Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

RADE: Learning Risk-Adjustable Driving Environment via Multi-Agent Conditional Diffusion

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read RADE claims that a PET-conditioned multi-agent diffusion model can generate statistically realistic traffic scenes whose crash rate rises on demand.

desk verdict Useful risk dial for AV simulation, but the high-risk setting may be extrapolation rather than learned behavior; needs ablations, baselines, and error bars before I'd trust it. read the letter →

arxiv 2505.03178 v1 pith:GX56MJTJ submitted 2025-05-06 cs.LG cs.RO

classification cs.LGcs.RO
keywords risk-adjustabletrafficsimulationmulti-agentdiffusionconditionalgenerationPost-EncroachmentTimesafety-criticalscenarioautonomousvehicletestingtokenizeddynamicscheckstatisticalrealism
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

The paper sets out to turn traffic-scenario generation for autonomous-vehicle testing into a dial: rather than scripting a single adversarial vehicle, it trains a multi-agent diffusion model on real driving data to generate the joint trajectories of all vehicles conditional on a requested risk level. Its central claim is that this risk-conditioned generation preserves statistical realism—distance, speed, and yielding distributions stay close to the observed roundabout data—while monotonically increasing the frequency of near-miss and crash events as the requested risk grows. If the claim holds, AV developers can expose a policy to a sweep of risk levels in closed-loop simulation without sacrificing realism or manually activating adversaries. The paper validates the approach on a real-world roundabout dataset, reporting crash-rate increases from 20.6% to 50.6% at fixed traffic volume and from 7.3% to 88.9% when volume scales with risk.

What carries the argument

The machinery is a conditional multi-agent diffusion model over state-only trajectories, with each vehicle state represented as $s=[p_x,p_y,\cos\theta,\sin\theta]$. The risk condition $r$ is computed from the minimum Post-Encroachment Time across occupancy-grid conflict cells through $r=\exp(-k\,\max(0,\mathrm{PET}/T-\sigma))$, and injected into every layer of a U-Net with multi-agent attention on skip connections; sampling uses classifier-free guidance $\hat{\epsilon}=\epsilon_\theta(\tau_k,\emptyset,k)+\omega(\epsilon_\theta(\tau_k,r,k)-\epsilon_\theta(\tau_k,\emptyset,k))$. A tokenized dynamics check then replaces each proposed motion with the nearest valid motion token from a vocabulary built by $k$-disks clustering of dataset transitions, enforcing physical plausibility without gradient-based guidance during denoising. Closed-loop simulation fixes the first denoised state to the current observed state, applies the dynamics check, advances one step, and spawns vehicles by a Poisson arrival process.

What would settle it

Train and run RADE with the same architecture, PET conditioning, and guidance scale on a second road geometry—say, a signalised intersection—using its own real trajectory data; if the crash rate no longer rises monotonically with requested risk or the realism distributions drift, the learned risk adjustment is specific to the roundabout and the PET surrogate rather than a general mapping. A cheaper check is to hold traffic volume fixed at the high-risk level while setting the condition to low risk $r=0.3$; if crash rates remain high, the conditioning signal itself is not causing the increase.

Watch

Extended reading notes

Core claim

RADE's central claim is that safety-critical traffic can be generated by learning the conditional distribution $p_\theta(\tau | r)$ over joint state trajectories of all vehicles, with $r$ a PET-derived risk index, and then rolling this distribution out autoregressively. The paper reports that this produces realistic gap-keeping, speed, and yielding behavior at every tested risk level, while the PET distribution shifts toward values below one second and crash rate rises consistently with $r$. The high-risk behaviors are not assigned to any vehicle; they emerge from the jointly generated futures, which the paper offers as evidence that multi-agent diffusion plus risk conditioning can serve as a scalable AV safety-validation environment. Empirically the claim is demonstrated on one two-lane roundabout dataset at 0.4 s resolution over 3.2 s prediction horizons, with 10 random-seeded runs of 100 episodes per risk level.

