pith. machine review for the scientific record. sign in

arxiv: 2512.07661 · v3 · submitted 2025-12-08 · 💻 cs.CV

Optimization-Guided Diffusion for Interactive Scene Generation

Shihao Li , Naisheng Ye , Tianyu Li , Kashyap Chitta , Tuo An , Peng Su , Boyang Wang , Haiou Liu
show 2 more authors
Chen Lv Hongyang Li
This is my paper

Pith reviewed 2026-05-17 00:30 UTC · model grok-4.3

classification 💻 cs.CV
keywords diffusion modelsscene generationdriving scenariosmulti-agent interactionsoptimization guidancesafety-critical eventsgame-theoretic formulation
0
0 comments X

The pith

Guided optimization during diffusion raises valid driving scenes from 32% to 72%

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

The paper introduces OMEGA, a training-free framework that applies constrained optimization at each reverse diffusion step to enforce physical plausibility and behavioral coherence in multi-agent driving scene generation. This addresses the scarcity of safety-critical events in real datasets by steering standard diffusion outputs toward trajectories that satisfy structural and interaction constraints. For adversarial generation, it casts ego-attacker dynamics as a game-theoretic problem solved in distribution space to approximate Nash equilibria. Experiments on nuPlan and Waymo demonstrate clear lifts in validity and controllability while preserving overall realism.

Core claim

OMEGA re-anchors each reverse diffusion step via constrained optimization, steering the generation towards physically plausible and behaviorally coherent trajectories. Building on this framework, ego-attacker interactions are formulated as a game-theoretic optimization in the distribution space, approximating Nash equilibria to generate realistic, safety-critical adversarial scenarios.

What carries the argument

re-anchoring each reverse diffusion step via constrained optimization to enforce structural consistency and interaction awareness

Load-bearing premise

The constrained optimization problem solved at each diffusion step can be solved accurately and efficiently without distorting the underlying diffusion trajectory or introducing new inconsistencies.

What would settle it

Run the same generation pipeline with and without the per-step optimization on a fixed set of prompts; if the fraction of scenes passing physical and behavioral checks shows no substantial increase or introduces visible artifacts, the guidance mechanism fails to deliver the claimed improvement.

Figures

Figures reproduced from arXiv: 2512.07661 by Boyang Wang, Chen Lv, Haiou Liu, Hongyang Li, Kashyap Chitta, Naisheng Ye, Peng Su, Shihao Li, Tianyu Li, Tuo An.

Figure 1
Figure 1. Figure 1: OMEGA (Ω). Our plug-and-play, training-free guidance module can be dropped into existent diffusion-based driving scene generators to provide significantly more realistic, controllable, and interactive traffic scenarios. Top. Our guidance produces diverse and realistic future scenarios from the same historical initialization, increasing the rate of physically and behaviorally valid scenes on both the nuPlan… view at source ↗
Figure 2
Figure 2. Figure 2: Standard diffusion sampling (top) vs. our optimization-guided OMEGA (bottom). We re-anchor each reverse diffusion step via constrained optimization within a KL-bounded trust region, steering the Markov chain towards behaviorally coherent samples. iteratively refines the scene tensor through denoising, pro￾gressively completing the partially observed scene toward a coherent multi-agent configuration. 3.2. O… view at source ↗
Figure 3
Figure 3. Figure 3: Toy example illustrating the effect of OMEGA on the generative manifold. We design a simple two-dimensional generation task to visualize how our method influences the learned data manifold. (a) Without any guidance, the diffusion model approximates the target data distribution and generates samples along its intrinsic manifold. (b) Using only the guidance objective R(x) pulls samples excessively toward the… view at source ↗
Figure 4
Figure 4. Figure 4: Illustration of two-phase noise scheduling. Darker blue means higher noise, lighter blue means lower noise. serve feasibility and realism within the learned generative manifold, the attacker seeks to induce maximal perturba￾tions that remain distributionally consistent, thereby expos￾ing safety-critical yet physically plausible interactions. This formulation enables adversarial scenario synthesis directly … view at source ↗
read the original abstract

