Pith. sign in

REVIEW 5 major objections 4 minor 59 references

SceneDiffuser++: City-Scale Traffic Simulation via a Generative World Model

T0 review · 5 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a single diffusion-based generative world model, trained end-to-end on one denoising loss, can simulate a full trip from point A to point B at city scale, jointly generating agent trajectories, agent validity…

desk verdict A real step forward for trip-level traffic simulation with a clever sparse-tensor validity trick, but the 'point A-to-B' claim outruns the experiments and the validity channel's sensor-relative semantics deserve closer scrutiny. read the letter →

arxiv 2506.21976 v1 pith:6GWOMZEH submitted 2025-06-27 cs.LG cs.AIcs.CVcs.MAcs.RO

Shuhan Tan , John Lambert , Hong Jeon , Sakshum Kulshrestha , Yijing Bai , Jing Luo , Dragomir Anguelov , Mingxing Tan
show 1 more author
Chiyu Max Jiang
This is my paper · ORCID
classification cs.LGcs.AIcs.CVcs.MAcs.RO
keywords trafficsimulationdiffusionmodelworldautoregressiverolloutsparsetensorsagentvaliditylightsWaymoOpenMotionDataset
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

SceneDiffuser++ sets out to show that one end-to-end generative model can replace the usual stack of specialised traffic simulators: given a large map, a start point, and an autonomous vehicle stack, it should populate the scene, animate vehicles and pedestrians, decide when agents appear and disappear, reason about occlusion, and set traffic light states for a trip lasting minutes rather than seconds. The paper claims all of this can be done by a single diffusion model trained on a single denoising loss, then rolled out autoregressively. To support the claim, the authors build a sparse scene tensor in which every agent and traffic light carries a validity channel indicating its presence in the ego vehicle's detection output, so spawning, removal, and occlusion become one prediction problem rather than separate modules. On a map-extended version of the Waymo Open Motion Dataset with kilometer-scale regions, they measure realism by comparing simulated and logged distributions over agent counts, entry and exit distances, speeds, offroad and collision rates, and traffic-light transitions, and report lower divergence than IDM and SceneDiffuser across most trip-level metrics. A sympathetic reader would take this as evidence that trip-level simulation of a generative city is feasible as one learned world model.

What carries the argument

The load-bearing object is the multi-tensor scene representation $X = \{x_{\mathrm{agent}}, x_{\mathrm{light}}\}$, where each $x_i \in \mathbb{R}^{E_i \times T \times D_i}$ stacks $E_i$ elements over $T$ timesteps and each element's last feature is a validity bit. The denoiser is an axial-attention transformer (the SceneDiffuser backbone) whose input is the element-concatenated, hidden-dimension-homogenized tensor, trained with the single v-prediction diffusion loss. Two sparse-tensor mechanisms carry the argument: during training, invalid entries are zeroed and the loss is masked so only the validity channel is supervised at invalid steps; during inference, soft clipping multiplies the denoised feature values by the predicted validity, so invalid entries decay toward zero instead of drifting out of distribution. That combination is what lets one model output a time-varying, sparse set of valid agents and lights, and it is the mechanism behind the paper's claimed ability to handle spawning, removal, occlusion, and traffic signals jointly.

What would settle it

Replay 60-second SceneDiffuser++ rollouts with a fixed route-unconditioned planner, then ray-cast from the simulated ego pose against the map geometry to compute the physically visible agent set at each step; if the model's predicted validity (or its entering and exiting events) matches ray-cast visibility no better than a baseline that ignores map occlusion, the claim that the validity channel encodes occlusion and visibility would fail.

Watch

Extended reading notes

Core claim

The central claim is that agent validity — whether an agent appears in the AV's detection output at a given timestep — can be treated as just another channel of the scene tensor and learned jointly with positions, sizes, and object types. Because validity is generated by the diffusion model itself rather than taken from the log, the model can emit sparse tensors without a prespecified sparsity structure: it learns when to insert a new agent, when to let an existing agent leave, and when an agent is occluded or disoccluded, by interpolating between valid and zero-invalid states during sampling. The same multi-tensor construction also carries a second tensor for traffic lights, with states and positions denoised alongside agents, so the entire trip is one inpainting problem. The paper reports state-of-the-art trip-level realism in 60-second rollouts, with substantially lower Jensen-Shannon divergence than IDM and SceneDiffuser for the distributions that depend directly on validity, while also matching logged traffic-light transition probabilities.

Load-bearing premise

The validity channel is supervised using the logged AV's detection output, so 'valid' means visible to that particular sensor trajectory; in route-unconditioned rollouts the simulated AV drives elsewhere, and nothing in the training objective ties predicted validity to actual visibility from the simulated ego, so the spawned and removed agents may not be the ones the simulated AV could actually see.