Load-bearing premise

The load-bearing premise is that a single scalar built from the minimum Post-Encroachment Time over all vehicle pairs captures enough of the interaction geometry that conditioning the diffusion model on that scalar reshapes joint behavior toward genuinely risky, still naturalistic driving; if the coupling is weak, the crash-rate trend could come from the arrival process or from motion-token snapping instead of from learned risk-conditioned behavior.

Editorial extensions

If this is right

  • At fixed traffic volume, raising the requested risk from $r=0.3$ to $r=1.0$ increases crash episodes from 20.6% to 50.6% and raises the density of near-miss PET values below 0.4 s by 113%.
  • When arrival rate is scaled with risk, the crash rate spans 7.3% to 88.9% and the near-miss PET density increases by more than 212%.
  • Inter-vehicle distance, speed, yielding distance, and yielding speed distributions stay close to the real roundabout data across all tested risk levels, so increased risk is not bought by visibly unrealistic spacing or speeds.
  • Crashes such as angle and sideswipe collisions emerge from the joint multi-agent generation without any per-vehicle adversarial activation, so the simulator can stress-test an AV over a continuous risk range without hand-scripted adversaries.

Reading between the lines

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

  • The paper leaves untested whether the learned risk adjustment transfers to other road geometries; conditioning a fresh RADE model on an intersection dataset would reveal whether a minimum-PET scalar is a general proxy for risky interaction.
  • Because the volume-scaled experiment changes arrival rate alongside risk, part of the 7.3% to 88.9% crash-range widening could be exposure rather than learned behavior; holding traffic volume fixed while sweeping $r$ would isolate the conditioning's effect.
  • A further step the authors do not take is to condition on alternative surrogate measures, such as time-to-collision, and compare which produces the steepest crash-rate gradient at fixed realism; the paper only notes this as future work.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes RADE, a risk-adjustable driving environment built on a multi-agent conditional diffusion model. Risk is encoded by a scalar PET-based index r (Eq. 9), which conditions trajectory generation via classifier-free guidance. A tokenized dynamics check module projects generated motions onto a vocabulary of real-world state transitions to enforce physical plausibility. Closed-loop simulations on the rounD roundabout dataset are evaluated across risk levels in [0.3, 1.0]. The paper reports that inter-vehicle distance, speed, and yielding distributions remain close to real data while the PET distribution shifts toward near-miss values and the crash rate rises monotonically from 20.6% to 50.6% (consistent traffic volume) or from 7.3% to 88.9% (risk-scaled traffic volume). The central claim is that RADE learns risk-conditioned multi-agent behaviors from data, enabling naturalistic generation of safety-critical events without explicit adversarial design.

Significance. If the central claim is established, RADE would be a practically useful tool for AV safety testing: it generates whole traffic scenes with controllable risk exposure while preserving statistical realism, in contrast to existing single-agent adversarial scenario generators. The multi-agent formulation and the use of a surrogate safety measure as a conditioning signal are reasonable and potentially generalizable. The paper also contributes a compute-friendly tokenized dynamics check. The crash-rate trend and the qualitative case studies are encouraging. However, the current evidence is not yet sufficient to validate the core mechanism: the evaluation is partly circular, the high-risk regime is likely extrapolation beyond the training label support, and the absence of ablations and quantitative baselines leaves open alternative explanations for the observed behavior.