Realistic and diverse multi-agent driving scenes are crucial for evaluating autonomous vehicles, but safety-critical events which are essential for this task are rare and underrepresented in driving datasets. Data-driven scene generation offers a low-cost alternative by synthesizing complex traffic behaviors from existing driving logs. However, existing models often lack controllability or yield samples that violate physical or social constraints, limiting their usability. We present OMEGA, an optimization-guided, training-free framework that enforces structural consistency and interaction awareness during diffusion-based sampling from a scene generation model. OMEGA re-anchors each reverse diffusion step via constrained optimization, steering the generation towards physically plausible and behaviorally coherent trajectories. Building on this framework, we formulate ego-attacker interactions as a game-theoretic optimization in the distribution space, approximating Nash equilibria to generate realistic, safety-critical adversarial scenarios. Experiments on nuPlan and Waymo show that OMEGA improves generation realism, consistency, and controllability, increasing the ratio of physically and behaviorally valid scenes from 32.35% to 72.27% for free exploration capabilities, and from 11% to 80% for controllability-focused generation. Our approach can also generate $5\times$ more near-collision frames with a time-to-collision under three seconds while maintaining the overall scene realism.

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 manuscript introduces OMEGA, a training-free optimization-guided diffusion framework for generating realistic multi-agent driving scenes. It re-anchors each reverse diffusion step by solving a constrained optimization problem to enforce physical and interaction constraints, and extends the approach to a game-theoretic formulation for approximating Nash equilibria in ego-attacker interactions. Experiments on nuPlan and Waymo report increasing the ratio of physically and behaviorally valid scenes from 32.35% to 72.27% for free exploration and from 11% to 80% for controllability-focused generation, along with generating 5× more near-collision frames (time-to-collision under three seconds) while preserving overall scene realism.

Significance. If the central claims hold, this work offers a practical training-free method to steer diffusion-based scene generators toward higher physical and behavioral validity, which is valuable for synthesizing diverse safety-critical scenarios needed for autonomous vehicle evaluation. The approach builds on standard public datasets and standard validity checks without introducing free parameters or ad-hoc axioms.

major comments (2)
  1. [§3.2] §3.2 (re-anchoring procedure): The manuscript provides no convergence analysis, solver tolerance study, or ablation on update magnitude for the constrained optimization solved at each reverse diffusion step. Without this, it is unclear whether the reported validity gains (32.35%→72.27% and 11%→80%) arise from faithful steering consistent with the score function or from the optimization projecting samples into regions where the downstream metrics are easier to satisfy.
  2. [§4.2] §4.2 and Table 2 (experimental results): The validity ratios and 5× near-collision increase are presented without statistical significance tests, run-to-run variance, or an explicit ablation isolating the effect of the per-step optimization from other implementation choices (e.g., baseline diffusion sampler, constraint weighting). This information is load-bearing for confirming that the improvements are robust and attributable to the proposed method.
minor comments (2)
  1. [Abstract] Abstract: The precise definition of 'physically and behaviorally valid scenes' and the exact number of generated scenes used to compute the reported percentages should be stated for reproducibility.
  2. [Figure 4] Figure 4 (qualitative examples): Captions should explicitly label which rows correspond to free-exploration versus controllability-focused settings and which scenes are produced by OMEGA versus baselines.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major comment below and describe the revisions we will make to strengthen the manuscript.

read point-by-point responses

  1. Referee: [§3.2] §3.2 (re-anchoring procedure): The manuscript provides no convergence analysis, solver tolerance study, or ablation on update magnitude for the constrained optimization solved at each reverse diffusion step. Without this, it is unclear whether the reported validity gains (32.35%→72.27% and 11%→80%) arise from faithful steering consistent with the score function or from the optimization projecting samples into regions where the downstream metrics are easier to satisfy.

    Authors: We thank the referee for this observation. The re-anchoring step solves a constrained quadratic program that minimizes Euclidean distance to the diffusion-predicted mean while enforcing the physical and interaction constraints; this formulation is intended to keep the update local and consistent with the score function. In the revised manuscript we will add: (i) convergence plots of the optimization residual across representative diffusion timesteps, (ii) an ablation varying solver tolerance (e.g., 10^{-3} vs. 10^{-4}) and maximum update magnitude, and (iii) a direct comparison of validity metrics obtained with and without the projection step. These additions will demonstrate that the reported gains arise from constraint enforcement within the diffusion manifold rather than from metric exploitation. revision: yes

  2. Referee: [§4.2] §4.2 and Table 2 (experimental results): The validity ratios and 5× near-collision increase are presented without statistical significance tests, run-to-run variance, or an explicit ablation isolating the effect of the per-step optimization from other implementation choices (e.g., baseline diffusion sampler, constraint weighting). This information is load-bearing for confirming that the improvements are robust and attributable to the proposed method.