Editorial extensions

If this is right

  • Trip-level statistics such as travel time, pick-up and drop-off behavior, and long-horizon interactions with emergency vehicles can be generated from logged data alone, without hand-written insertion and removal rules.
  • Because the world model predicts traffic-light state transitions, violations of light rules can be measured in simulation rather than assumed from map priors.
  • The validity channel removes the hard limit that previously tied simulation duration to the length of the log: agents can be recycled into previously vacated slots, so arbitrarily many agents can appear over long horizons.
  • More frequent planner-world-model interaction lowers collision rates and improves speed realism, suggesting the model is most realistic when the AV and background agents continuously react to each other.
  • Rollout realism degrades as the horizon grows, especially for the timing of agent insertion, which indicates that autoregressive error accumulation, not map coverage, is the next bottleneck for trip-level simulation.

Reading between the lines

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

  • We would expect the sparse-tensor validity trick to transfer to other generative settings where outputs are naturally sparse, such as point-cloud completion with missing measurements or simultaneous generation of landmarks and objects; the paper does not test this.
  • A natural extension, not pursued in the paper, is to condition validity on the simulated ego's sensor footprint rather than on the logged AV's detection output, so that spawn and removal events remain physically meaningful when the simulated route diverges from the log.
  • The histogram-based Jensen-Shannon divergence protocol is the paper's own proposal; a reasonable next step for the field is to make such correspondence-free, sliding-window distributional metrics standard for long-horizon simulators.
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 / 4 minor

Summary. The paper proposes SceneDiffuser++, a diffusion-based generative world model for city-scale traffic simulation. The model jointly predicts, from a single denoising objective, agent trajectories, a validity channel that encodes agent presence/visibility, and traffic light states, via an autoregressive rollout over a sparse scene tensor. The authors introduce soft clipping at inference to stabilize the generation of sparse tensors, a multi-tensor architecture to jointly handle agents and traffic lights with different feature dimensions, and a map-extended version of the Waymo Open Motion Dataset (WOMD-XLMap) for trip-level simulation. They evaluate realism using JS divergence between histograms of simulated and logged metrics (agent counts, entering/exiting distances, offroad and collision rates, speeds, traffic-light transitions), reporting quantitative gains over IDM and SceneDiffuser baselines on 60s rollouts.

Significance. If the central claims are upheld, the paper would make a notable contribution: a single end-to-end model that handles scene generation, dynamic agent insertion/removal, occlusion, and traffic-light simulation without task-specific heads or heuristics. The sparse-tensor learning method is an interesting technical idea, and the joint modeling of agents and traffic lights is a useful step beyond prior work. The paper also provides an enlarged dataset and a nontrivial evaluation protocol for long-horizon simulation, which is valuable to the community. However, the claim of 'point A-to-B simulation' is not actually exercised in the experiments, and the validity channel, which is central to the new capabilities, has a train/inference semantic mismatch. These issues currently prevent the paper's strongest claims from being supported.