major comments (4)
  1. [§IV-A, Eq. (7) vs. Algorithm 1] There is an internal inconsistency between the sampling formula and the algorithm. Eq. (7) writes the guided noise as ϵθ(τk, r(τk), k), i.e., conditioning on the risk of the noisy trajectory, while Algorithm 1 line 7 uses the user-specified desired risk r. For a noisy state τk, PET is not well defined and r(τk) does not equal the intended condition. Since this equation defines the core generation mechanism, the authors must state which conditioning signal is actually used and correct the inconsistent notation. The training loss in Eq. (8) conditions on r(τ0) with Bernoulli dropout, so the mismatch between training and the stated sampling procedure is a reproducibility issue.
  2. [§V-C, Figs. 6–7] The primary evidence for risk adjustability is the shift in the PET distribution, but PET is exactly the quantity used to define the conditioning signal r in training and sampling. This evaluation is therefore partly self-referential. The crash-rate result is a more independent outcome, but it is reported only as the mean over 10 seeds with min–max shading; there are no confidence intervals, significance tests, or per-seed scatter plots, so the monotonicity claim is not statistically established. The paper should add proper uncertainty quantification and an ablation that removes or scrambles the risk condition, to show that the effect is not produced by the guidance scale or by the tokenized dynamics check snapping trajectories near the data manifold.
  3. [§IV-C and §V-C] The high-risk conditioning regime is likely outside the training label support. With k=5, σ=0.05, and T=3.2, Eq. (9) gives r=1 for PET≤0.16 s and r=0.8 for PET≤0.30 s, both of which are near-collision events. Such events are rare in the naturalistic rounD dataset, so the conditional model at r∈[0.8, 1.0] is extrapolating far beyond the values of r(τ0) seen during training. The observed crash-rate increase could then be a guidance artifact (e.g., pushing trajectories off the data manifold, followed by token snapping into kinematically valid but behaviorally implausible motions) rather than a learned risk-behavior mapping. The paper neither reports the empirical distribution of r(τ0) over the training set nor provides a control experiment with randomized conditioning labels, so the central claim that RADE 'learns risk-conditioned behaviors directly from data' is not yet demonstrated.
  4. [§V-C, Fig. 5] Statistical realism is only assessed qualitatively through plotted distributions. The text states that RADE achieves a 'highly accurate match' with real-world distance and yielding distributions, but no quantitative divergence measure (e.g., KL divergence, Wasserstein distance) or standard errors are reported, and there is no comparison against an unguided or baseline generative model. Since realism preservation across risk levels is one of the two central claims, the authors should quantify the fit and, ideally, compare with an unconditional version of their model and with prior traffic simulation methods.
minor comments (5)
  1. [§IV-C, Eq. (9)] The derivation of PET over occupancy grids should clarify how conflict areas are defined for all vehicle pairs and how the minimum is taken across grid cells; currently the description is ambiguous and may be hard to reproduce.
  2. [Algorithm 1] The 'Denoise' function on line 8 is not defined in the pseudocode; specify how μk−1 and Σk−1 are computed from the estimated noise, and how the low-temperature scaling α enters the variance Σk−1.
  3. [Fig. 5] The curves for different risk levels are not clearly distinguishable in grayscale; use distinct line styles or add a legend so that the claim of alignment across risk levels can be visually verified.
  4. [§V-B] The case study says generation is performed in a single forward pass with no autoregressive generation, whereas the closed-loop evaluation in Algorithm 1 is autoregressive; this distinction should be stated more explicitly to avoid confusion.
  5. [§IV-D] The term 'k-disks clustering' may be unfamiliar to the broader AV audience; briefly define it or cite the origin, and state whether the motion vocabulary is constructed separately for each vehicle type or shared across all vehicles.

Circularity Check

1 steps flagged · score 4.0 of 10

PET-based evaluation is partly self-referential, but crash-rate and realism results provide independent support.

  1. self definitional [Section V-C 'Adjustable Risk'; risk defined in Eq. 9 in Section IV-C]
    "We proceed to evaluate whether the actual traffic risk has been successfully controlled in RADE by analyzing the PET distribution and the crash rate under varying risk levels."

    The conditioning label r is defined in Eq. 9 as r = exp(-5·max(0, PET/3.2 - 0.05)), a monotone transform of the scenario's minimum PET. Training (Eq. 8) uses r(τ0) as the condition, and sampling conditions on a desired r. The 'Adjustable Risk' evaluation then measures the PET distribution of the generated trajectories to show that 'actual traffic risk' is controlled. Because the conditioning variable and the evaluation metric are linked by the paper's own definition, a PET shift under varying r is a self-consistency check of the label function, not an independent confirmation that an extrinsic risk is being adjusted. The crash-rate result is a separate, more independent outcome, which prevents the whole claim from being circular.

