Pith. sign in

REVIEW 4 major objections 6 minor 53 references

Causal Composition Diffusion Model for Closed-loop Traffic Generation

T0 review · 4 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read A diffusion model that discovers which agents causally matter can generate controllable, realistic safety-critical traffic scenes.

desk verdict A useful masked-guidance recipe for controllable traffic simulation, but the paper's causal story is backed by a kinematic heuristic rather than the formal causal criterion it defines. read the letter →

arxiv 2412.17920 v3 pith:3VPK66WC submitted 2024-12-23 cs.AI cs.LGcs.RO

Haohong Lin , Xin Huang , Tung Phan-Minh , David S. Hayden , Huan Zhang , Ding Zhao , Siddhartha Srinivasa , Eric M. Wolff
show 1 more author
Hongge Chen
This is my paper · ORCID
classification cs.AIcs.LGcs.RO
keywords causalcompositiondiffusionmodelclosed-looptrafficsimulationsafety-criticalscenariogenerationcontrollabilityrealismclassifier-freeguidancetime-to-collisionattentiondecisiongraph
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 tries to show that the conflict between making a traffic simulation obey a user's safety-critical instructions and keeping the simulated behavior realistic can be broken by exploiting which agents actually influence each other. It proposes CCDiff, a diffusion-based generator that first learns a sparse causal graph among agents from time-to-collision masked attention, then applies controllability guidance only to the top-ranked causal agents. On the nuScenes validation set, CCDiff reports higher or comparable controllability and realism than five existing methods, with the largest gains when many agents are controlled and over long horizons. The result matters for autonomous-vehicle safety testing, where a simulator must be able to force rare dangerous interactions without producing implausible traffic.

What carries the argument

The central object is the Decision Causal Graph (DCG): a matrix over agents in which $G_{ij}=1$ means the $i$-th agent's state is a causal parent of the $j$-th agent's action. CCDiff estimates the DCG as the elementwise product of a time-to-collision (TTC) mask and spatial attention weights; the mask keeps only edges where TTC is below a threshold $C_{\mathrm{ttc}}$. The DCG does two jobs: it selects the conditioning parents for an interventional classifier-free guidance, and it feeds a community-detection ranking that decides which agents receive classifier-based reward guidance at inference. Masking guidance to the top-ranked causal agents is what lets the model push for collisions or other safety-critical outcomes while keeping the unguided agents realistic.

What would settle it

One could test this by running CCDiff on scenes where a ground-truth causal graph is known from a controlled simulator, then checking whether replacing the learned TTC mask with the ground truth improves or worsens the controllability-realism trade-off; if the learned graph does not track the ground truth, the causal discovery is not what carries the gain.

Watch

Extended reading notes

Core claim

CCDiff's central claim is that the constrained optimization of controllability subject to a realism bound can be solved more effectively by causal composition than by weighting all objectives together. The paper formalizes closed-loop traffic generation as a Constrained Factored MDP and derives a factored score combining reward terms (controllability) and policy likelihoods (realism). To avoid gradient conflict between these terms, CCDiff discovers a Decision Causal Graph using a time-to-collision mask applied to spatial attention, ranks agents by causal importance via community detection, and then applies classifier-free and classifier-based guidance only to the top-K agents. The paper reports that this yields better controllability and realism scores than SimNet, TrafficSim, STRIVE, BITS, and CTG, and that replacing the causal ranking with distance-based or human ranking degrades performance.

Load-bearing premise

The load-bearing premise is that the time-to-collision masked attention graph identifies the true causal parents of each agent's behavior, and that the community-detection ranking orders agents by how much guiding them improves the safety objective.

Editorial extensions

If this is right

  • If CCDiff is right, safety-critical scenario generation can be scaled to larger sets of controllable agents without the usual realism collapse: its controllability score rises from 0.33 with 2 agents to 0.74 with full scene control, while realism stays near the best baseline.
  • Long-horizon closed-loop generation becomes more reliable: at 4-5 second planning horizons CCDiff's controllability is higher than CTG with comparable realism, suggesting causal masking limits the compounding errors of closed-loop rollout.
  • The guidance can be steered to different safety-critical objectives: the overspeed experiments show the same causal composition mechanism works beyond collision generation.
  • The ranking module, not just the guidance strength, determines the outcome: distance-based and human ranking both degrade collision rate, off-road rate, and FDE, so agent selection is a first-class control knob.

