Pith. sign in

REVIEW 5 major objections 6 minor 22 references

DeepMF: Deep Motion Factorization for Closed-Loop Safety-Critical Driving Scenario Simulation

T0 review · 5 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 DeepMF casts safety-critical traffic simulation as a Bayesian factorization, then maximizes the posterior risk to turn any observed scene into an accident-prone one in seconds, without replaying recorded crashes.

desk verdict DeepMF is a solid closed-loop adversarial scenario generator with substantial experiments, but the theory overclaims and the conditional AV predictor is never validated against the simulator. read the letter →

arxiv 2412.17487 v1 pith:FAA2D3LH submitted 2024-12-23 cs.AI cs.LG

Yizhe Li , Linrui Zhang , Xueqian Wang , Houde Liu , Bin Liang This is my paper · ORCID
classification cs.AIcs.LG
keywords safety-criticalscenariogenerationclosed-loopsimulationBayesianfactorizationadversarialdrivingmotionpredictionautonomousrobustnessopponentselectionMetaDrive
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

DeepMF is a framework for generating safety-critical driving scenarios on demand. It treats the problem as maximizing the posterior probability of a collision, factorized into four tractable pieces: which surrounding vehicle should attack, how that opponent will move, how the autonomous vehicle will react, and whether the resulting pair collides. Because each piece is a decoupled neural network fed only current observations and history, the whole pipeline runs in seconds and can re-plan whenever the scene changes, making it closed-loop rather than a replay of past accident clips. The paper reports that on 1,500 Waymo scenarios imported into MetaDrive, DeepMF achieves the highest collision rate and better naturalness than STRIVE, ART, BBO, BGA, and BRS. If correct, this gives AD developers a fast, interactive stress-test generator for open-ended traffic situations.

What carries the argument

The load-bearing object is the Bayesian factorization in Eq. (5): $\max_{Y_{\mathrm{OV}}} P(Y_{\mathrm{OV}}|X) \int P(Y_{\mathrm{AV}}|Y_{\mathrm{OV}},X) P(\mathrm{Coll}=\mathrm{True}|Y_{\mathrm{AV}},Y_{\mathrm{OV}},X)\, dY_{\mathrm{AV}}$, where $X$ is the observed scene. It separates opponent selection from marginal opponent prediction, conditional AV reaction prediction, and collision likelihood. These terms are implemented by decoupled networks: an adversarial-score predictor trained with pseudo-labels from a geometric proximity heuristic, two DenseTNT-style trajectory predictors (one marginal for the opponent, one conditional for the AV), and a bounding-box intersection check. The formula is what turns scenario generation into an optimization over predicted trajectories rather than a search over raw simulator parameters.

What would settle it

Take a set of natural driving logs where a vehicle follows closely but never intends to cause a collision, run DeepMF-S1 with the IDM planner, and check whether the selected opponent is consistently that close follower and whether the reported collision rate holds. If the collision rate drops far below 88% or the selected opponent is often a vehicle that a human labeler would not consider hostile, the pseudo-label heuristic is not transferring.

Watch

Extended reading notes

Core claim

The paper's central claim is that safety-critical scenario generation can be reduced to maximizing the compounded posterior probability of traffic risk, written as the maximum over opponent futures of the prior probability of that future, times the conditional probability of the AV's reaction given it, times the probability of collision given both. This factorization turns an intractable joint search over ego and opponent trajectories into a sequence of standard prediction sub-problems: an opponent selection module scores every surrounding vehicle and picks the most dangerous one; a marginal trajectory predictor proposes the opponent's likely futures; a conditional predictor generates the AV's reaction to each; and a collision check selects the opponent future with the highest combined score. The authors argue that this decomposition, with all components trained on real driving logs and re-run periodically during simulation, is what lets DeepMF generate natural, high-risk scenarios at any triggered time and for any duration in closed loop.

Load-bearing premise

The whole framework rests on the assumption that a vehicle flagged as risky by a simple geometric rule—predicted trajectory bounding boxes overlapping the AV's and a centroid distance below the AV length—is the right opponent to attack, and that a network trained on these labels will keep working when the scene is replayed and repeatedly changed in closed loop.

Editorial extensions

If this is right

  • Closed-loop generation: because DeepMF replans at every update cycle (e.g., every 1, 2, or 4 seconds), it can respond to the AV's actual planner behavior instead of assuming a fixed future, and the reported 93% and 88% collision rates with replay and IDM planners show the attack adapts to both.
  • Speed: generating a scenario takes about 3.6 seconds, versus tens of seconds to minutes for STRIVE and ART, making online adversarial testing during development feasible.
  • No reliance on recorded crashes: since inputs are only current observation and history, DeepMF can attack newly encountered scenes rather than only snippets of previously recorded accident-prone traffic.
  • Naturalness: the generated opponents' action and trajectory distributions are closer to real human driving logs than the compared baselines, according to the KL, Wasserstein, SSP, and Hausdorff metrics, which supports using DeepMF for evaluating human-like risk.
  • Diversity: attacks from eight directions (front, rear, left, right, and diagonals) are produced, giving broader coverage of accident configurations.

