pith. sign in

arxiv: 2605.18026 · v1 · pith:VEWY6HWHnew · submitted 2026-05-18 · 💻 cs.RO

Scenario Generation in Roundabouts with Adjustable Interaction Intensity

Li Li , Till Temmen , Tobias Brinkmann , Bj\"orn Krautwig , Markus Eisenbarth , Jakob Andert This is my paper

Pith reviewed 2026-05-20 10:22 UTC · model grok-4.3

classification 💻 cs.RO
keywords scenario generationroundaboutsinteraction intensityyield codeWGANautoencodersautonomous drivingsafety testing
0
0 comments X

The pith

Scaling a compact yield code by factor λ during approach to entry continuously adjusts interaction intensity in WGAN-generated roundabout scenarios.

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

The paper decouples geometric routes and temporal progress profiles in roundabouts and maps them to latent codes with pretrained autoencoders. Conditional generation then uses Wasserstein GANs, with yielding treated as a timing intervention controlled by a compact yield code that is scaled by λ in the approach-to-entry segment. This scaling modulates interaction intensity in a continuous and controlled manner. Experiments show improved timing fidelity over baselines and plausible yielding responses. When the scaling is calibrated to criticality, larger λ values produce expanded safety margins for systematic testing of driving functions.

Core claim

Yielding is modeled as a controllable timing intervention via a compact yield code during the approach-to-entry segment, where interaction intensity is modulated by scaling the code with a factor λ. Geometric routes and temporal progress profiles are first decoupled and mapped to latent codes using pretrained autoencoders, after which conditional latent generation is performed with Wasserstein Generative Adversarial Networks to produce scenarios that exhibit enhanced timing-latent fidelity and plausible interaction responses.

What carries the argument

The compact yield code, scaled by factor λ during the approach-to-entry segment, which intervenes on yielding timing to modulate interaction intensity while staying within the learned latent distribution.

If this is right

  • Higher λ expands the safety margin under criticality-calibrated scaling.
  • The generator produces scenarios with enhanced timing-latent fidelity relative to a baseline model.
  • Plausible interaction responses emerge that support systematic safety analysis.
  • The approach supplies a scalable mechanism for controlled testing of intelligent driving functions in roundabouts.

Where Pith is reading between the lines

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

  • Demand for rare near-critical naturalistic data could decrease if adjustable generators reliably produce desired intensity levels on request.
  • The same yield-code scaling idea could transfer to other merging scenarios such as unsignalized intersections or lane changes.
  • Automated criticality calibration across multiple roundabout geometries would make the intensity control more portable.

Load-bearing premise

That scaling the compact yield code by λ during the approach-to-entry segment produces meaningful and plausible changes in interaction intensity without introducing unrealistic artifacts or breaking the learned distribution.

What would settle it

Generate scenarios across a range of λ values and check whether measured safety margins such as minimum time-to-collision increase monotonically with λ while all trajectories remain kinematically feasible and consistent with observed yielding patterns.

Figures

Figures reproduced from arXiv: 2605.18026 by Bj\"orn Krautwig, Jakob Andert, Li Li, Markus Eisenbarth, Till Temmen, Tobias Brinkmann.

Figure 1
Figure 1. Figure 1: Neuweiler Roundabout in rounD Dataset [29]. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the Three-Stage Scenario Generation [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overlay of Real and Generated routes. TABLE IV: Comparison Between Baseline and Interaction￾aware Model. Metric Baseline Interaction-aware MMD in zt ↓ 0.3997 0.1796 Fréchet distance in zt ↓ 0.1894 0.0020 resulting in 110 points per scatter plot with color indicating λ. V. RESULTS A. Realism Evaluation [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Visualization of Representative Generated Scenarios [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
read the original abstract