major comments (5)
  1. [Abstract and Appendix A.4] The abstract and Section 1 claim that SceneDiffuser++ is capable of 'point A-to-B simulation on a city scale', but Appendix A.4 explicitly states: 'SceneDiffuser and SceneDiffuser++ do not use goal-oriented routing; in other words, they do not use or ingest a goal location in any way, shape, or form.' Figure 1 depicts a 'trip end' star but no conditioning mechanism is described or tested. The route-unconditioned rollouts used in Section 4.1 do not implement point-to-point trips, so this central claim is unsupported as stated. Either the model must be extended with goal/route conditioning, or the paper should be reframed around route-unconditioned world-model simulation and the point-to-point claim removed from the abstract and contributions.
  2. [Appendix A.4 (Validity Definition) and Section 3 (Learning Sparse Tensors)] The validity channel is trained to reproduce whether an agent appears in the logged AV's detection output, and all positions are normalized by the logged AV's ego pose (Section 3). In route-unconditioned rollouts, the simulated AV follows an arbitrary path that diverges from the logged trajectory, and nothing in the loss, architecture, or sampling procedure ties predicted validity to the simulated AV's actual sensor frustum or detection range. The generated validity can therefore encode 'agent exists on the map' rather than 'agent is visible to this AV'. This undermines the paper's occlusion-reasoning claims and makes the validity-based metrics (# valid agents, entering/exiting agents and distances) an unvalidated proxy for sensor-relative realism. The authors should either constrain validity at inference with the simulated AV's sensor geometry (e.g., raycasting or frustum culling) or provide evidence that the learned validity remains consistent with the simulated AV's visibility.
  3. [Section 4.1 (Metrics)] The evaluation relies on JS divergence between histograms of simulated and logged metric values, but the paper does not specify histogram bin width or number of bins, report error bars or confidence intervals, state the number of independent rollout seeds, or validate that the divergence is sensitive to the sample sizes used. The numbers in Table 1, such as 0.3132 vs. 0.2206 for # Valid Agents, may not be statistically distinguishable. In addition, the Composite score averages metrics with different units and scales without justification. Without these details, the headline 'state-of-the-art trip-level simulation realism' is not established. Adding standard errors over seeds and explicit histogram construction would make the comparison interpretable.
  4. [Table 1 and Section 4.2] Several comparison cells are missing for the IDM and SceneDiffuser baselines: Entering and Exiting Distances are not reported for these world models, and Traffic Light Violation and Traffic Light Transition are not reported for any baseline. The text states that SceneDiffuser++ 'achieves significantly better performance in all metrics that relate to agent insertion and removal', but two of the four insertion/removal metrics lack baseline values. Moreover, because IDM and SceneDiffuser never insert agents, their validity distributions are degenerate by construction, which can inflate or deflate JS divergence in ways not discussed. The comparison should either report the missing values (with a principled definition for non-inserting baselines) or restrict the superiority claim to the metrics actually measured against both baselines.
  5. [Section 4.3, Table 3] The text in Section 3 states that imputing invalid values with zeros 'cannot work', yet Table 3 includes a 'No Clipping' row described as 'a model trained to directly predict invalid agents' features to be 0', which yields a lower JS divergence for # Valid Agents (0.2426) than Soft Clipping (0.3053) and a comparable # Entering Agents (0.2035 vs. 0.2120). The claim that this approach 'cannot work' is internally inconsistent with the reported data. The ablation should present the full trade-off (e.g., no clipping is better on some agent-count metrics, worse on collision/offroad/TL), and the choice of soft clipping should be justified by the metric priorities of the target use case rather than by a blanket statement.
minor comments (4)
  1. [Figure 7] The legend in the bottom plot uses 'SceneDiffusion', which should be 'SceneDiffuser' for consistency with the rest of the paper.
  2. [Section 3, Eq. (2)] The loss weight w is described only as 'a loss weighting term'; its exact form (including the loss mask described later in the same section) should be specified in the equation or immediately after, since it is central to the sparse-tensor training.
  3. [Figure 6] The 'Ground-truth Log' panel covers only 91 steps but the predicted panels show 600 steps; the caption should state this difference explicitly to avoid implying direct temporal alignment beyond step 91.
  4. [Section 4.2] The phrase 'Average Speed likelihood' appears to mix a distributional metric with a probabilistic notion; consider rewording to 'Average Speed distribution realism'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the validity-channel concern is a train/deployment mismatch, not an input-output reduction.

full rationale

The paper's derivation chain is self-contained against external data: it defines scene tensors, trains a denoiser with the v-prediction loss in Eq. (2) on logged WOMD-XLMap data, rolls out autoregressively, and evaluates realism via JS divergence between simulated and held-out log histograms. No parameter is fitted to the evaluation metrics; the soft-clipping sampler is an inference-time decoding choice ablated on validation behavior, not a quantity derived from the metrics it later ranks. The SceneDiffuser backbone is a self-citation, but it is used as an architectural component rather than as a premise that forces the validity, traffic-light, or spawning results. The closest concern is Appendix A.4's sensor-relative validity definition combined with route-unconditioned rollout: log validity means 'appears in the logged AV's detection output,' while deployment uses a different simulated route with no explicit sensor model. This is a genuine generalization and correctness risk, but it is not circularity under the paper's equations, because predicted validity is not defined in terms of the simulated AV's visible set, nor is it forced to equal an input by construction. The model could in principle learn an implicit proxy from the simulated AV's history. No load-bearing argument reduces to a self-citation, no fitted input is renamed as a prediction, and no uniqueness claim is imported from the authors' prior work. Therefore the correct finding is no significant circularity.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the validity channel and its training target, on generalization from short clips to long rollouts, and on the evaluation metric being a meaningful measure of realism. No new physical entities are introduced.

free parameters (2)
  • Validity soft-clipping threshold = 0.5
    Used to decide which agents are considered valid during inference. Hand-chosen, not fitted to data, and only ablated across clipping variants.
  • Feature normalization constants = 1/80 for x,y,z; mu_l=4.5, mu_w=2.0, mu_h=1.75, mu_k=0.5; sigma_l=2.5, sigma_w=0.8, sigma_h=0.6, sigma_k=0.5
    Hand-picked or statistics-derived scaling to map features to the (-1,1) range. These are preprocessing choices, not the result of fitting to the evaluation metric.
assumptions (4)
  • domain assumption The WOMD-XLMap map expansion covers a sufficient road network for 1km-radius trip-level simulation.
    The paper requests expanded maps but does not specify coverage statistics; long rollouts may leave the map extent.
  • domain assumption Agent validity defined by appearance in the AV's detection output is a sufficient training target for learning occlusion and spawning behavior.
    This equates visibility with the logged perception system, and assumes the model can generalize validity to off-log routes without simulating sensor geometry.
  • domain assumption A model trained on 9.1s WOMD clips generalizes to 60-300s rollouts and unseen map regions.
    The paper's own results show degradation with longer horizons, but the central claim depends on useful generalization.
  • standard math Standard diffusion v-prediction and cosine schedule are valid for multi-tensor denoising.
    Adopted from Eq. 2 and references [20,24]; not re-derived here.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SceneDiffuser++: City-Scale Traffic Simulation via a Generative World Model." pith.science (2026). https://pith.science/paper/6GWOMZEH