full rationale

The training and sampling procedures condition the diffusion model on r, which Eq. 9 defines as a monotone function of the scenario minimum PET. The 'Adjustable Risk' evaluation then uses the PET distribution as the primary evidence that risk is controlled, so that portion of the validation is a self-consistency check of the conditioning label rather than an independent prediction. This is a genuine but mild self-definitional element. The crash-rate curve (Fig. 7) is independent of the PET label: a collision is not defined by Eq. 9, and the monotone increase from 20.6% to 50.6% (or 7.3% to 88.9% with scaled volume) is a nontrivial empirical result. The realism metrics (distance, speed, yielding distance/speed) are also independent distributional comparisons. Self-citations to [2], [4], and [19] are contextual and not load-bearing. A separate correctness caveat, not circularity, is that Eq. 7 conditions on r(τk) while Algorithm 1 conditions on the desired r; this is an internal inconsistency but does not make the derivation circular. Overall, the central risk-adjustability claim retains independent empirical content, so a moderate score of 4 is appropriate.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the learned risk-conditioned distribution. The risk mapping (Eq. 9) introduces three hand-chosen constants, and the sampling procedure uses two unreported hyperparameters (guidance scale and low-temperature factor). The main domain assumptions are the validity of PET as a safety surrogate, the coverage of the motion vocabulary, and the representativeness of the rounD dataset. No invented entities are introduced.

free parameters (5)
  • k (risk exponent coefficient) = 5
    Appears in Eq. 9 to map PET to risk. Chosen by hand; it sets the steepness of the risk curve and determines that PET<1 corresponds to r>0.3. The paper provides no fitting procedure or sensitivity analysis.
  • σ (bias) = 0.05
    Bias in Eq. 9, set to 0.05 to shift the threshold. Hand-chosen.
  • T (normalization limit) = 3.2
    Normalizes PET in Eq. 9; equals the simulation horizon. Hand-chosen, conflates the risk mapping with the episode length.
  • guidance scale ω = not reported
    Controls the strength of the risk conditioning in Eq. 7 and Algorithm 1. The paper never states the value used in the experiments, so the reported results are not reproducible without it.
  • low-temperature scaling α = not reported
    Initial noise scale in sampling (τ0 ~ N(0, αI)). Not reported; affects stochasticity of rollouts.
assumptions (4)
  • domain assumption The diffusion model with classifier-free guidance can learn the conditional distribution p(τ|r) accurately.
    This is an architectural trust assumption: no conditional density evaluation is performed, and the paper relies on the expressive power of the U-Net and the classifier-free guidance to make the conditioning effective. Invoked in Section IV-A.
  • domain assumption PET is a valid surrogate safety measure for crash risk in this scenario.
    The paper cites [13] and [34] but does not establish that the minimum PET across grid cells captures the multi-agent interactions that lead to the observed crashes. This underlies the definition of the training label r(τ).
  • domain assumption The motion vocabulary built by k-disks clustering covers the feasible maneuver space well enough for the dynamics check to preserve high-risk behaviors.
    If high-risk tokens are absent or underrepresented in the vocabulary, the filter either removes the riskiest moves or snaps them to a safer token, which would break the risk-control claim. The paper does not analyze the coverage of the vocabulary for PET<1 transitions. Introduced in Section IV-D.
  • domain assumption rounD provides a representative distribution of naturalistic roundabout interactions, including enough low-PET events to learn the risk-conditioned model.
    The conditioning is learned from the dataset's label distribution; if the dataset has no near-crashes, the model cannot learn high-risk generation. The paper does not report the PET distribution of the training data. Used throughout Section V.

how reviews work

0 comments
Cite this review

Pith. "Pith review of RADE: Learning Risk-Adjustable Driving Environment via Multi-Agent Conditional Diffusion." pith.science (2026). https://pith.science/paper/GX56MJTJ

@misc{pith2026250503178,
  author       = {Pith},
  title        = {Pith review of: RADE: Learning Risk-Adjustable Driving Environment via Multi-Agent Conditional Diffusion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GX56MJTJ}},
  note         = {Machine review of arXiv:2505.03178}
}
read the original abstract