Reading between the lines

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

  • Editorial inference: because the TTC-masked graph is interpretable (nonzero entries mark which agents must react to whom), the same DCG could be reused as a human-readable explanation of why a generated scenario is safety-critical, not just as an internal mask.
  • Editorial inference: the causal composition principle is not tied to diffusion; a world model or autoregressive traffic policy could adopt the same masked-guidance idea whenever gradient conflict appears between a reward and an imitation objective.
  • Editorial inference: the manual threshold $C_{\mathrm{ttc}}$ and the community-detection hyperparameters are the main tuning surface, and a learned or adaptive threshold would be a testable extension that could remove the reported sensitivity.
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

4 major / 6 minor

Summary. The paper proposes CCDiff, a diffusion-based closed-loop traffic simulator for safety-critical scenario generation. It formulates the task as a Constrained Factored MDP, defines a Decision Causal Graph from time-to-collision (TTC) masked attention, ranks agents by a community-detection heuristic, and applies masked classifier guidance to controllability rewards only to top-ranked agents while keeping other agents close to the learned behavior policy. The method is evaluated on nuScenes against SimNet, TrafficSim, STRIVE, BITS, and CTG under varying numbers of controllable agents and planning horizons, using standardized Controllability and Realism Scores plus GD/IGD. Ablations compare the causal ranking with distance-based and human rankings.

Significance. If the causal mechanism were established, structure-guided diffusion for closed-loop traffic would be a meaningful advance over reweighting controllability and realism losses, and the paper provides a clean problem formulation and a broad baseline comparison. Strengths include the constrained MDP formulation, the use of external baselines and a public dataset, the detailed qualitative scenario analyses, and the additional studies of overspeed guidance and gradient-conflict statistics. However, the significance currently rests on an unsupported identification between the formal causal criterion in Eq. (5) and the implemented TTC/clique heuristic, on an unenforced constrained-optimization claim, and on relative standardized scores without variance; these weaken the central contribution.

major comments (4)
  1. [§4.2, Eq. (5) and Algorithm 3] The paper defines the importance ranking via the gradient inner product ρ_i = argmax_i ⟨∇_τ log π^(i), ∇_τ R^(j)⟩, but this quantity is never computed or reported. Algorithm 3 instead ranks agents by greedily growing cliques in G = M ⊙ α, where M is a TTC threshold (Cttc = 3.0 s) and α is an attention matrix; no gradient inner product or conditional-independence test is run. The ablations in Table 6 compare the resulting ranking against distance-based and human rankings, but not against the Eq. (5) criterion, a random ranking with the same Nc, or a ground-truth causal graph. The evidence therefore supports a TTC-clique selector, not the causal ranking claimed in the title and abstract; the conclusion's own admission that the causal reasoning pipeline relies on hyperparameter tuning and is hard to evaluate further underscores this gap.
  2. [§3.1 and §4.2, Eq. (3) and Algorithm 1] The constrained optimization in Eq. (3), announced with a Lagrangian multiplier and structured projected gradient descent, is not reflected in Algorithm 1. The inference loop applies classifier-free and classifier guidance and then enforces vehicle dynamics, but it never forms a Lagrangian, projects onto the constraint set, or checks whether E[C(τ)] ≤ κ is satisfied. Thus the claim that CCDiff maximizes controllability while adhering to realism as a solution of the constrained problem is not supported by the implemented procedure; realism is only measured post hoc via ORR/FDE/CFD. The authors should either implement the constraint or clearly describe the approximation and verify constraint satisfaction.
  3. [§5.1, Tables 1–3] No standard deviations, random seeds, or significance tests are reported for the 100 validation scenes. The Controllability and Realism Scores are standardized within the baseline set, making them relative; the raw metrics in Table 5 are summarized only as best/second best. Without error bars or tests, the claimed substantial gains and the '10 out of 12 metrics' comparisons cannot be assessed for statistical significance. Please report variances across seeds or scene resamples and run paired significance tests for the key comparisons.
  4. [Appendix C.4, Table 9] The post-hoc correlation between TTC-clique count and controllability score (R² = 0.49, p = 2.2 × 10⁻⁷) is offered as evidence for the TTC mask, but it is computed on scenarios generated by the same model in which the TTC mask was already used for guidance. This does not validate causal discovery; it may simply reflect the model's own selection mechanism. A stronger test would compare ranking quality using ground-truth interaction labels or counterfactual interventions, for example by removing a predicted parent and measuring the change in the target agent's action distribution.
minor comments (6)
  1. [Algorithm 1, line 7] The update bπ ← (1 − ω)πϕ,ψ(τ(k), c, k; G) + ωπ_uncond appears to invert the roles of the conditional and unconditional models relative to the text, where classifier-free guidance is (1 − w)∇ log π_uncond + w∇ log π_cond. Please correct the sign convention and define bπ consistently.
  2. [Eq. (4), Definition 2, Eq. (3)] Definition 2 and Eq. (3) treat G as a binary adjacency matrix, but Eq. (4) defines Gij as a continuous masked attention weight Mij · softmax(...). Specify how the continuous graph is binarized, or how the binary constraint |G| ≤ C_sparsity applies to a weighted graph.
  3. [Algorithm 3] The pseudo-code for community detection is underspecified: it does not define the edge weight w(v_j, v) for the product graph, the clique condition is only checked against the current clique Ci, and the final argsort over (C, w) is ambiguous. Please provide a precise expression for the clique score and the tie-breaking rule.
  4. [§5.1, metrics] The 'comfort distance (CFD)' metric is referenced to [3, 29] but never defined; please provide the formula used so the realism score can be reproduced.
  5. [Figure 4] The GD/IGD values printed inside the subcaptions are difficult to parse; please restructure them into a table or a legend.
  6. [References] Reference [14] and several appendix citations contain malformed author lists or stray page-number fragments; please clean up the bibliography.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the benchmark results stand on external baselines and metrics, and the causal-ranking gap is a validation issue rather than a circular reduction.