@misc{pith2026250621976,
  author       = {Pith},
  title        = {Pith review of: SceneDiffuser++: City-Scale Traffic Simulation via a Generative World Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6GWOMZEH}},
  note         = {Machine review of arXiv:2506.21976}
}
read the original abstract

The goal of traffic simulation is to augment a potentially limited amount of manually-driven miles that is available for testing and validation, with a much larger amount of simulated synthetic miles. The culmination of this vision would be a generative simulated city, where given a map of the city and an autonomous vehicle (AV) software stack, the simulator can seamlessly simulate the trip from point A to point B by populating the city around the AV and controlling all aspects of the scene, from animating the dynamic agents (e.g., vehicles, pedestrians) to controlling the traffic light states. We refer to this vision as CitySim, which requires an agglomeration of simulation technologies: scene generation to populate the initial scene, agent behavior modeling to animate the scene, occlusion reasoning, dynamic scene generation to seamlessly spawn and remove agents, and environment simulation for factors such as traffic lights. While some key technologies have been separately studied in various works, others such as dynamic scene generation and environment simulation have received less attention in the research community. We propose SceneDiffuser++, the first end-to-end generative world model trained on a single loss function capable of point A-to-B simulation on a city scale integrating all the requirements above. We demonstrate the city-scale traffic simulation capability of SceneDiffuser++ and study its superior realism under long simulation conditions. We evaluate the simulation quality on an augmented version of the Waymo Open Motion Dataset (WOMD) with larger map regions to support trip-level simulation.

Figures

Figures reproduced from arXiv: 2506.21976 by the authors.

Figure 1
Figure 1. Overview. SceneDiffuser++ is a single unified, end-to-end trained generative world model that enables CitySim: city-scale traffic [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Multi-Tensor Diffusion jointly denoises the set of scene elements in a joint fashion. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Learning sparse signals with diffusion models. We illustrate the scene tensor using one single slice (for a single agent). Learning the validity field in this sparse tensor allows us to model (a) agent spawning, (b) occlusion, (c) disocclusion and (d) removal. Left: During training, we impute the corresponding values for all invalid steps to always be zero, before adding noise to train the denoiser. A corresponding … view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Visualization of a 60s rollout with SceneDiffuser++ (top row) vs. SceneDiffuser, where all agents become stuck in a single [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 6
Figure 6. Figure 6: Plot of predicted validity for all 128 agents across 600 [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 5
Figure 5. Figure 5: Plots of SceneDiffuser++ results (green→blue indicates temporal progression). over long horizons. Note that although the realism of the number of entering and exiting agents degrades over time, the respective entering or exiting distances stay quite stable. This might …
Figure 8
Figure 8. Figure 8: Traffic light transition probability matrix for log vs ours. [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Additional qualitative results of SceneDiffuser++. [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: Additional qualitative results of SceneDiffuser++. [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

59 extracted references · 52 canonical work pages

  1. [1]

    A survey of statistical model checking

    Gul Agha and Karl Palmskog. A survey of statistical model checking. ACM Transactions on Modeling and Computer Simulation (TOMACS), 28(1):1–39, 2018. 2

    work page 2018

  2. [2]

    SimNet: Learning reactive self-driving sim- ulations from real-world observations

    Luca Bergamini, Yawei Ye, Oliver Scheel, Long Chen, Chih Hu, Luca Del Pero, Bła ˙zej Osi´nski, Hugo Grimmet, and Pe- ter Ondruska. SimNet: Learning reactive self-driving sim- ulations from real-world observations. In ICRA, 2021. 2, 3

    work page 2021

  3. [3]

    Video generation models as world simulators

    Tim Brooks, Bill Peebles, Connor Holmes, Will DePue, Yufei Guo, Li Jing, David Schnurr, Joe Taylor, Troy Luh- man, Eric Luhman, Clarence Ng, Ricky Wang, and Aditya Ramesh. Video generation models as world simulators

  4. [4]

    nuScenes: A multi- modal dataset for autonomous driving

    Holger Caesar, Varun Bankiti, Alex H Lang, Sourabh V ora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Gi- ancarlo Baldan, and Oscar Beijbom. nuScenes: A multi- modal dataset for autonomous driving. In CVPR, 2020. 2

    work page 2020

  5. [5]