Generating safety-critical scenarios in high-fidelity simulations offers a promising and cost-effective approach for efficient testing of autonomous vehicles. Existing methods typically rely on manipulating a single vehicle's trajectory through sophisticated designed objectives to induce adversarial interactions, often at the cost of realism and scalability. In this work, we propose the Risk-Adjustable Driving Environment (RADE), a simulation framework that generates statistically realistic and risk-adjustable traffic scenes. Built upon a multi-agent diffusion architecture, RADE jointly models the behavior of all agents in the environment and conditions their trajectories on a surrogate risk measure. Unlike traditional adversarial methods, RADE learns risk-conditioned behaviors directly from data, preserving naturalistic multi-agent interactions with controllable risk levels. To ensure physical plausibility, we incorporate a tokenized dynamics check module that efficiently filters generated trajectories using a motion vocabulary. We validate RADE on the real-world rounD dataset, demonstrating that it preserves statistical realism across varying risk levels and naturally increases the likelihood of safety-critical events as the desired risk level grows up. Our results highlight RADE's potential as a scalable and realistic tool for AV safety evaluation.

Figures

Figures reproduced from arXiv: 2505.03178 by the authors.

Figure 1
Figure 1. The proposed RADE framework. Conditioned on the desired risk [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Network architecture based on the MADiff backbone [12]. The [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Tokenized dynamics check. (a) Schematic for dynamics check. This [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Snapshots of two scenarios under different risk levels. For each scenario, the left and right panels represent vehicle states at [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 6
Figure 6. Figure 6: Comparison of PET distribution at different risk levels. Each metric [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 5
Figure 5. Figure 5: Statistical realism at different risk levels. Each metric is calculated [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 7
Figure 7. Figure 7: Crash rate across varying risk levels under two traffic volume [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Generative AI for Testing of Autonomous Driving Systems: A Survey

    cs.SE 2025-08 conditional novelty 5.0 of 10

    A systematic survey that organizes 91 studies of generative AI for autonomous driving testing into six scenario-based tasks and catalogs 27 limitations.

Reference graph

Works this paper leans on

34 extracted references · 17 canonical work pages · cited by 1 Pith paper

  1. [1]

    Continuous improvement of self-driving cars using dynamic confidence-aware reinforcement learning,

    Z. Cao, K. Jiang, W. Zhou, S. Xu, H. Peng, and D. Yang, “Continuous improvement of self-driving cars using dynamic confidence-aware reinforcement learning,” Nature Machine Intelligence , vol. 5, no. 2, pp. 145–158, 2023

  2. [2]

    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

  3. [3]

    Trafficsim: Learning to simulate realistic multi-agent behaviors,

    S. Suo, S. Regalado, S. Casas, and R. Urtasun, “Trafficsim: Learning to simulate realistic multi-agent behaviors,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2021, pp. 10 400–10 409

  4. [4]

    Learn- ing naturalistic driving environment with statistical realism,

    X. Yan, Z. Zou, S. Feng, H. Zhu, H. Sun, and H. X. Liu, “Learn- ing naturalistic driving environment with statistical realism,” Nature Communications, vol. 14, no. 1, p. 2037, 2023

  5. [5]

    Adaptive stress testing for autonomous vehicles,

    M. Koren, S. Alsaif, R. Lee, and M. J. Kochenderfer, “Adaptive stress testing for autonomous vehicles,” in 2018 IEEE Intelligent Vehicles Symposium (IV). IEEE, 2018, pp. 1–7

  6. [6]

    Training adversarial agents to exploit weaknesses in deep control policies,

    S. Kuutti, S. Fallah, and R. Bowden, “Training adversarial agents to exploit weaknesses in deep control policies,” in 2020 IEEE Interna- tional Conference on Robotics and Automation (ICRA) . IEEE, 2020, pp. 108–114

  7. [7]

    Generat- ing useful accident-prone driving scenarios via a learned traffic prior,

