotion, vol. 27, no. 2, pp. 121–135, 2020. Max Bastian Mertensearned his M.Sc. degree in Communications and Computer Engineering at Ulm U...

    work page 2020

  4. [1]

    Handling Occlusions in Automated Driving Using a Multiaccess Edge Computing Server-Based Environment Model From Infrastructure Sensors,

    M. Buchholz, J. C. M ¨uller, M. Herrmann, J. Strohbeck, B. V ¨olz, M. Maier, J. Paczia, O. Stein, H. Rehborn, and R.-W. Henn, “Handling Occlusions in Automated Driving Using a Multiaccess Edge Computing Server-Based Environment Model From Infrastructure Sensors,”IEEE Intell. Transp. Syst. Magazine, pp. 2–16, 2021

    work page 2021

  5. [2]

    Cooperative Maneuver Planning for Mixed Traffic at Unsignalized Intersections Using Prob- abilistic Predictions,

    M. B. Mertens, J. M ¨uller, and M. Buchholz, “Cooperative Maneuver Planning for Mixed Traffic at Unsignalized Intersections Using Prob- abilistic Predictions,” in2022 IEEE Intell. Veh. Symp. (IV), 2022, pp. 1174–1180

    work page 2022

  6. [3]

    Fast Long- Term Multi-Scenario Prediction for Maneuver Planning at Unsignalized Intersections,

    M. B. Mertens, J. Ruof, J. Strohbeck, and M. Buchholz, “Fast Long- Term Multi-Scenario Prediction for Maneuver Planning at Unsignalized Intersections,” in2024 American Control Conf. (ACC), 2024, pp. 111– 116

    work page 2024

  7. [5]

    DeepSIL: A Software-in-the-Loop Framework for Evaluating Motion Planning Schemes Using Multiple Trajectory Prediction Networks *,

    J. Strohbeck, J. M ¨uller, A. Holzbock, and M. Buchholz, “DeepSIL: A Software-in-the-Loop Framework for Evaluating Motion Planning Schemes Using Multiple Trajectory Prediction Networks *,” in2021 IEEE/RSJ Int. Conf. on Intell. Robots and Syst. (IROS), 2021, pp. 7075– 7081

    work page 2021

  8. [6]

    Real-Time Spatial Trajectory Planning for Urban Environments Using Dynamic Optimization,

    J. Ruof, M. B. Mertens, M. Buchholz, and K. Dietmayer, “Real-Time Spatial Trajectory Planning for Urban Environments Using Dynamic Optimization,” in2023 IEEE Intell. Veh. Symp. (IV), 2023

    work page 2023

Show all 29 references
  1. [7]

    An extended maneuver coordination protocol with support for urban scenarios and mixed traffic,

    M. B. Mertens, J. M ¨uller, R. Dehler, M. Klimke, M. Maier, S. Gherekhloo, B. V ¨olz, R.-W. Henn, and M. Buchholz, “An extended maneuver coordination protocol with support for urban scenarios and mixed traffic,” in2021 IEEE Veh. Netw. Conf. (VNC), vol. 2021, 2021, pp. 32–35

    2021

  2. [8]

    Multiagent traffic management: a reservation- based intersection control mechanism,

    K. Dresner and P. Stone, “Multiagent traffic management: a reservation- based intersection control mechanism,” inProc. of the Third Int. Joint Conf. on Autonomous Agents and Multiagent Syst., 2004. AAMAS 2004., 2004, pp. 530–537

    2004

  3. [9]

    Autonomous and Semiautonomous Intersection Management: A Survey,

    Z. Zhong, M. Nejad, and E. E. Lee, “Autonomous and Semiautonomous Intersection Management: A Survey,”IEEE Intell. Transp. Syst. Maga- zine, vol. 13, no. 2, pp. 53–70, 2021

    2021

  4. [10]

    Cooperative Driving at Blind Crossings Using Intervehicle Communication,

    L. Li and F.-Y . Wang, “Cooperative Driving at Blind Crossings Using Intervehicle Communication,”IEEE Transactions on Veh. Technol., vol. 55, no. 6, pp. 1712–1724, 2006

    2006

  5. [11]

    Cooperative driving: an ant colony system for autonomous intersection management,

    J. Wu, A. Abbas-Turki, and A. El Moudni, “Cooperative driving: an ant colony system for autonomous intersection management,”Appl Intell, vol. 37, no. 2, pp. 207–222, 2012

    2012

  6. [12]

    Vehicle to infrastructure based safe trajectory planning for Autonomous Intersection Management,

    C. Wuthishuwong and A. Traechtler, “Vehicle to infrastructure based safe trajectory planning for Autonomous Intersection Management,” in 2013 13th Int. Conf. on ITS Telecommun. (ITST), 2013, pp. 175–180

    2013

  7. [13]

    Modeling and controlling an isolated urban intersection based on cooperative vehicles,

    M. Ahmane, A. Abbas-Turki, F. Perronnet, J. Wu, A. E. Moudni, J. Buisson, and R. Zeo, “Modeling and controlling an isolated urban intersection based on cooperative vehicles,”Transp. Research Part C: Emerging Technologies, vol. 28, pp. 44–62, 2013