    Argo- verse: 3d tracking and forecasting with rich maps

    Ming-Fang Chang, John Lambert, Patsorn Sangkloy, Jag- jeet Singh, Slawomir Bak, Andrew Hartnett, De Wang, Peter Carr, Simon Lucey, Deva Ramanan, and James Hays. Argo- verse: 3d tracking and forecasting with rich maps. In CVPR,

    work page

  6. [6]

    Controllable safety- critical closed-loop traffic simulation via guided diffusion,

    Wei-Jer Chang, Francesco Pittaluga, Masayoshi Tomizuka, Wei Zhan, and Manmohan Chandraker. Controllable safety- critical closed-loop traffic simulation via guided diffusion,

    work page

  7. [7]

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

    Wei-Jer Chang, Francesco Pittaluga, Masayoshi Tomizuka, Wei Zhan, and Manmohan Chandraker. Safe-sim: Safety-critical closed-loop traffic simulation with diffusion- controllable adversaries. In ECCV, 2024. 3

    work page 2024

  8. [8]

    Sledge: Synthesizing driving environments with generative models and rule-based traffic

    Kashyap Chitta, Daniel Dauner, and Andreas Geiger. Sledge: Synthesizing driving environments with generative models and rule-based traffic. In ECCV, 2024. 2

    work page 2024

Show all 59 references
  1. [9]

    Dice: Diverse diffu- sion model with scoring for trajectory prediction, 2023

    Younwoo Choi, Ray Coden Mercurius, Soheil Mohamad Al- izadeh Shabestary, and Amir Rasouli. Dice: Diverse diffu- sion model with scoring for trajectory prediction, 2023. 2

    2023

  2. [10]

    A survey of algorithms for black-box safety validation of cyber-physical systems

    Anthony Corso, Robert Moss, Mark Koren, Ritchie Lee, and Mykel Kochenderfer. A survey of algorithms for black-box safety validation of cyber-physical systems. Journal of Arti- ficial Intelligence Research, 72:377–428, 2021. 2

    2021

  3. [11]

    CARLA: An open urban driving simulator

    Alexey Dosovitskiy, German Ros, Felipe Codevilla, Antonio Lopez, and Vladlen Koltun. CARLA: An open urban driving simulator. In CoRL, 2017. 3

    2017

  4. [12]

    Qi, Yin Zhou, Zoey Yang, Aur´elien Chouard, Pei Sun, Jiquan Ngiam, Vijay Vasudevan, Alexander McCauley, Jonathon Shlens, and Dragomir Anguelov

    Scott Ettinger, Shuyang Cheng, Benjamin Caine, Chenxi Liu, Hang Zhao, Sabeek Pradhan, Yuning Chai, Ben Sapp, Charles R. Qi, Yin Zhou, Zoey Yang, Aur´elien Chouard, Pei Sun, Jiquan Ngiam, Vijay Vasudevan, Alexander McCauley, Jonathon Shlens, and Dragomir Anguelov. Large scale i...

    2021

  5. [13]

    Trafficgen: Learning to generate diverse and realistic traffic scenarios

    Lan Feng, Quanyi Li, Zhenghao Peng, Shuhan Tan, and Bolei Zhou. Trafficgen: Learning to generate diverse and realistic traffic scenarios. In ICRA, 2023. 3

    2023

  6. [14]

    Co-Reyes, Rishabh Agarwal, Re- becca Roelofs, Yao Lu, Nico Montali, Paul Mougin, Zoey Yang, Brandyn White, Aleksandra Faust, Rowan McAllister, Dragomir Anguelov, and Benjamin Sapp

    Cole Gulino, Justin Fu, Wenjie Luo, George Tucker, Eli Bronstein, Yiren Lu, Jean Harb, Xinlei Pan, Yan Wang, Xiangyu Chen, John D. Co-Reyes, Rishabh Agarwal, Re- becca Roelofs, Yao Lu, Nico Montali, Paul Mougin, Zoey Yang, Brandyn White, Aleksandra Faust, Rowan McAllister, Dra...

  7. [15]

    SceneDM: Scene-level multi-agent trajectory gen- eration with consistent diffusion models, 2023

    Zhiming Guo, Xing Gao, Jianlan Zhou, Xinyu Cai, and Bo- tian Shi. SceneDM: Scene-level multi-agent trajectory gen- eration with consistent diffusion models, 2023. 2

    2023

  8. [16]

    Recurrent world models facilitate policy evolution

    David Ha and J ¨urgen Schmidhuber. Recurrent world models facilitate policy evolution. In NeurIPS, 2018. 2

    2018

  9. [17]

    Dream to control: Learning behaviors by la- tent imagination

    Danijar Hafner, Timothy Lillicrap, Jimmy Ba, and Moham- mad Norouzi. Dream to control: Learning behaviors by la- tent imagination. In ICLR, 2020

    2020

  10. [18]