full rationale

The paper's central claims are evaluated against external baselines (SimNet, TrafficSim, STRIVE, BITS, CTG) and external quantitative metrics (SCR, ORR, FDE, CFD) in a closed-loop simulator. The reported Controllability and Realism Scores are standardized across methods, so CCDiff's own TTC mask and causal ranking are not used to define the reported performance. The constrained optimization in Eq. (3) is not literally solved by Algorithm 1, and the gradient-conflict ranking defined in Eq. (5) is replaced by the TTC-clique heuristic in Algorithm 3; these are implementation and validation gaps, not instances where a prediction reduces to a fitted input by construction. The appendix correlation between TTC-clique count and collision rate is post-hoc and partly self-induced by the guidance, but it is not part of the derivation of the headline benchmark comparisons. Self-citations, such as CausalAF [11], appear only in related-work context and are not load-bearing for the method's correctness. Therefore no load-bearing step of the paper equates a claimed result with its own input by definition or by self-citation.

Assumptions & free parameters 8 free parameters · 6 assumptions · 2 invented entities

The model introduces the DCG and a causal ranking as invented constructs. The graph is a TTC-thresholded attention matrix, and the ranking is a community detection heuristic. Several key hyperparameters (Cttc, Nc, guidance weights, lambda_sparsity) are chosen by hand or by tuning and are load-bearing for the reported results. The formal background (diffusion, control-as-inference) is standard, but the causal assumptions are not independently validated.

free parameters (8)
  • TTC threshold Cttc = 3.0 s
    Hard threshold for the causal mask M in Eq. (4); controls graph sparsity and is tuned by hand.
  • Number of controllable agents Nc (top-K) = 1, 2, 3, 4, 5, 10, Full
    User-specified constraint level; varied during evaluation, not learned.
  • Guidance weight w = not reported (text says range [1,2))
    Classifier-free guidance scale; no value given in Table 8.
  • Collision guidance weight = -50.0
    Weight for agent collision gradient in guidance loss (Table 8).
  • Off-road guidance weight = 1.0
    Weight for map off-road gradient in guidance loss (Table 8).
  • Guidance gradient steps = 30
    Number of gradient steps in guidance update (Table 8).
  • Distance threshold = 50 m
    Cutoff for relative features in the encoders (Table 8).
  • Sparsity regularization lambda_sparsity = not reported
    Used for the causal composition scene encoder; removed in the w/o encoder ablation, value not given.
assumptions (6)
  • domain assumption Each agent's actions depend primarily on the states of a subset of nearby agents (key insight, Section 1).
    Motivates the factored policy and the causal graph; if false, causal masking has no basis.
  • ad hoc to paper The TTC threshold Cttc (3.0 s) identifies causal parent relationships between agents.
    Used in Eq. (4) to build the hard mask M; no evidence that TTC threshold equals causal structure.
  • domain assumption The realism constraint C(tau) is the total variation distance between dataset policy and learned policy (Section 3.1).
    This is a modeling choice that is never computed at inference; the reported realism metrics are ORR, FDE, CFD instead.
  • standard math Control-as-inference optimality variable p(O_t=1 | tau_t) proportional to exp(R(tau_t)) (Eq. (1)).
    Standard in RL-as-inference literature, used to factorize the objective.
  • domain assumption Vehicle dynamics P are deterministic and known (Section 3.1).
    Limits generality to environments with known dynamics models.
  • ad hoc to paper Community detection on the DCG produces a meaningful importance ranking (Algorithm 3).
    The ranking is sketched without formal justification or external validation.
invented entities (2)
  • Decision Causal Graph (DCG) G_ij = M_ij * softmax attention
    purpose: To select causal parents of each agent's action and to rank agent importance for guidance masking.
    The graph is a thresholded attention mask (Eq. 4). The paper provides only a post-hoc correlation (Table 9) between TTC-clique counts and collision rate, not a falsifiable prediction outside this paper.
  • Importance ranking rho (causal ranking via community detection)
    purpose: To choose the top-Nc agents whose gradient guidance is applied, preserving realism for the rest.
    Algorithm 3 is under-specified and no external benchmark validates the ranking; its value is only demonstrated through in-paper ablations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Causal Composition Diffusion Model for Closed-loop Traffic Generation." pith.science (2026). https://pith.science/paper/3VPK66WC