    Authors: We agree that statistical rigor and targeted ablations are required. In the revision we will: (i) report all metrics as mean ± standard deviation over at least five independent random seeds, (ii) include paired statistical tests (e.g., Wilcoxon signed-rank) with p-values for the key improvements, and (iii) add an explicit ablation that isolates the per-step optimization by comparing the full OMEGA pipeline against variants that apply optimization only at the final step or with uniform constraint weights. These changes will confirm both robustness and the specific contribution of the proposed method. revision: yes

Circularity Check

0 steps flagged

No significant circularity; performance gains validated externally

full rationale

The paper describes OMEGA as a training-free method that applies constrained optimization at each reverse diffusion step to enforce physical and interaction constraints on a pre-trained scene generation model. The reported improvements (32.35% to 72.27% valid scenes for free exploration, 11% to 80% for controllability, and 5× more near-collision frames) are computed on external public datasets (nuPlan, Waymo) using standard validity metrics for collisions, trajectories, and interactions. These metrics are not defined in terms of the optimization variables or any fitted parameters internal to OMEGA. No self-definitional equations, fitted-input predictions, load-bearing self-citations, or ansatz smuggling appear in the abstract or described framework. The derivation chain remains independent of its evaluation outputs and is directly falsifiable against the cited benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The framework rests on standard diffusion sampling assumptions and the premise that constrained optimization can be interleaved without breaking the generative process; no new free parameters, invented entities, or ad-hoc axioms are introduced in the abstract.

axioms (1)
  • domain assumption Reverse diffusion steps can be re-anchored by solving a constrained optimization problem while preserving the overall generative distribution.
    Central operating assumption of the OMEGA framework stated in the abstract.

pith-pipeline@v0.9.0 · 5552 in / 1234 out tokens · 76773 ms · 2026-05-17T00:30:20.021048+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

43 extracted references · 43 canonical work pages · 3 internal anchors

  1. [1]

    nuScenes: A multimodal dataset for autonomous driving

    Holger Caesar, Varun Bankiti, Alex H Lang, Sourabh Vora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Gi- ancarlo Baldan, and Oscar Beijbom. nuScenes: A multimodal dataset for autonomous driving. InCVPR, 2020. 2

    work page 2020

  2. [2]

    NuPlan: A closed-loop ML-based planning benchmark for autonomous vehicles

    Holger Caesar, Juraj Kabzan, Kok Seang Tan, Whye Kit Fong, Eric Wolff, Alex Lang, Luke Fletcher, Oscar Beijbom, and Sammy Omari. nuPlan: A closed-loop ml-based plan- ning benchmark for autonomous vehicles.arXiv preprint arXiv:2106.11810, 2021. 7, 14

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  3. [3]

    Pseudo- simulation for autonomous driving

    Wei Cao, Marcel Hallgarten, Tianyu Li, Daniel Dauner, Xunjiang Gu, Caojun Wang, Yakov Miron, Marco Aiello, Hongyang Li, Igor Gilitschenski, Boris Ivanovic, Marco Pavone, Andreas Geiger, and Kashyap Chitta. Pseudo- simulation for autonomous driving. InCoRL, 2025. 2

    work page 2025

  4. [4]

    End-to-end autonomous driving: Challenges and frontiers.IEEE TPAMI, 2024

    Li Chen, Penghao Wu, Kashyap Chitta, Bernhard Jaeger, Andreas Geiger, and Hongyang Li. End-to-end autonomous driving: Challenges and frontiers.IEEE TPAMI, 2024. 3

    work page 2024

  5. [5]

    Diffusion policy: Visuomotor policy learning via action dif- fusion.The International Journal of Robotics Research, 44 (10-11):1684–1704, 2025

    Cheng Chi, Zhenjia Xu, Siyuan Feng, Eric Cousineau, Yilun Du, Benjamin Burchfiel, Russ Tedrake, and Shuran Song. Diffusion policy: Visuomotor policy learning via action dif- fusion.The International Journal of Robotics Research, 44 (10-11):1684–1704, 2025. 7, 18, 20

    work page 2025

  6. [6]

    SLEDGE: Synthesizing driving environments with genera- tive models and rule-based traffic