    Mastering atari with discrete world models

    Danijar Hafner, Timothy P Lillicrap, Mohammad Norouzi, and Jimmy Ba. Mastering atari with discrete world models. In ICLR, 2021. 2

    2021

  11. [19]

    A holistic framework towards vision-based traffic sig- nal control with microscopic simulation

    Pan He, Quanyi Li, Xiaoyong Yuan, and Bolei Zhou. A holistic framework towards vision-based traffic sig- nal control with microscopic simulation. arXiv preprint arXiv:2403.06884, 2024. 3

    2024

  12. [20]

    sim- ple diffusion: End-to-end diffusion for high resolution im- ages

    Emiel Hoogeboom, Jonathan Heek, and Tim Salimans. sim- ple diffusion: End-to-end diffusion for high resolution im- ages. In ICML. PMLR, 2023. 3, 4

    2023

  13. [21]

    Gaia-1: A generative world model for au- tonomous driving

    Anthony Hu, Lloyd Russell, Hudson Yeo, Zak Murez, George Fedoseev, Alex Kendall, Jamie Shotton, and Gian- luca Corrado. Gaia-1: A generative world model for au- tonomous driving. arXiv preprint arXiv:2309.17080, 2023. 2

    2023 arXiv

  14. [22]

    Versatile scene- consistent traffic scenario generation as optimization with diffusion, 2024

    Zhiyu Huang, Zixu Zhang, Ameya Vaidya, Yuxiao Chen, Chen Lv, and Jaime Fern ´andez Fisac. Versatile scene- consistent traffic scenario generation as optimization with diffusion, 2024. 2, 3

    2024

  15. [23]

    Motion- diffuser: Controllable multi-agent motion prediction using diffusion

    Chiyu “Max” Jiang, Andre Cornman, Cheolho Park, Ben- jamin Sapp, Yin Zhou, and Dragomir Anguelov. Motion- diffuser: Controllable multi-agent motion prediction using diffusion. In CVPR, 2023. 2

    2023

  16. [24]

    Scenediffuser: Efficient and controllable driving simulation initialization and rollout

    Chiyu Max Jiang, Yijing Bai, Andre Cornman, Christopher Davis, Xiukun Huang, Hong Jeon, Sakshum Kulshrestha, John Lambert, Shuangyu Li, Xuanyu Zhou, Carlos Fuertes, Chang Yuan, Mingxing Tan, Yin Zhou, and Dragomir Anguelov. Scenediffuser: Efficient and controllable driving sim...

    2024

  17. [25]

    Towards learning- based planning:the nuplan benchmark for real-world au- tonomous driving

    Napat Karnchanachari, Dimitris Geromichalos, Kok Seang Tan, Nanxiang Li, Christopher Eriksen, Shakiba Yaghoubi, Noushin Mehdipour, Gianmarco Bernasconi, Whye Kit Fong, Yiluan Guo, and Holger Caesar. Towards learning- based planning:the nuplan benchmark for real-world au- tonom...

    2024

  18. [26]

    Metadrive: Composing diverse driving scenarios for generalizable reinforcement learning

    Quanyi Li, Zhenghao Peng, Lan Feng, Qihang Zhang, Zhenghai Xue, and Bolei Zhou. Metadrive: Composing diverse driving scenarios for generalizable reinforcement learning. TPAMI, 2022. 3

    2022

  19. [27]

    A deep reinforcement learning network for traffic light cycle control

    Xiaoyuan Liang, Xunsheng Du, Guiling Wang, and Zhu Han. A deep reinforcement learning network for traffic light cycle control. IEEE Transactions on Vehicular Technology, 68(2):1243–1253, 2019. 3

    2019

  20. [28]

    Divergence measures based on the shannon en- tropy

    Jianhua Lin. Divergence measures based on the shannon en- tropy. IEEE Transactions on Information theory, 37(1):145– 151, 1991. 5

    1991

  21. [29]

    Microscopic traffic simulation using sumo

    Pablo Alvarez Lopez, Michael Behrisch, Laura Bieker-Walz, Jakob Erdmann, Yun-Pang Fl¨otter¨od, Robert Hilbrich, Leon- hard L ¨ucken, Johannes Rummel, Peter Wagner, and Eva- marie Wießner. Microscopic traffic simulation using sumo. In 2018 21st international conference on intel...