@misc{pith2026241217920,
  author       = {Pith},
  title        = {Pith review of: Causal Composition Diffusion Model for Closed-loop Traffic Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3VPK66WC}},
  note         = {Machine review of arXiv:2412.17920}
}
read the original abstract

Simulation is critical for safety evaluation in autonomous driving, particularly in capturing complex interactive behaviors. However, generating realistic and controllable traffic scenarios in long-tail situations remains a significant challenge. Existing generative models suffer from the conflicting objective between user-defined controllability and realism constraints, which is amplified in safety-critical contexts. In this work, we introduce the Causal Compositional Diffusion Model (CCDiff), a structure-guided diffusion framework to address these challenges. We first formulate the learning of controllable and realistic closed-loop simulation as a constrained optimization problem. Then, CCDiff maximizes controllability while adhering to realism by automatically identifying and injecting causal structures directly into the diffusion process, providing structured guidance to enhance both realism and controllability. Through rigorous evaluations on benchmark datasets and in a closed-loop simulator, CCDiff demonstrates substantial gains over state-of-the-art approaches in generating realistic and user-preferred trajectories. Our results show CCDiff's effectiveness in extracting and leveraging causal structures, showing improved closed-loop performance based on key metrics such as collision rate, off-road rate, FDE, and comfort.

Figures

Figures reproduced from arXiv: 2412.17920 by the authors.

Figure 1
Figure 1. Comparison of safety-critical scenario generation meth [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. (a): Overview of Causal Composition Diffusion Model. The scene encoder encodes the history and then uses causal reasoning for a structured scene encoding and causal ranking. Finally, we exert guidance only to the top-K agents and eliminate the non-causal agents that would not contribute to the guidance objective to maintain better realism. (b): Summing up the score functions over all the agents achieves sub-optimal … view at source ↗
Figure 3
Figure 3. Detailed model structure of CCDiff, which incorporates temporal tokenizer, spatial attention, and action decoding. The decision causal graph helps to extract the spatial patterns to identify the most relevant actions, then use the ranking outputs to mask the output of the action. t means trainable modules, and ` means non-trainable parts during training. the most likely action under a parsimonious decision causal gr… view at source ↗
Figures from the paper (26 more)
Figure 4
Figure 4. Figure 4: Plot of the controllability v.s. realism in the multi-agent and long-horizon generation settings. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: As is shown in [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]
Figure 6
Figure 6. Figure 6: Inference speed with respect to the number of agents. [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]
Figure 7
Figure 7. Figure 7: Comparison of CCDiff and CTG on the controllability and realism under different sizes of controllable agents. We can see that CCDiff can consistently generate realistic cross-traffic violation scenarios, yet CTG can only generate one with shorter planning cycle in 1s …
Figure 8
Figure 8. Figure 8: Comparison of CCDiff and CTG on the controllability and realism under different sizes of controllable agents. We can see that when the number of controllable agents is greater than 1, CCDiff can consistently generate realistic unprotected left-turn violations, yet CTG …
Figure 9
Figure 9. Figure 9: Examples of GD and IGD used to evaluate the multi-objective optimization. Two axes [PITH_FULL_IMAGE:figures/full_fig_p018_9.png]
Figure 10
Figure 10. Figure 10: The number of cliques in the TTC graph is more informative causal features of safety-critical incidents (higher Pearson correlation) [PITH_FULL_IMAGE:figures/full_fig_p019_10.png]
Figure 11
Figure 11. Figure 11: Qualitative of CCDiff and baselines in unprotected left turn scenarios [PITH_FULL_IMAGE:figures/full_fig_p020_11.png]
Figure 12
Figure 12. Figure 12: Qualitative of CCDiff and baselines in cross traffic violation scenarios [PITH_FULL_IMAGE:figures/full_fig_p021_12.png]
Figure 13
Figure 13. Figure 13: Qualitative comparison of CCDiff and CTG under cross traffic violation generation under different sizes of controllable agents [PITH_FULL_IMAGE:figures/full_fig_p022_13.png]
Figure 14
Figure 14. Figure 14: Qualitative comparison of CCDiff and CTG under cross traffic violation generation under different generation horizons [PITH_FULL_IMAGE:figures/full_fig_p023_14.png]
Figure 15
Figure 15. Figure 15: Qualitative of CCDiff and baselines in lane cut-in scenarios [PITH_FULL_IMAGE:figures/full_fig_p024_15.png]
Figure 16
Figure 16. Figure 16: Qualitative comparison of CCDiff and CTG under cross traffic violation generation under different sizes of controllable agents [PITH_FULL_IMAGE:figures/full_fig_p025_16.png]
Figure 17
Figure 17. Figure 17: Qualitative comparison of CCDiff and CTG under cross traffic violation generation under different generation horizons [PITH_FULL_IMAGE:figures/full_fig_p026_17.png]
Figure 18
Figure 18. Figure 18: Qualitative of CCDiff and baselines in the emergency break scenarios [PITH_FULL_IMAGE:figures/full_fig_p027_18.png]
Figure 19
Figure 19. Figure 19: Qualitative comparison of CCDiff and CTG under cross traffic violation generation under different sizes of controllable agents [PITH_FULL_IMAGE:figures/full_fig_p028_19.png]
Figure 20
Figure 20. Figure 20: Qualitative comparison of CCDiff and CTG under cross traffic violation generation under different generation horizons [PITH_FULL_IMAGE:figures/full_fig_p029_20.png]
Figure 21
Figure 21. Figure 21: Qualitative of CCDiff and baselines in the chain-reaction crash scenarios [PITH_FULL_IMAGE:figures/full_fig_p030_21.png]
Figure 22
Figure 22. Figure 22: Qualitative comparison of CCDiff and CTG under cross traffic violation generation under different sizes of controllable agents [PITH_FULL_IMAGE:figures/full_fig_p031_22.png]
Figure 23
Figure 23. Figure 23: Qualitative comparison of CCDiff and CTG under cross traffic violation generation under different generation horizons [PITH_FULL_IMAGE:figures/full_fig_p032_23.png]
Figure 24
Figure 24. Figure 24: Qualitative of CCDiff and baselines in the adjacent left-turn side-wipe scenario [PITH_FULL_IMAGE:figures/full_fig_p033_24.png]
Figure 25
Figure 25. Figure 25: Qualitative comparison of CCDiff and CTG under cross traffic violation generation under different sizes of controllable agents [PITH_FULL_IMAGE:figures/full_fig_p034_25.png]
Figure 26
Figure 26. Figure 26: Qualitative comparison of CCDiff and CTG under cross traffic violation generation under different generation horizons [PITH_FULL_IMAGE:figures/full_fig_p035_26.png]
Figure 27
Figure 27. Figure 27: Qualitative of CCDiff and baselines in the multi-vehicle lane merge-in scenarios [PITH_FULL_IMAGE:figures/full_fig_p036_27.png]
Figure 28
Figure 28. Figure 28: Qualitative comparison of CCDiff and CTG under cross traffic violation generation under different sizes of controllable agents [PITH_FULL_IMAGE:figures/full_fig_p037_28.png]
Figure 29
Figure 29. Figure 29: Qualitative comparison of CCDiff and CTG under cross traffic violation generation under different generation horizons [PITH_FULL_IMAGE:figures/full_fig_p038_29.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 39 canonical work pages