    Kashyap Chitta, Daniel Dauner, and Andreas Geiger. SLEDGE: Synthesizing driving environments with genera- tive models and rule-based traffic. InECCV, 2024. 3

    work page 2024

  7. [7]

    Diffusion Posterior Sampling for General Noisy Inverse Problems

    Hyungjin Chung, Jeongsol Kim, Michael T Mccann, Marc L Klasky, and Jong Chul Ye. Diffusion posterior sam- pling for general noisy inverse problems.arXiv preprint arXiv:2209.14687, 2022. 3

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  8. [8]

    NA VSIM: Data-driven non-reactive autonomous vehicle simulation and benchmark- ing

    Daniel Dauner, Marcel Hallgarten, Tianyu Li, Xinshuo Weng, Zhiyu Huang, Zetong Yang, Hongyang Li, Igor Gilitschenski, Boris Ivanovic, Marco Pavone, et al. NA VSIM: Data-driven non-reactive autonomous vehicle simulation and benchmark- ing. InNeurIPS Datasets and Benchmarks, 2024. 2

    work page 2024

  9. [9]

    Diffusion models beat gans on image synthesis

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis. InNeurIPS, 2021. 3

    work page 2021

  10. [10]

    CARLA: An open urban driving simulator

    Alexey Dosovitskiy, German Ros, Felipe Codevilla, Antonio Lopez, and Vladlen Koltun. CARLA: An open urban driving simulator. InCoRL, 2017. 2

    work page 2017

  11. [11]

    Large scale interactive mo- tion forecasting for autonomous driving: The waymo open motion dataset

    Scott Ettinger, Shuyang Cheng, Benjamin Caine, Chenxi Liu, Hang Zhao, Sabeek Pradhan, Yuning Chai, Ben Sapp, Charles R Qi, Yin Zhou, et al. Large scale interactive mo- tion forecasting for autonomous driving: The waymo open motion dataset. InICCV, 2021. 8

    work page 2021

  12. [12]

    TrafficGen: Learning to generate diverse and realistic traffic scenarios.arXiv preprint arXiv:2210.06609, 2022

    Lan Feng, Quanyi Li, Zhenghao Peng, Shuhan Tan, and Bolei Zhou. TrafficGen: Learning to generate diverse and realistic traffic scenarios.arXiv preprint arXiv:2210.06609, 2022. 2

    work page arXiv 2022

  13. [13]

    Dense reinforcement learning for safety validation of autonomous vehicles.Nature, 615(7953):620–627, 2023

    Shuo Feng, Haowei Sun, Xintao Yan, Haojie Zhu, Zhengxia Zou, Shengyin Shen, and Henry X Liu. Dense reinforcement learning for safety validation of autonomous vehicles.Nature, 615(7953):620–627, 2023. 3

    work page 2023

  14. [14]

    Introduction to sensitivity and stability analysis in non linear programming

    Anthony V Fiacco. Introduction to sensitivity and stability analysis in non linear programming. 1983. 2

    work page 1983

  15. [15]

    King: Generating safety-critical driving scenarios for robust imitation via kine- matics gradients

    Niklas Hanselmann, Katrin Renz, Kashyap Chitta, Apra- tim Bhattacharyya, and Andreas Geiger. King: Generating safety-critical driving scenarios for robust imitation via kine- matics gradients. InECCV, 2022. 3

    work page 2022

  16. [16]

    Denoising diffu- sion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffu- sion probabilistic models. InNeurIPS, 2020. 3

    work page 2020

  17. [17]

    Planning-oriented autonomous driving

    Yihan Hu, Jiazhi Yang, Li Chen, Keyu Li, Chonghao Sima, Xizhou Zhu, Siqi Chai, Senyao Du, Tianwei Lin, et al. Planning-oriented autonomous driving. InCVPR, 2023. 3

    work page 2023

  18. [18]

    Solv- ing motion planning tasks with a scalable generative model

    Yihan Hu, Siqi Chai, Zhening Yang, Jingyu Qian, Kun Li, Wenxin Shao, Haichao Zhang, Wei Xu, and Qiang Liu. Solv- ing motion planning tasks with a scalable generative model. InECCV, 2024. 2, 3

    work page 2024

  19. [19]

    Versatile scene- consistent traffic scenario generation as optimization with diffusion.arXiv preprint arXiv:2404.02524, 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.arXiv preprint arXiv:2404.02524, 2024. 2

    work page arXiv 2024

  20. [20]