Roundabouts, characterized by frequent merging and yielding interactions, remain a safety-critical corner case for the development and testing of intelligent driving functions. However, extracting sufficient near-critical scenarios from naturalistic data is inefficient. Most existing scenario generation methods provide limited controllability over interaction intensity and criticality, making systematic safety testing and detailed analysis difficult. This paper presents an interaction-aware roundabout scenario generator with continuously adjustable interaction intensity. Geometric routes and temporal progress profiles are first decoupled and mapped to latent codes using pretrained autoencoders. Conditional latent generation is then performed with Wasserstein Generative Adversarial Networks (WGAN) to generate scenarios. Yielding is modeled as a controllable timing intervention via a compact yield code during the approach-to-entry segment, where interaction intensity is modulated by scaling the code with a factor $\lambda$. Results demonstrate enhanced timing-latent fidelity and plausible interaction responses compared to a baseline model. Under criticality-calibrated scaling, increasing $\lambda$ expands the safety margin, providing a scalable and controlled testing mechanism.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The paper presents a roundabout scenario generator that decouples geometric routes and temporal progress profiles, encodes them via pretrained autoencoders into latent codes, and uses conditional WGAN generation to produce trajectories. Yielding behavior is modeled as a timing intervention encoded in a compact yield code that is scaled by a factor λ during the approach-to-entry segment to control interaction intensity. The abstract reports enhanced timing-latent fidelity and plausible responses relative to a baseline, and claims that criticality-calibrated increases in λ expand the safety margin to enable scalable testing.

Significance. If the controllability claim holds with demonstrated fidelity, the work would supply a practical mechanism for generating families of near-critical roundabout scenarios with tunable interaction strength. This addresses a recognized gap in systematic safety validation for autonomous driving, where naturalistic data yields too few edge cases. The combination of autoencoders for structured latent spaces and WGAN for conditional generation is technically standard, but the explicit intensity knob via yield-code scaling could be a useful engineering contribution if properly validated.

major comments (2)
  1. [Abstract] Abstract: the central claim that 'increasing λ expands the safety margin' and provides a 'scalable and controlled testing mechanism' rests on the unverified assumption that scaled yield codes remain inside the support of the trained WGAN distribution. No discriminator scores, latent-space distances, collision-rate statistics, or human plausibility ratings are reported for λ ≠ 1, making the downstream safety-margin result impossible to evaluate.
  2. [Methods] Methods (yield-code scaling paragraph): multiplying the compact yield code by λ during the approach-to-entry segment is presented as a direct intensity control, yet the manuscript supplies no analysis showing that the resulting trajectories stay within the learned manifold or avoid non-physical artifacts (e.g., abrupt accelerations or invalid merging geometries). This is load-bearing for the 'plausible interaction responses' assertion.
minor comments (2)
  1. [Abstract] The abstract states 'enhanced timing-latent fidelity' without naming the baseline model or the quantitative metric (e.g., reconstruction MSE, FID, or timing-error distribution) used for comparison.
  2. [Methods] Notation for the yield code and the precise definition of 'criticality-calibrated scaling' should be introduced with an equation or pseudocode in the methods section for reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. The comments correctly identify that stronger empirical support is needed for the controllability claims when scaling the yield code. We address each point below and will incorporate additional validation in the revision.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that 'increasing λ expands the safety margin' and provides a 'scalable and controlled testing mechanism' rests on the unverified assumption that scaled yield codes remain inside the support of the trained WGAN distribution. No discriminator scores, latent-space distances, collision-rate statistics, or human plausibility ratings are reported for λ ≠ 1, making the downstream safety-margin result impossible to evaluate.

    Authors: We agree that the abstract claim would benefit from explicit supporting metrics for λ ≠ 1. The current results section reports timing-latent fidelity and baseline comparisons at λ = 1, but does not include the requested discriminator scores, latent distances, or collision statistics across a range of λ. In the revised manuscript we will add these quantities (WGAN critic scores, nearest-neighbor latent distances, and collision rates) for λ in [0.5, 1.5] together with a short discussion of the range in which the generated trajectories remain plausible. We will also revise the abstract to qualify the safety-margin statement as holding under the reported criticality-calibrated scaling. revision: yes

  2. Referee: [Methods] Methods (yield-code scaling paragraph): multiplying the compact yield code by λ during the approach-to-entry segment is presented as a direct intensity control, yet the manuscript supplies no analysis showing that the resulting trajectories stay within the learned manifold or avoid non-physical artifacts (e.g., abrupt accelerations or invalid merging geometries). This is load-bearing for the 'plausible interaction responses' assertion.