  1. [1]

    Dense reinforcement learning for safety validation of autonomous vehicles

    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,

    work page

  2. [2]

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

    Wenhao Ding, Chejian Xu, Mansur Arief, Haohong Lin, Bo Li, and Ding Zhao. A survey on safety-critical driving scenario generation—a methodological per- spective. IEEE Transactions on Intelligent Transporta- tion Systems, 2023. 1

    work page 2023

  3. [3]

    Guided conditional diffusion for controllable traffic simulation

    Ziyuan Zhong, Davis Rempe, Danfei Xu, Yuxiao Chen, Sushant Veer, Tong Che, Baishakhi Ray, and Marco Pavone. Guided conditional diffusion for controllable traffic simulation. In 2023 IEEE International Con- ference on Robotics and Automation (ICRA) , pages 3560–3566. IEEE, 2023. 1, 2, 3, 6, 7

    work page 2023

  4. [4]

    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 Conference on Robot Learn- ing, pages 144–177. PMLR, 2023. 2, 5, 1

    work page 2023

  5. [5]

    Realgen: Retrieval augmented gen- eration for controllable traffic scenarios

    Wenhao Ding, Yulong Cao, Ding Zhao, Chaowei Xiao, and Marco Pavone. Realgen: Retrieval augmented gen- eration for controllable traffic scenarios. arXiv preprint arXiv:2312.13303, 2023. 1, 2, 6, 7

    arXiv 2023

  6. [6]

    How simulation helps au- tonomous driving: A survey of sim2real, digital twins, and parallel intelligence

    Xuemin Hu, Shen Li, Tingyu Huang, Bo Tang, Roux- ing Huai, and Long Chen. How simulation helps au- tonomous driving: A survey of sim2real, digital twins, and parallel intelligence. IEEE Transactions on Intelli- gent Vehicles, 2023. 1

    work page 2023

  7. [7]

    Carla: An open urban driving simulator

    Alexey Dosovitskiy, German Ros, Felipe Codevilla, Antonio Lopez, and Vladlen Koltun. Carla: An open urban driving simulator. In Conference on robot learn- ing, pages 1–16. PMLR, 2017. 2

    work page 2017

  8. [8]

    Microscopic traffic simulation using sumo

    Pablo Alvarez Lopez, Michael Behrisch, Laura Bieker- Walz, Jakob Erdmann, Yun-Pang Flötteröd, Robert Hilbrich, Leonhard Lücken, Johannes Rummel, Peter Wagner, and Evamarie Wießner. Microscopic traffic simulation using sumo. In 2018 21st international con- ference on intelligent transportation systems (ITSC) , pages 2575–2582. IEEE, 2018. 2

    work page 2018

Show all 53 references
  1. [9]

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

    Jingkang Wang, Ava Pun, James Tu, Sivabalan Mani- vasagam, Abbas Sadat, Sergio Casas, Mengye Ren, and Raquel Urtasun. Advsim: Generating safety-critical scenarios for self-driving vehicles. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, p...