    MotionDiffuser: Con- trollable multi-agent motion prediction using diffusion

    Chiyu Jiang, Andre Cornman, Cheolho Park, Benjamin Sapp, Yin Zhou, Dragomir Anguelov, et al. MotionDiffuser: Con- trollable multi-agent motion prediction using diffusion. In CVPR, 2023. 2

    work page 2023

  21. [21]

    SceneDiffuser: Ef- ficient and controllable driving simulation initialization and rollout

    Max Jiang, Yijing Bai, Andre Cornman, Christopher Davis, Xiukun Huang, Hong Jeon, Sakshum Kulshrestha, John Lam- bert, Shuangyu Li, Xuanyu Zhou, et al. SceneDiffuser: Ef- ficient and controllable driving simulation initialization and rollout. InNeurIPS, 2024. 2, 3, 7, 8, 18, 20 9

    work page 2024

  22. [22]

    Mtgs: Multi-traversal gaussian splatting.arXiv preprint arXiv:2503.12552, 2025

    Tianyu Li, Yihang Qiu, Zhenhua Wu, Carl Lind- str¨om, Peng Su, Matthias Nießner, and Hongyang Li. Mtgs: Multi-traversal gaussian splatting.arXiv preprint arXiv:2503.12552, 2025. 2

    work page arXiv 2025

  23. [23]

    Model-based policy adaptation for closed-loop end-to-end autonomous driving

    Haohong Lin, Yunzhi Zhang, Wenhao Ding, Jiajun Wu, and Ding Zhao. Model-based policy adaptation for closed-loop end-to-end autonomous driving. InNeurIPS, 2025. 2

    work page 2025

  24. [24]

    Reinforced refinement with self-aware ex- pansion for end-to-end autonomous driving.arXiv preprint arXiv:2506.09800, 2025

    Haochen Liu, Tianyu Li, Haohan Yang, Li Chen, Caojun Wang, Ke Guo, Haochen Tian, Hongchen Li, Hongyang Li, and Chen Lv. Reinforced refinement with self-aware ex- pansion for end-to-end autonomous driving.arXiv preprint arXiv:2506.09800, 2025. 3

    work page arXiv 2025

  25. [25]

    Curse of rarity for autonomous vehicles.Nature communications, 15(1):4808, 2024

    Henry X Liu and Shuo Feng. Curse of rarity for autonomous vehicles.Nature communications, 15(1):4808, 2024. 2

    work page 2024

  26. [26]

    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 Evamarie Wießner. Microscopic traffic simulation using sumo. InITSC,

    work page

  27. [27]

    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. InICRA, 2024. 17

    work page 2024

  28. [28]

    Unleashing generalization of end-to-end au- tonomous driving with controllable long video generation

    Enhui Ma, Lijun Zhou, Tao Tang, Zhan Zhang, Dong Han, Junpeng Jiang, Kun Zhan, Peng Jia, Xianpeng Lang, Haiyang Sun, et al. Unleashing generalization of end-to-end au- tonomous driving with controllable long video generation. arXiv preprint arXiv:2406.01349, 2024. 2

    work page arXiv 2024

  29. [29]

    Scenario dreamer: Vector- ized latent diffusion for generating driving simulation envi- ronments

    Luke Rowe, Roger Girgis, Anthony Gosselin, Liam Paull, Christopher Pal, and Felix Heide. Scenario dreamer: Vector- ized latent diffusion for generating driving simulation envi- ronments. InCVPR, 2025. 17

    work page 2025

  30. [30]

    MotionLM: Multi-agent motion forecasting as language modeling

    Ari Seff, Brian Cera, Dian Chen, Mason Ng, Aurick Zhou, Nigamaa Nayakanti, Khaled S Refaat, Rami Al-Rfou, and Benjamin Sapp. MotionLM: Multi-agent motion forecasting as language modeling. InICCV, 2023. 2, 3

    work page 2023

  31. [31]

    Deep unsupervised learning using nonequilibrium thermodynamics

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. InICML, 2015. 3

    work page 2015

  32. [32]

    Denoising diffusion implicit models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. InICLR, 2021. 3

    work page 2021

  33. [33]

    A real-time game theoretic planner for autonomous two-player drone racing.IEEE Transactions on Robotics, 36(5):1389–1403, 2020