    Authors: We accept that the manuscript currently lacks a dedicated analysis of manifold adherence and artifact avoidance for scaled codes. The Methods section describes the scaling operation but does not quantify acceleration profiles, merging geometry validity, or distance to the training manifold. We will add a short paragraph (or subsection) in the Results section that reports these checks for representative λ values, including example trajectory plots and aggregate statistics on jerk and lane-center deviation. This will directly support the plausibility claim. revision: yes

Circularity Check

0 steps flagged

No significant circularity; controllability via external λ is a design parameter, not a self-derived quantity

full rationale

The paper decouples geometry and timing, maps them to latent codes with pretrained autoencoders, then uses conditional WGAN generation. Interaction intensity is modulated by an externally chosen scalar λ applied to a compact yield code. This scaling is introduced as a modeling choice to achieve controllability; the resulting safety-margin expansion is measured on the generated trajectories rather than being algebraically forced by the definition of λ itself. No equations reduce a claimed prediction to a fitted input by construction, and no load-bearing self-citation or uniqueness theorem is invoked. The pipeline therefore remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim depends on the pretrained autoencoders faithfully separating geometric routes from temporal profiles and on the WGAN learning a latent distribution that remains plausible under external scaling of the yield code.

free parameters (1)
  • λ
    Scaling factor applied to the yield code to modulate interaction intensity; chosen to achieve desired criticality levels.
axioms (2)
  • domain assumption Pretrained autoencoders accurately map geometric routes and temporal progress profiles to latent codes without significant information loss.
    Invoked when decoupling routes and profiles before conditional generation.
  • domain assumption WGAN conditional generation preserves realistic interaction dynamics when the yield code is scaled.
    Required for the claim that increasing λ produces plausible and controllable safety-margin changes.

pith-pipeline@v0.9.0 · 5712 in / 1308 out tokens · 27411 ms · 2026-05-20T10:22:34.528631+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

29 extracted references · 29 canonical work pages

  1. [1]

    Imperfect advanced driver assistance systems in the eyes of imperfect users,

    Y . Chu, W. Tang, S. Chen, and P. Liu, “Imperfect advanced driver assistance systems in the eyes of imperfect users,”Transportation Research Part F: Traffic Psychology and Behaviour, vol. 116, p. 103447, 2026

    work page 2026

  2. [2]

    Driving to safety: How many miles of driving would it take to demonstrate autonomous vehicle reliability?

    N. Kalra and S. M. Paddock, “Driving to safety: How many miles of driving would it take to demonstrate autonomous vehicle reliability?” Transportation Research Part A: Policy and Practice, vol. 94, pp. 182– 193, 2016

    work page 2016

  3. [3]

    Data-driven traffic simulation: A comprehensive review,

    D. Chen, M. Zhu, H. Yang, X. Wang, and Y . Wang, “Data-driven traffic simulation: A comprehensive review,”IEEE Transactions on Intelligent Vehicles, vol. 9, no. 4, pp. 4730–4748, 2024

    work page 2024

  4. [4]

    A survey on safety-critical driving scenario generation—a methodological per- spective,

    W. Ding, C. Xu, M. Arief, H. Lin, B. Li, and D. Zhao, “A survey on safety-critical driving scenario generation—a methodological per- spective,”IEEE Transactions on Intelligent Transportation Systems, vol. 24, no. 7, pp. 6971–6988, 2023

    work page 2023

  5. [5]

    New Car Assess- ment Program Final Decision Notice: Advanced Driver Assistance Systems Roadmap,

    National Highway Traffic Safety Administration, “New Car Assess- ment Program Final Decision Notice: Advanced Driver Assistance Systems Roadmap,” U.S. Department of Transportation, Tech. Rep., Nov. 2024, final Decision Notice (NCAP ADAS Roadmap), published 18 Nov 2024

    work page 2024

  6. [6]

    Is too much system caution counterproductive? effects of varying sensi- tivity and automation levels in vehicle collision avoidance systems,

    E. Fu, M. Johns, D. A. B. Hyde, S. Sibi, M. Fischer, and D. Sirkin, “Is too much system caution counterproductive? effects of varying sensi- tivity and automation levels in vehicle collision avoidance systems,” inProceedings of the 2020 CHI Conference on Human Factors in Computing Systems, ser. CHI ’20. New York, NY , USA: Association for Computing Machin...

    work page 2020

  7. [7]