    2021

  2. [10]

    Sce- negen: Learning to generate realistic traffic scenes

    Shuhan Tan, Kelvin Wong, Shenlong Wang, Sivabalan Manivasagam, Mengye Ren, and Raquel Urtasun. Sce- negen: Learning to generate realistic traffic scenes. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, pages 892–901,

  3. [11]

    Causalaf: Causal autoregressive flow for safety-critical driving scenario generation

    Wenhao Ding, Haohong Lin, Bo Li, and Ding Zhao. Causalaf: Causal autoregressive flow for safety-critical driving scenario generation. In Conference on robot learning, pages 812–823. PMLR, 2023. 2, 1

    2023

  4. [12]

    Rulefuser: Inject- ing rules in evidential networks for robust out-of- distribution trajectory prediction

    Jay Patrikar, Sushant Veer, Apoorva Sharma, Marco Pavone, and Sebastian Scherer. Rulefuser: Inject- ing rules in evidential networks for robust out-of- distribution trajectory prediction. arXiv preprint arXiv:2405.11139, 2024

    2024 arXiv

  5. [13]

    Language con- ditioned traffic generation

    Shuhan Tan, Boris Ivanovic, Xinshuo Weng, Marco Pavone, and Philipp Kraehenbuehl. Language con- ditioned traffic generation. In Conference on Robot Learning, pages 2714–2752. PMLR, 2023. 2, 6, 1, 7

    2023

  6. [14]

    nuscenes: A multimodal dataset for autonomous driv- ing

    H Caesar, V Bankiti, AH Lang, S V ora, VE Liong, Q Xu, A Krishnan, Y Pan, G Baldan, and O Beijbom. nuscenes: A multimodal dataset for autonomous driv- ing. arxiv. 2019. 2, 6

    2019

  7. [15]

    Towards robust and adaptive motion forecasting: A causal represen- tation perspective

    Yuejiang Liu, Riccardo Cadei, Jonas Schweizer, Sher- win Bahmani, and Alexandre Alahi. Towards robust and adaptive motion forecasting: A causal represen- tation perspective. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion, pages 17081–170...

    2022

  8. [16]

    Causal intervention for human trajectory prediction with cross attention mechanism

    Chunjiang Ge, Shiji Song, and Gao Huang. Causal intervention for human trajectory prediction with cross attention mechanism. In Proceedings of the AAAI Con- ference on Artificial Intelligence , volume 37, pages 658–666, 2023. 2

    2023

  9. [17]

    Human trajectory prediction via counterfactual anal- ysis

    Guangyi Chen, Junlong Li, Jiwen Lu, and Jie Zhou. Human trajectory prediction via counterfactual anal- ysis. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 9824–9833,

  10. [18]

    Cadet: a causal disentanglement approach for robust trajectory prediction in autonomous driving

    Mozhgan Pourkeshavarz, Junrui Zhang, and Amir Ra- souli. Cadet: a causal disentanglement approach for robust trajectory prediction in autonomous driving. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, pages 14874– 14884, 2024. 2, 5

    2024

  11. [19]

    Large scale interactive motion 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 motion forecasting for autonomous driving: The waymo open motion dataset. In Proceedings of the IEEE/CVF Internat...

    2021

  12. [20]

    Causalagents: A robustness benchmark for motion forecasting using causal relationships

    Rebecca Roelofs, Liting Sun, Ben Caine, Khaled S Refaat, Ben Sapp, Scott Ettinger, and Wei Chai. Causalagents: A robustness benchmark for motion forecasting using causal relationships. arXiv preprint arXiv:2207.03586, 2022. 2

    2022 arXiv

  13. [21]

    Scene transformer: A unified architecture for predicting multiple agent trajectories

    Jiquan Ngiam, Benjamin Caine, Vijay Vasudevan, Zhengdong Zhang, Hao-Tien Lewis Chiang, Jeffrey Ling, Rebecca Roelofs, Alex Bewley, Chenxi Liu, Ashish Venugopal, et al. Scene transformer: A unified architecture for predicting multiple agent trajectories. arXiv preprint arXiv:21...