    2018

  22. [30]

    Scenecontrol: Diffusion for controllable traffic scene generation

    Jack Lu, Kelvin Wong, Chris Zhang, Simon Suo, and Raquel Urtasun. Scenecontrol: Diffusion for controllable traffic scene generation. In ICRA, 2024. 2

    2024

  23. [31]

    Unigen: Unified modeling of initial agent states and trajectories for generating autonomous driving scenarios

    Reza Mahjourian, Rongbing Mu, Valerii Likhosherstov, Paul Mougin, Xiukun Huang, Joao Messias, and Shimon White- son. Unigen: Unified modeling of initial agent states and trajectories for generating autonomous driving scenarios. In ICRA, 2024. 3

    2024

  24. [32]

    The waymo open sim agents challenge

    Nico Montali, John Lambert, Paul Mougin, Alex Kuefler, Nick Rhinehart, Michelle Li, Cole Gulino, Tristan Em- rich, Zoey Yang, Shimon Whiteson, Brandyn White, and Dragomir Anguelov. The waymo open sim agents challenge. In Advances in Neural Information Processing Systems Track ...

    2023

  25. [33]

    Scene transformer: A unified architecture for pre- dicting future trajectories of multiple agents

    Jiquan Ngiam, Vijay Vasudevan, Benjamin Caine, Zheng- dong Zhang, Hao-Tien Lewis Chiang, Jeffrey Ling, Re- becca Roelofs, Alex Bewley, Chenxi Liu, Ashish Venugopal, David J Weiss, Benjamin Sapp, Zhifeng Chen, and Jonathon Shlens. Scene transformer: A unified architecture for p...

    2022

  26. [34]

    A diffusion-model of joint interactive nav- igation

    Matthew Niedoba, Jonathan Lavington, Yunpeng Liu, Vasileios Lioutas, Justice Sefas, Xiaoxuan Liang, Dylan Green, Setareh Dabiri, Berend Zwartsenberg, Adam Scibior, and Frank Wood. A diffusion-model of joint interactive nav- igation. In NeurIPS, 2023. 2

    2023

  27. [35]

    Improving agent behaviors with rl fine-tuning for autonomous driving

    Zhenghao Peng, Wenjie Luo, Yiren Lu, Tianyi Shen, Cole Gulino, Ari Seff, and Justin Fu. Improving agent behaviors with rl fine-tuning for autonomous driving. In ECCV, 2024. 5

    2024

  28. [36]

    Trajeglish: Learning the language of driving scenarios

    Jonah Philion, Xue Bin Peng, and Sanja Fidler. Trajeglish: Learning the language of driving scenarios. In ICLR, 2024. 3

    2024

  29. [37]

    Sampson, Shikai Li, Simone Parmeggiani, Steve Fine, Tara Fowler, Vladan Petro- vic, and Yuming Du

    Adam Polyak, Amit Zohar, Andrew Brown, Andros Tjandra, Animesh Sinha, Ann Lee, Apoorv Vyas, Bowen Shi, Chih- Yao Ma, Ching-Yao Chuang, David Yan, Dhruv Choudhary, Dingkang Wang, Geet Sethi, Guan Pang, Haoyu Ma, Ishan Misra, Ji Hou, Jialiang Wang, Kiran Jagadeesh, Kunpeng Li, L...

    2024

  30. [38]

    Scenario diffusion: Controllable driving scenario gen- eration with diffusion

    Ethan Pronovost, Meghana Reddy Ganesina, Noureldin Hendy, Zeyu Wang, Andres Morales, Kai Wang, and Nick Roy. Scenario diffusion: Controllable driving scenario gen- eration with diffusion. In NeurIPS, 2023. 2, 3

    2023

  31. [39]

    Generating driv- ing scenes with diffusion

    Ethan Pronovost, Kai Wang, and Nick Roy. Generating driv- ing scenes with diffusion. In ICRA Workshop on Scalable Autonomous Driving, 2023. 2

    2023

  32. [40]

    Introducing general world models, 2024

    Runway ML. Introducing general world models, 2024. Ac- cessed on 2024-10-24. 2

    2024

  33. [41]

    Progressive distillation for fast sampling of diffusion models

    Tim Salimans and Jonathan Ho. Progressive distillation for fast sampling of diffusion models. In ICLR, 2022. 4

    2022

  34. [42]

    Adafactor: Adaptive learning rates with sublinear memory cost

    Noam Shazeer and Mitchell Stern. Adafactor: Adaptive learning rates with sublinear memory cost. In ICML, 2018. 11

    2018

  35. [43]

    Drivescenegen: Generating diverse and realistic driving sce- narios from scratch

    Shuo Sun, Zekai Gu, Tianchen Sun, Jiawei Sun, Chen- gran Yuan, Yuhang Han, Dongen Li, and Marcelo H Ang. Drivescenegen: Generating diverse and realistic driving sce- narios from scratch. IEEE Robotics and Automation Letters,

  36. [44]

    Trafficsim: Learning to simulate realistic multi- agent behaviors

    Simon Suo, Sebastian Regalado, Sergio Casas, and Raquel Urtasun. Trafficsim: Learning to simulate realistic multi- agent behaviors. In CVPR, 2021. 3

    2021

  37. [45]