    The role of traffic conflicts in roundabout safety evaluation: A review,

    L. Li, Z. Zhang, Z.-G. Xu, W.-C. Yang, and Q.-C. Lu, “The role of traffic conflicts in roundabout safety evaluation: A review,”Accident Analysis & Prevention, vol. 196, p. 107430, 2024

    work page 2024

  8. [8]

    Identifying crash pat- terns on roundabouts: An exploratory study,

    E. Polders, S. Daniels, W. Casters, and T. Brijs, “Identifying crash pat- terns on roundabouts: An exploratory study,”Traffic injury prevention, vol. 16, no. 2, pp. 202–207, 2015

    work page 2015

  9. [9]

    Wasserstein generative adversarial networks,

    M. Arjovsky, S. Chintala, and L. Bottou, “Wasserstein generative adversarial networks,” inProceedings of the 34th International Con- ference on Machine Learning, ser. Proceedings of Machine Learning Research, D. Precup and Y . W. Teh, Eds., vol. 70. PMLR, 06–11 Aug 2017, pp. 214–223

    work page 2017

  10. [10]

    Improved training of wasserstein gans,

    I. Gulrajani, F. Ahmed, M. Arjovsky, V . Dumoulin, and A. C. Courville, “Improved training of wasserstein gans,”Advances in neural information processing systems, vol. 30, 2017

    work page 2017

  11. [11]

    Testing scenario library generation for connected and automated vehicles: An adaptive framework,

    S. Feng, Y . Feng, H. Sun, Y . Zhang, and H. X. Liu, “Testing scenario library generation for connected and automated vehicles: An adaptive framework,”IEEE Transactions on Intelligent Transportation Systems, vol. 23, no. 2, pp. 1213–1222, 2022

    work page 2022

  12. [12]

    Intelligent driving intelligence test for autonomous vehicles with naturalistic and adversarial environment,

    S. Feng, X. Yan, H. Sun, Y . Feng, and H. X. Liu, “Intelligent driving intelligence test for autonomous vehicles with naturalistic and adversarial environment,”Nature Communications, vol. 12, no. 1, pp. 1–14, 2021

    work page 2021

  13. [13]

    Curse of rarity for autonomous vehicles,

    H. X. Liu and S. Feng, “Curse of rarity for autonomous vehicles,” Nature Communications, vol. 15, no. 1, p. 4808, 2024

    work page 2024

  14. [14]

    Scenario parameter generation method and scenario representativeness metric for scenario-based assessment of automated vehicles,

    E. de Gelder, J. Hof, E. Cator, J.-P. Paardekooper, O. O. d. Camp, J. Ploeg, and B. de Schutter, “Scenario parameter generation method and scenario representativeness metric for scenario-based assessment of automated vehicles,”IEEE Transactions on Intelligent Transporta- tion Systems, vol. 23, no. 10, pp. 18 794–18 807, 2022

    work page 2022

  15. [15]

    Multi- agent scenario generation in roundabouts with a transformer-enhanced conditional variational autoencoder,

    L. Li, T. Brinkmann, T. Temmen, M. Eisenbarth, and J. Andert, “Multi- agent scenario generation in roundabouts with a transformer-enhanced conditional variational autoencoder,”arXiv preprint arXiv:2510.24671, 2025

    work page arXiv 2025

  16. [16]

    Data generation for connected and automated vehicle tests using deep learning models,

    Y . Li, F. Liu, L. Xing, Y . He, C. Dong, C. Yuan, J. Chen, and L. Tong, “Data generation for connected and automated vehicle tests using deep learning models,”Accident Analysis & Prevention, vol. 190, p. 107192, 2023

    work page 2023

  17. [17]

    Gener- ating intersection pre-crash trajectories for autonomous driving safety testing using transformer time-series generative adversarial networks,

    X. Liu, H. Huang, J. Bian, R. Zhou, Z. Wei, and H. Zhou, “Gener- ating intersection pre-crash trajectories for autonomous driving safety testing using transformer time-series generative adversarial networks,” Engineering Applications of Artificial Intelligence, vol. 160, p. 111995, 2025

    work page 2025

  18. [18]

    Diffscene: diffusion- based safety-critical scenario generation for autonomous vehicles,

    C. Xu, A. Petiushko, D. Zhao, and B. Li, “Diffscene: diffusion- based safety-critical scenario generation for autonomous vehicles,” inProceedings of the Thirty-Ninth AAAI Conference on Artificial Intelligence and Thirty-Seventh Conference on Innovative Applications of Artificial Intelligence and Fifteenth Symposium on Educational Advances in Artificial In...