    2021 arXiv

  14. [22]

    Hivt: Hierarchical vector transformer for multi-agent motion prediction

    Zikang Zhou, Luyao Ye, Jianping Wang, Kui Wu, and Kejie Lu. Hivt: Hierarchical vector transformer for multi-agent motion prediction. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8823–8833, 2022. 5

    2022

  15. [23]

    Query-centric trajectory prediction

    Zikang Zhou, Jianping Wang, Yung-Hui Li, and Yu-Kai Huang. Query-centric trajectory prediction. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17863–17873, 2023

    2023

  16. [24]

    Wayformer: Motion forecasting via simple & efficient attention networks

    Nigamaa Nayakanti, Rami Al-Rfou, Aurick Zhou, Kratarth Goel, Khaled S Refaat, and Benjamin Sapp. Wayformer: Motion forecasting via simple & efficient attention networks. In 2023 IEEE International Con- ference on Robotics and Automation (ICRA) , pages 2980–2987. IEEE, 2023. 2

    2023

  17. [25]

    Realistic extreme behavior generation for im- proved av testing

    Robert Dyro, Matthew Foutter, Ruolin Li, Luigi Di Lillo, Edward Schmerling, Xilin Zhou, and Marco Pavone. Realistic extreme behavior generation for im- proved av testing. arXiv preprint arXiv:2409.10669,

    arXiv

  18. [26]

    Trafficsim: Learning to simulate re- alistic multi-agent behaviors

    Simon Suo, Sebastian Regalado, Sergio Casas, and Raquel Urtasun. Trafficsim: Learning to simulate re- alistic multi-agent behaviors. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 10400–10409, 2021. 2, 6, 1, 7

    2021

  19. [27]

    Simnet: Learning reactive self-driving simulations from real-world observations

    Luca Bergamini, Yawei Ye, Oliver Scheel, Long Chen, Chih Hu, Luca Del Pero, Bła˙zej Osi´nski, Hugo Grim- mett, and Peter Ondruska. Simnet: Learning reactive self-driving simulations from real-world observations. In 2021 IEEE International Conference on Robotics and Automation ...

    2021

  20. [28]

    Generating useful accident-prone driving scenarios via a learned traffic prior

    Davis Rempe, Jonah Philion, Leonidas J Guibas, Sanja Fidler, and Or Litany. Generating useful accident-prone driving scenarios via a learned traffic prior. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17305–17315, 2022. 2, 6, 1

    2022

  21. [29]

    Bits: Bi-level imitation for traffic simulation

    Danfei Xu, Yuxiao Chen, Boris Ivanovic, and Marco Pavone. Bits: Bi-level imitation for traffic simulation. In 2023 IEEE International Conference on Robotics and Automation (ICRA), pages 2929–2936. IEEE, 2023. 2, 3, 6, 1, 7

    2023

  22. [30]

    Generative modeling by estimating gradients of the data distribution

    Yang Song and Stefano Ermon. Generative modeling by estimating gradients of the data distribution. Ad- vances in neural information processing systems, 32,

  23. [31]

    Denoising diffusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in neural information processing systems, 33:6840–6851, 2020. 6

    2020

  24. [32]

    Classifier-free diffu- sion guidance

    Jonathan Ho and Tim Salimans. Classifier-free diffu- sion guidance. arXiv preprint arXiv:2207.12598, 2022. 2, 5, 6

    2022 arXiv

  25. [33]

    Scenario diffusion: Controllable driving scenario generation with diffusion

    Ethan Pronovost, Meghana Reddy Ganesina, Noureldin Hendy, Zeyu Wang, Andres Morales, Kai Wang, and Nick Roy. Scenario diffusion: Controllable driving scenario generation with diffusion. Advances in Neu- ral Information Processing Systems, 36:68873–68894,

  26. [34]

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

    Chiyu Jiang, Andre Cornman, Cheolho Park, Benjamin Sapp, Yin Zhou, Dragomir Anguelov, et al. Motion- diffuser: Controllable multi-agent motion prediction using diffusion. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 9644–9653, 2023. 2

    2023

  27. [35]

    Scenediffuser: Efficient and control- lable driving simulation initialization and rollout

    Chiyu Max Jiang, Yijing Bai, Andre Cornman, Christo- pher Davis, Xiukun Huang, Hong Jeon, Sakshum Kul- shrestha, John Wheatley Lambert, Shuangyu Li, Xu- anyu Zhou, et al. Scenediffuser: Efficient and control- lable driving simulation initialization and rollout. In The Thirty-e...

  28. [36]

    Diffscene: Diffusion-based safety-critical scenario generation for autonomous ve- hicles

    Chejian Xu, Ding Zhao, Alberto Sangiovanni- Vincentelli, and Bo Li. Diffscene: Diffusion-based safety-critical scenario generation for autonomous ve- hicles. In The Second Workshop on New Frontiers in Adversarial Machine Learning, 2023. 2

    2023

  29. [37]

    Controllable safety-critical closed-loop traffic sim- ulation via guided diffusion

    Wei-Jer Chang, Francesco Pittaluga, Masayoshi Tomizuka, Wei Zhan, and Manmohan Chandraker. Controllable safety-critical closed-loop traffic sim- ulation via guided diffusion. arXiv preprint arXiv:2401.00391, 2023