    D. Rempe, J. Philion, L. J. Guibas, S. Fidler, and O. Litany, “Generat- ing useful accident-prone driving scenarios via a learned traffic prior,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 17 305–17 315

  8. [8]

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

    C. Xu, D. Zhao, A. Sangiovanni-Vincentelli, and B. Li, “Diffscene: Diffusion-based safety-critical scenario generation for autonomous vehicles,” in The Second Workshop on New Frontiers in Adversarial Machine Learning, 2023

Show all 34 references
  1. [9]

    Safe-sim: Safety-critical closed-loop traffic simulation with diffusion- controllable adversaries,

    W.-J. Chang, F. Pittaluga, M. Tomizuka, W. Zhan, and M. Chandraker, “Safe-sim: Safety-critical closed-loop traffic simulation with diffusion- controllable adversaries,” in European Conference on Computer Vi- sion. Springer, 2024, pp. 242–258

  2. [10]

    Advdiffuser: Generating adversarial safety-critical driving scenarios via guided diffusion,

    Y . Xie, X. Guo, C. Wang, K. Liu, and L. Chen, “Advdiffuser: Generating adversarial safety-critical driving scenarios via guided diffusion,” in 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . IEEE, 2024, pp. 9983–9989

  3. [11]

    Terasim: Uncovering unknown unsafe events for autonomous vehicles through generative simulation,

    H. Sun, X. Yan, Z. Qiao, H. Zhu, Y . Sun et al., “Terasim: Uncovering unknown unsafe events for autonomous vehicles through generative simulation,” arXiv preprint arXiv:2503.03629 , 2025

  4. [12]

    Madiff: Offline multi-agent learning with diffusion models,

    Z. Zhu, M. Liu, L. Mao, B. Kang, M. Xu, Y . Yu, S. Ermon, and W. Zhang, “Madiff: Offline multi-agent learning with diffusion models,” Advances in Neural Information Processing Systems, vol. 37, pp. 4177–4206, 2024

  5. [13]

    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

  6. [14]

    Congested traffic states in empirical observations and microscopic simulations,

    M. Treiber, A. Hennecke, and D. Helbing, “Congested traffic states in empirical observations and microscopic simulations,” Physical review E, vol. 62, no. 2, p. 1805, 2000

  7. [15]

    General lane-changing model mobil for car-following models,

    A. Kesting, M. Treiber, and D. Helbing, “General lane-changing model mobil for car-following models,” Transportation Research Record, vol. 1999, no. 1, pp. 86–94, 2007

  8. [16]

    Bits: Bi-level imitation for traffic simulation,

    D. Xu, Y . Chen, B. Ivanovic, and M. Pavone, “Bits: Bi-level imitation for traffic simulation,” in 2023 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, 2023, pp. 2929–2936

  9. [17]

    Trafficbots: Towards world models for autonomous driving simulation and motion prediction,

    Z. Zhang, A. Liniger, D. Dai, F. Yu, and L. Van Gool, “Trafficbots: Towards world models for autonomous driving simulation and motion prediction,” in 2023 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2023, pp. 1522–1529

  10. [18]

    Smart: Scalable multi-agent real-time motion generation via next-token prediction,

    W. Wu, X. Feng, Z. Gao, and Y . Kan, “Smart: Scalable multi-agent real-time motion generation via next-token prediction,” Advances in Neural Information Processing Systems, vol. 37, pp. 114 048–114 071, 2024

  11. [19]

    Dense reinforcement learning for safety validation of autonomous vehicles,

    S. Feng, H. Sun, X. Yan, H. Zhu, Z. Zou, S. Shen, and H. X. Liu, “Dense reinforcement learning for safety validation of autonomous vehicles,” Nature, vol. 615, no. 7953, pp. 620–627, 2023

  12. [20]

    King: Generating safety-critical driving scenarios for robust imitation via kinematics gradients,

    N. Hanselmann, K. Renz, K. Chitta, A. Bhattacharyya, and A. Geiger, “King: Generating safety-critical driving scenarios for robust imitation via kinematics gradients,” in European Conference on Computer Vision. Springer, 2022, pp. 335–352