    Riccardo Spica, Eric Cristofalo, Zijian Wang, Eduardo Mon- tijano, and Mac Schwager. A real-time game theoretic planner for autonomous two-player drone racing.IEEE Transactions on Robotics, 36(5):1389–1403, 2020. 2, 6, 12

    work page 2020

  34. [34]

    SceneDiffuser++: City-scale traffic simulation via a generative world model

    Shuhan Tan, John Lambert, Hong Jeon, Sakshum Kul- shrestha, Yijing Bai, Jing Luo, Dragomir Anguelov, Mingx- ing Tan, and Chiyu Max Jiang. SceneDiffuser++: City-scale traffic simulation via a generative world model. InCVPR,

    work page

  35. [35]

    SimScale: Learning to Drive via Real-World Simulation at Scale

    Haochen Tian, Tianyu Li, Haochen Liu, Jiazhi Yang, Yihang Qiu, Guang Li, Junli Wang, Yinfeng Gao, Zhang Zhang, Liang Wang, Hangjun Ye, Tieniu Tan, Long Chen, and Hongyang Li. Simscale: Learning to drive via real-world simulation at scale.arXiv preprint arXiv:2511.23369, 2025. 2

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  36. [36]

    AdvSim: Generating safety-critical scenarios for self- driving vehicles

    Jingkang Wang, Ava Pun, James Tu, Sivabalan Manivasagam, Abbas Sadat, Sergio Casas, Mengye Ren, and Raquel Urta- sun. AdvSim: Generating safety-critical scenarios for self- driving vehicles. InCVPR, 2021. 3

    work page 2021

  37. [37]

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

    Yuting Xie, Xianda Guo, Cong Wang, Kunhua Liu, and Long Chen. Advdiffuser: Generating adversarial safety-critical driving scenarios via guided diffusion. InIROS. 2, 3

    work page

  38. [38]

    DiffScene: Diffusion-based safety-critical scenario genera- tion for autonomous vehicles

    Chejian Xu, Aleksandr Petiushko, Ding Zhao, and Bo Li. DiffScene: Diffusion-based safety-critical scenario genera- tion for autonomous vehicles. InAAAI, 2025. 2

    work page 2025

  39. [39]

    Resim: Reliable world simulation for autonomous driving

    Jiazhi Yang, Kashyap Chitta, Shenyuan Gao, Long Chen, Yuqian Shao, Xiaosong Jia, Hongyang Li, Andreas Geiger, Xiangyu Yue, and Li Chen. Resim: Reliable world simulation for autonomous driving. 2025. 2

    work page 2025

  40. [40]

    CAT: Closed-loop adversarial training for safe end-to-end driving

    Linrui Zhang, Zhenghao Peng, Quanyi Li, and Bolei Zhou. CAT: Closed-loop adversarial training for safe end-to-end driving. InCoRL, 2023. 3, 18

    work page 2023

  41. [41]

    Diffusion-based planning for autonomous driving with flexi- ble guidance

    Yinan Zheng, Ruiming Liang, Kexin ZHENG, Jinliang Zheng, Liyuan Mao, Jianxiong Li, Weihao Gu, Rui Ai, Shengbo Eben Li, Xianyuan Zhan, and Jingjing Liu. Diffusion-based planning for autonomous driving with flexi- ble guidance. InICLR, 2025. 3, 8, 20

    work page 2025

  42. [42]

    Guided conditional diffusion for controllable traffic simula- tion.arXiv preprint arXiv:2210.17366, 2022

    Ziyuan Zhong, Davis Rempe, Danfei Xu, Yuxiao Chen, Sushant Veer, Tong Che, Baishakhi Ray, and Marco Pavone. Guided conditional diffusion for controllable traffic simula- tion.arXiv preprint arXiv:2210.17366, 2022. 2, 3, 8, 18, 20

    work page arXiv 2022

  43. [43]

    Decoupled diffusion sparks adap- tive scene generation

    Yunsong Zhou, Naisheng Ye, William Ljungbergh, Tianyu Li, Jiazhi Yang, Zetong Yang, Hongzi Zhu, Christoffer Pe- tersson, and Hongyang Li. Decoupled diffusion sparks adap- tive scene generation. InICCV, 2025. 2, 3, 7, 8, 14, 18, 20 10 Optimization-Guided Diffusion for Interactive Scene Generation Supplementary Material A. Theoretical Foundations A.1. Deriv...

    work page 2025