    work page 2025

  19. [19]

    Evolving testing scenario generation and intelligence evaluation for automated vehicles,

    Y . Ma, W. Jiang, L. Zhang, J. Chen, H. Wang, C. Lv, X. Wang, and L. Xiong, “Evolving testing scenario generation and intelligence evaluation for automated vehicles,”Transportation Research Part C: Emerging Technologies, vol. 163, p. 104620, 2024

    work page 2024

  20. [20]

    Deep generative modelling: A comparative review of vaes, gans, normalizing flows, energy-based and autoregressive models,

    S. Bond-Taylor, A. Leach, Y . Long, and C. G. Willcocks, “Deep generative modelling: A comparative review of vaes, gans, normalizing flows, energy-based and autoregressive models,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44, no. 11, pp. 7327– 7347, 2022

    work page 2022

  21. [21]

    A survey on generative diffusion models,

    H. Cao, C. Tan, Z. Gao, Y . Xu, G. Chen, P.-A. Heng, and S. Z. Li, “A survey on generative diffusion models,”IEEE Transactions on Knowledge and Data Engineering, vol. 36, no. 7, pp. 2814–2830, 2024

    work page 2024

  22. [22]

    Near miss determination through use of a scale of danger,

    J. C. Hayward, “Near miss determination through use of a scale of danger,”Highway Research Record, 1972

    work page 1972

  23. [23]

    Can post encroach- ment time substitute intersection characteristics in crash prediction models?

    L. N. Peesapati, M. P. Hunter, and M. O. Rodgers, “Can post encroach- ment time substitute intersection characteristics in crash prediction models?”Journal of Safety Research, vol. 66, pp. 205–211, 2018

    work page 2018

  24. [24]

    Surrogate safety measures from traffic simulation models a comparison of different models for intersection safety evaluation,

    V . Astarita, D. C. Festa, V . P. Giofrè, and G. Guido, “Surrogate safety measures from traffic simulation models a comparison of different models for intersection safety evaluation,”Transportation research procedia, vol. 37, pp. 219–226, 2019

    work page 2019

  25. [25]

    Gap acceptance behaviour and crash risks of mobile phone distracted young drivers at roundabouts: A random parameters survival model,

    E. Memeh, Y . Ali, F. Javier Rubio, C. Hancock, and M. Mazharul Haque, “Gap acceptance behaviour and crash risks of mobile phone distracted young drivers at roundabouts: A random parameters survival model,”Accident Analysis & Prevention, vol. 206, p. 107720, 2024

    work page 2024

  26. [26]

    Accelerated evaluation of automated vehicles in car-following maneuvers,

    D. Zhao, X. Huang, H. Peng, H. Lam, and D. J. LeBlanc, “Accelerated evaluation of automated vehicles in car-following maneuvers,”IEEE Transactions on Intelligent Transportation Systems, vol. 19, no. 3, pp. 733–744, 2017

    work page 2017

  27. [27]

    Adaptive design of experiments for safety evaluation of automated vehicles,

    J. Sun, H. Zhou, H. Xi, H. Zhang, and Y . Tian, “Adaptive design of experiments for safety evaluation of automated vehicles,”IEEE Transactions on Intelligent Transportation Systems, vol. 23, no. 9, pp. 14 497–14 508, 2022

    work page 2022

  28. [28]

    Hazardous scenario enhanced generation for automated vehicle testing based on optimization search- ing method,

    B. Zhu, P. Zhang, J. Zhao, and W. Deng, “Hazardous scenario enhanced generation for automated vehicle testing based on optimization search- ing method,”IEEE Transactions on Intelligent Transportation Systems, vol. 23, no. 7, pp. 7321–7331, 2022

    work page 2022

  29. [29]

    The round dataset: A drone dataset of road user trajectories at roundabouts in germany,

    R. Krajewski, T. Moers, J. Bock, L. Vater, and L. Eckstein, “The round dataset: A drone dataset of road user trajectories at roundabouts in germany,” in2020 IEEE 23rd International Conference on Intelligent Transportation Systems (ITSC), 2020, pp. 1–6

    work page 2020