    2023 arXiv

  30. [38]

    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. arXiv preprint arXiv:2410.08453, 2024. 2

    2024 arXiv

  31. [39]

    Versatile scene-consistent traffic scenario generation as optimiza- tion with diffusion

    Zhiyu Huang, Zixu Zhang, Ameya Vaidya, Yuxiao Chen, Chen Lv, and Jaime Fernández Fisac. Versatile scene-consistent traffic scenario generation as optimiza- tion with diffusion. arXiv preprint arXiv:2404.02524,

  32. [40]

    Causal multi-agent reinforcement learning: Review and open problems

    St John Grimbly, Jonathan Shock, and Arnu Pretorius. Causal multi-agent reinforcement learning: Review and open problems. arXiv preprint arXiv:2111.06721,

    arXiv

  33. [41]

    Planning with diffusion for flexible behavior synthesis

    Michael Janner, Yilun Du, Joshua B Tenenbaum, and Sergey Levine. Planning with diffusion for flexible behavior synthesis. arXiv preprint arXiv:2205.09991,

    arXiv

  34. [42]

    Is condi- tional generative modeling all you need for decision- making? arXiv preprint arXiv:2211.15657, 2022

    Anurag Ajay, Yilun Du, Abhi Gupta, Joshua Tenen- baum, Tommi Jaakkola, and Pulkit Agrawal. Is condi- tional generative modeling all you need for decision- making? arXiv preprint arXiv:2211.15657, 2022. 3

    2022 arXiv

  35. [43]

    Reinforcement learning and control as probabilistic inference: Tutorial and review

    Sergey Levine. Reinforcement learning and control as probabilistic inference: Tutorial and review. arXiv preprint arXiv:1805.00909, 2018. 3

    2018 arXiv

  36. [44]

    Gradient surgery for multi-task learning

    Tianhe Yu, Saurabh Kumar, Abhishek Gupta, Sergey Levine, Karol Hausman, and Chelsea Finn. Gradient surgery for multi-task learning. Advances in Neural Information Processing Systems, 33:5824–5836, 2020. 4, 7

    2020

  37. [45]

    Pixe- lasparam: A gradient view on diffusion sampling with guidance

    Anh-Dung Dinh, Daochang Liu, and Chang Xu. Pixe- lasparam: A gradient view on diffusion sampling with guidance. In International Conference on Machine Learning, pages 8120–8137. PMLR, 2023

    2023

  38. [46]

    Gradient shaping for multi-constraint safe reinforcement learn- ing

    Yihang Yao, Zuxin Liu, Zhepeng Cen, Peide Huang, Tingnan Zhang, Wenhao Yu, and Ding Zhao. Gradient shaping for multi-constraint safe reinforcement learn- ing. In 6th Annual Learning for Dynamics & Control Conference, pages 25–39. PMLR, 2024

    2024

  39. [47]

    Decouple-then-merge: Towards bet- ter training for diffusion models

    Qianli Ma, Xuefei Ning, Dongrui Liu, Li Niu, and Linfeng Zhang. Decouple-then-merge: Towards bet- ter training for diffusion models. arXiv preprint arXiv:2410.06664, 2024. 4

    2024 arXiv

  40. [48]

    Constrained policy optimization

    Joshua Achiam, David Held, Aviv Tamar, and Pieter Abbeel. Constrained policy optimization. In Interna- tional Conference on Machine Learning, pages 22–31. PMLR, 2017. 4

    2017

  41. [49]

    Learning model-based sparsity via projected gradient descent

    Sohail Bahmani, Petros T Boufounos, and Bhiksha Raj. Learning model-based sparsity via projected gradient descent. IEEE Transactions on Information Theory , 62(4):2092–2099, 2016. 4

    2016

  42. [50]

    Differentiable causal discovery from interventional data

    Philippe Brouillard, Sébastien Lachapelle, Alexandre Lacoste, Simon Lacoste-Julien, and Alexandre Drouin. Differentiable causal discovery from interventional data. Advances in Neural Information Processing Sys- tems, 33:21865–21877, 2020. 4

    2020

  43. [51]

    Learning neural causal models with active interventions

    Nino Scherrer, Olexa Bilaniuk, Yashas Annadani, Anirudh Goyal, Patrick Schwab, Bernhard Schölkopf, Michael C Mozer, Yoshua Bengio, Stefan Bauer, and Nan Rosemary Ke. Learning neural causal models with active interventions. arXiv preprint arXiv:2109.02429,

    arXiv

  44. [52]

    Causality

    Judea Pearl. Causality. Cambridge university press,

  45. [53]

    Evolutionary algorithms for solving multi-objective problems

    Carlos A Coello Coello. Evolutionary algorithms for solving multi-objective problems. Springer, 2007. 7 CCDiff: Causal Composition Diffusion Model for Closed-loop Traffic Generation (Supplementary Materials) A. Additional Related Works Table 4. Key features of related works in...

    2007

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.