    2013

  8. [14]

    Decentralized Cooperative Planning for Automated Vehicles with Hierarchical Monte Carlo Tree Search,

    K. Kurzer, C. Zhou, and J. Marius Z ¨ollner, “Decentralized Cooperative Planning for Automated Vehicles with Hierarchical Monte Carlo Tree Search,” in2018 IEEE Intell. Veh. Symp. (IV), 2018, pp. 529–536

    2018

  9. [15]

    Developing an Optimal Intersection Control System for Automated Connected Vehicles,

    Y . Bichiou and H. A. Rakha, “Developing an Optimal Intersection Control System for Automated Connected Vehicles,”IEEE Transactions on Intell. Transp. Syst., vol. 20, no. 5, pp. 1908–1916, 2019

    1908

  10. [16]

    Traffic management with autonomous and connected vehicles at single-lane roundabouts,

    M. Martin-Gasulla and L. Elefteriadou, “Traffic management with autonomous and connected vehicles at single-lane roundabouts,”Transp. Research Part C: Emerging Technologies, vol. 125, p. 102964, 2021

    2021

  11. [17]

    Intelligent traffic management at intersections supported by V2V and V2I communications,

    L. C. Bento, R. Parafita, and U. Nunes, “Intelligent traffic management at intersections supported by V2V and V2I communications,” in2012 15th Int. IEEE Conf. on Intell. Transp. Syst., 2012, pp. 1495–1502

    2012

  12. [18]

    Sharing the Road: Autonomous Vehicles Meet Human Drivers,

    K. Dresner and P. Stone, “Sharing the Road: Autonomous Vehicles Meet Human Drivers,” inIJCAI, 2007

    2007

  13. [19]

    Intelligent traffic management at intersections: Legacy mode for vehicles not equipped with V2V and V2I communications,

    L. C. Bento, R. Parafita, S. Santos, and U. Nunes, “Intelligent traffic management at intersections: Legacy mode for vehicles not equipped with V2V and V2I communications,” in16th Int. IEEE Conf. on Intell. Transp. Syst. (ITSC 2013), 2013, pp. 726–731

    2013

  14. [20]

    Priority-based coordination of autonomous and legacy vehicles at intersection,

    X. Qian, J. Gregoire, F. Moutarde, and A. De La Fortelle, “Priority-based coordination of autonomous and legacy vehicles at intersection,” in17th Int. IEEE Conf. on Intell. Transp. Syst. (ITSC), 2014, pp. 1166–1171

    2014

  15. [21]

    Isolated intersection control for various levels of vehicle technology: Conventional, connected, and automated vehicles,

    K. Yang, S. I. Guler, and M. Menendez, “Isolated intersection control for various levels of vehicle technology: Conventional, connected, and automated vehicles,”Transp. Research Part C: Emerging Technologies, vol. 72, pp. 109–129, 2016

    2016

  16. [23]

    Space Time Reservation Procedure (STRP) for V2X-Based Maneuver Coordination of Cooperative Automated Vehicles in Diverse Conflict Scenarios,

    M. Nichting, D. Heß, J. Schindler, T. Hesse, and F. K ¨oster, “Space Time Reservation Procedure (STRP) for V2X-Based Maneuver Coordination of Cooperative Automated Vehicles in Diverse Conflict Scenarios,” in 2020 IEEE Intell. Veh. Symp. (IV), 2020, pp. 502–509

    2020

  17. [24]

    Automatic Intersection Manage- ment in Mixed Traffic Using Reinforcement Learning and Graph Neural Networks,

    M. Klimke, B. V ¨olz, and M. Buchholz, “Automatic Intersection Manage- ment in Mixed Traffic Using Reinforcement Learning and Graph Neural Networks,” in2023 IEEE Intell. Veh. Symp. (IV), 2023, pp. 1–8

    2023

  18. [25]

    The inD Dataset: A Drone Dataset of Naturalistic Road User Trajec- tories at German Intersections,

    J. Bock, R. Krajewski, T. Moers, S. Runde, L. Vater, and L. Eckstein, “The inD Dataset: A Drone Dataset of Naturalistic Road User Trajec- tories at German Intersections,” in2020 IEEE Intell. Veh. Symp. (IV), 2020, pp. 1929–1934

    2020

  19. [26]

    openDD: A Large-Scale Roundabout Drone Dataset,

    A. Breuer, J.-A. Term ¨ohlen, S. Homoceanu, and T. Fingscheidt, “openDD: A Large-Scale Roundabout Drone Dataset,” 2020, arXiv:2007.08463

    2020 arXiv

  20. [27]

    Proximal Policy Optimization Algorithms,

    J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov, “Proximal Policy Optimization Algorithms,” 2017, arXiv:1707.06347

    2017 arXiv

  21. [28]

    Generative Adversarial Imitation Learning,

    J. Ho and S. Ermon, “Generative Adversarial Imitation Learning,” 2016, arXiv:1606.03476

    2016 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.