    Scenegen: Learning to generate realistic traffic scenes

    Shuhan Tan, Kelvin Wong, Shenlong Wang, Sivabalan Mani- vasagam, Mengye Ren, and Raquel Urtasun. Scenegen: Learning to generate realistic traffic scenes. In CVPR, 2021. 3

    2021

  38. [46]

    Language conditioned traffic gen- eration

    Shuhan Tan, Boris Ivanovic, Xinshuo Weng, Marco Pavone, and Philipp Kr ¨ahenb¨uhl. Language conditioned traffic gen- eration. 7th Annual Conference on Robot Learning (CoRL),

  39. [47]

    Promptable closed-loop traffic simulation

    Shuhan Tan, Boris Ivanovic, Yuxiao Chen, Boyi Li, Xinshuo Weng, Yulong Cao, Philipp Kr¨ahenb¨uhl, and Marco Pavone. Promptable closed-loop traffic simulation. 8th Annual Con- ference on Robot Learning (CoRL), 2024. 3

    2024

  40. [48]

    Srinivasan, Jonathan T

    Matthew Tancik, Vincent Casser, Xinchen Yan, Sabeek Prad- han, Ben Mildenhall, Pratul P. Srinivasan, Jonathan T. Bar- ron, and Henrik Kretzschmar. Block-nerf: Scalable large scene neural view synthesis. In CVPR, 2022. 3

    2022

  41. [49]

    Con- gested traffic states in empirical observations and micro- scopic simulations

    Martin Treiber, Ansgar Hennecke, and Dirk Helbing. Con- gested traffic states in empirical observations and micro- scopic simulations. Physical review E , 62(2):1805, 2000. 5

    2000

  42. [50]

    Diffusion models are real-time game engines

    Dani Valevski, Yaniv Leviathan, Moab Arar, and Shlomi Fruchter. Diffusion models are real-time game engines. In ICLR, 2025. 2

    2025

  43. [51]

    A survey on traffic signal control methods

    Hua Wei, Guanjie Zheng, Vikash Gayah, and Zhenhui Li. A survey on traffic signal control methods. arXiv preprint arXiv:1904.08117, 2019. 3

    1904 arXiv

  44. [52]

    Argoverse 2: Next generation datasets for self-driving perception and fore- casting

    Benjamin Wilson, William Qi, Tanmay Agarwal, John Lam- bert, Jagjeet Singh, Siddhesh Khandelwal, Bowen Pan, Rat- nesh Kumar, Andrew Hartnett, Jhony Kaesemodel Pontes, Deva Ramanan, Peter Carr, and James Hays. Argoverse 2: Next generation datasets for self-driving perception an...

    2021

  45. [53]

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

    Wei Wu, Xiaoxin Feng, Ziyan Gao, and Yuheng Kan. Smart: Scalable multi-agent real-time motion generation via next- token prediction. In NeurIPS, 2024. 2

    2024

  46. [54]

    Wcdt: World-centric diffusion trans- former for traffic scene generation, 2024

    Chen Yang, Aaron Xuxiang Tian, Dong Chen, Tianyu Shi, and Arsalan Heydarian. Wcdt: World-centric diffusion trans- former for traffic scene generation, 2024. 2

    2024

  47. [55]

    Unisim: A neural closed-loop sensor simulator

    Ze Yang, Yun Chen, Jingkang Wang, Sivabalan Mani- vasagam, Wei-Chiu Ma, Anqi Joyce Yang, and Raquel Ur- tasun. Unisim: A neural closed-loop sensor simulator. In CVPR, 2023. 3

    2023

  48. [56]

    Learning to drive via asymmetric self-play

    Chris Zhang, Sourav Biswas, Kelvin Wong, Kion Fallah, Lunjun Zhang, Dian Chen, Sergio Casas, and Raquel Urta- sun. Learning to drive via asymmetric self-play. In ECCV,

  49. [57]

    Tedi: Temporally-entangled diffusion for long-term motion synthesis

    Zihan Zhang, Richard Liu, Rana Hanocka, and Kfir Aber- man. Tedi: Temporally-entangled diffusion for long-term motion synthesis. In ACM SIGGRAPH 2024 Conference Pa- pers, 2024. 2

    2024

  50. [58]

    Language-guided traffic simulation via scene-level diffusion

    Ziyuan Zhong, Davis Rempe, Yuxiao Chen, Boris Ivanovic, Yulong Cao, Danfei Xu, Marco Pavone, and Baishakhi Ray. Language-guided traffic simulation via scene-level diffusion. In CoRL, 2023. 2

    2023

  51. [59]

    trip end

    Xiaoyu Zhou, Zhiwei Lin, Xiaojun Shan, Yongtao Wang, Deqing Sun, and Ming-Hsuan Yang. Drivinggaussian: Composite gaussian splatting for surrounding dynamic au- tonomous driving scenes. In CVPR, 2024. 3 A. Appendix A.1. Video We provide a video 2 that features a brief and intui...

    2024

Pith tools

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