  13. [21]

    Cat: Closed-loop adversarial training for safe end-to-end driving,

    L. Zhang, Z. Peng, Q. Li, and B. Zhou, “Cat: Closed-loop adversarial training for safe end-to-end driving,” in Conference on Robot Learn- ing. PMLR, 2023, pp. 2357–2372

  14. [22]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” Advances in neural information processing systems , vol. 33, pp. 6840–6851, 2020

  15. [23]

    Planning with diffusion for flexible behavior synthesis,

    M. Janner, Y . Du, J. Tenenbaum, and S. Levine, “Planning with diffusion for flexible behavior synthesis,” in International Conference on Machine Learning . PMLR, 2022, pp. 9902–9915

  16. [24]

    Is conditional generative modeling all you need for decision- making?

    A. Ajay, Y . Du, A. Gupta, J. Tenenbaum, T. Jaakkola, and P. Agrawal, “Is conditional generative modeling all you need for decision- making?” arXiv preprint arXiv:2211.15657 , 2022

  17. [25]

    Guided conditional diffusion for controllable traffic simulation,

    Z. Zhong, D. Rempe, D. Xu, Y . Chen, S. Veer, T. Che, B. Ray, and M. Pavone, “Guided conditional diffusion for controllable traffic simulation,” in 2023 IEEE international conference on robotics and automation (ICRA). IEEE, 2023, pp. 3560–3566

  18. [26]

    Language-guided traffic simulation via scene-level diffu- sion,

    Z. Zhong, D. Rempe, Y . Chen, B. Ivanovic, Y . Cao, D. Xu, M. Pavone, and B. Ray, “Language-guided traffic simulation via scene-level diffu- sion,” in Conference on Robot Learning. PMLR, 2023, pp. 144–177

  19. [27]

    Versatile scene-consistent traffic scenario generation as optimization with diffusion,

    Z. Huang, Z. Zhang, A. Vaidya, Y . Chen, C. Lv, and J. F. Fisac, “Versatile scene-consistent traffic scenario generation as optimization with diffusion,” arXiv preprint arXiv:2404.02524 , vol. 3, 2024

  20. [28]

    Scenediffuser: Efficient and controllable driving simulation initialization and rollout,

    M. Jiang, Y . Bai, A. Cornman, C. Davis, X. Huang, H. Jeon, S. Kul- shrestha, J. Lambert, S. Li, X. Zhouet al., “Scenediffuser: Efficient and controllable driving simulation initialization and rollout,” Advances in Neural Information Processing Systems , vol. 37, pp. 55 729–55...

  21. [29]

    Classifier-free diffusion guidance,

    J. Ho and T. Salimans, “Classifier-free diffusion guidance,” arXiv preprint arXiv:2207.12598, 2022

  22. [30]

    A review of surrogate safety measures and their applications in connected and automated vehicles safety modeling,

    C. Wang, Y . Xie, H. Huang, and P. Liu, “A review of surrogate safety measures and their applications in connected and automated vehicles safety modeling,” Accident Analysis & Prevention, vol. 157, p. 106157, 2021

  23. [31]

    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,” in 2020 IEEE 23rd International Conference on Intelligent Transportation Systems (ITSC) . IEEE, 2020, pp. 1–6

  24. [32]

    Trajeglish: Traffic modeling as next-token prediction,

    J. Philion, X. B. Peng, and S. Fidler, “Trajeglish: Traffic modeling as next-token prediction,” arXiv preprint arXiv:2312.04535 , 2023

  25. [33]

    Motion transformer with global intention localization and local movement refinement,

    S. Shi, L. Jiang, D. Dai, and B. Schiele, “Motion transformer with global intention localization and local movement refinement,” Ad- vances in Neural Information Processing Systems , vol. 35, pp. 6531– 6543, 2022

  26. [34]

    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,” International journal of injury control and safety promotion , vol. 27, no. 2, pp. 121–135, 2020

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.