Reading between the lines

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

  • An implicit consequence is that the same factorization could be applied to other reactive agents, such as pedestrians, cyclists, or multiple coordinated opponents, since the four terms are decoupled and only need retraining of the predictors.
  • The geometric pseudo-label heuristic that defines a risky opponent by bounding-box overlap and a centroid distance below the AV length may underestimate attacks that begin from further away, such as a fast lane-change from two lanes over; a learned or motion-forecast-based labeler might broaden the attack space.
  • Because the framework maximizes collision probability, it will naturally focus on scenarios where collision is easiest; adversarial safety testing may want a modified objective that also rewards near-misses or sensor-failure triggers, not only actual crashes.
  • The closed-loop replanning frequency is a new control knob: the comparison between the S1, S2, S4, and open-loop versions suggests that more frequent replanning raises collision rate but also computation time, so a deployment choice can be tuned to available compute.
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 / 6 minor

Summary. DeepMF proposes a closed-loop, safety-critical driving scenario simulation framework that factorizes the task into opponent selection, marginal opponent trajectory prediction, conditional AV reaction prediction, and collision probability evaluation. The framework is formulated as a Bayesian posterior maximization (Eq. (5)), implemented by Algorithm 1, and evaluated on 1,500 Waymo scenarios imported into MetaDrive under replay and IDM planners. The paper reports that DeepMF achieves the highest collision rate and the most natural trajectories compared with STRIVE, ART, BBO, BGA, and BRS, while being significantly faster than all baselines except BRS.

Significance. If the claims are supported, DeepMF would be a practically useful and fast closed-loop adversarial scenario generator for testing the robustness of autonomous driving systems. The paper's modular design, use of public data and simulator, and explicit reporting of runtime and collision statistics are strengths. The framework is a plausible engineering contribution. However, the central theoretical derivation is mathematically flawed, the implemented selection rule is not shown to match the stated objective, and the empirical evaluation omits error bars and a direct validation of the conditional AV prediction model under distribution shift. These issues currently prevent the paper from substantiating its central claim of maximizing the compounded posterior risk.

major comments (5)
  1. [III, Eq. (5)] The derivation of Eq. (5) is not a valid statement of max E(Coll|X). Starting from Eq. (2), the expected collision probability is an integral over both Y_OV and Y_AV. Replacing the integral over Y_OV with a max over Y_OV changes the optimization objective: the quantity being maximized in Eq. (5) is not equivalent to the maximum of E(Coll|X) as defined in Eq. (1). If the intended objective is to select the single most dangerous OV trajectory, the formulation should state this explicitly, e.g., max_{Y_OV} P(Y_OV|X) * P(Coll=True|Y_OV,X), or a corresponding integral over Y_AV. As written, the theoretical justification for the Bayesian factorization is incorrect and must be corrected.
  2. [Algorithm 1, line 13; Section IV.C.4] The implemented selection criterion is a heuristic product of trajectory scores and a collision indicator: P(Y_OV^j) * sum_k P(Y_AV^k) * Coll_{jk}. This is not the integral in Eq. (5), which would require integrating over the conditional distribution of Y_AV given Y_OV and X. The paper does not specify the sample counts N1 and N2, establish that the Monte Carlo estimate is unbiased, or justify why the unnormalized product of scores equals the posterior probability. Moreover, the notation P(Y_AV^k) in Algorithm 1 omits the conditioning on Y_OV^j and X, which obscures what the conditional predictor actually computes. The relationship between the stated Bayesian objective and the implemented selection rule must be made explicit and justified.
  3. [IV.C.2 and Tables I-II] The conditional AV trajectory predictor is trained on ground-truth OV future trajectories as conditioning input, but at inference it conditions on sampled, adversarially selected OV predictions, which lie outside its training distribution. The paper never quantifies this mismatch: there is no comparison of predicted AV trajectories against the actual simulator trajectories, no correlation analysis between predicted collision indicators and actual collision outcomes, and no ablation that replaces the conditional predictor with a direct simulator roll-out. Without such validation, the claim that the selection procedure maximizes the probability of an actual collision is unsupported; the observed collision rates might arise from the marginal OV prior or from biases in the collision criterion rather than from the stated Bayesian maximization.
  4. [Tables I, II, III] The evaluation reports single numerical values for all metrics, without error bars, confidence intervals, or statistical significance tests. For example, the collision-rate advantage of DeepMF over STRIVE (93% vs. 86% under the replay planner) may be within sampling noise given 1,500 scenarios, and the naturalness metrics show differences that likewise require variance estimates. The authors should report means and standard deviations over multiple independent replicates (or per-scenario bootstrap confidence intervals) and apply appropriate paired significance tests before claiming that DeepMF 'excels' over the baselines.
  5. [V.A.3] The baseline implementations BBO, BGA, and BRS are described only as bicycle-based models built on the AdvSim framework. No hyperparameters are reported (e.g., optimization budget, number of function evaluations, population size, termination criteria), which makes the runtime comparison and the performance differences difficult to interpret. Without these details, a reader cannot assess whether the baselines were given a fair computational budget or whether their results are representative of the published methods.
minor comments (6)
  1. [IV.C.3] The collision criterion is defined only as 'bounding boxes overlap at the same time.' The paper should specify the bounding-box dimensions (e.g., fixed footprint or dynamic), the temporal discretization used, and whether simultaneous overlap across multiple sampled time steps is required.
  2. [Algorithm 1, line 2] The notation 'IND ∼ Softmax{s1,...,sM}' is ambiguous. The text elsewhere says the opponent with the highest score is selected, so line 2 appears to describe a probabilistic sample rather than the deterministic argmax used in Sections IV.B.3 and IV.B.2. Please align the notation with the described behavior.
  3. [IV.B.1] The heuristic pseudo-label generation is described in prose; a formal definition of the trajectory-bounding-box overlap and the centroid-distance threshold ('less than the length of the AV') would make the labeling rule reproducible.
  4. [Abstract and Section I] The claim that DeepMF 'can effectively and efficiently simulate safety-critical traffic scenarios at any triggered time and for any duration' is not supported by the experiments, which are limited to fixed 9-second episodes (1 s history, 8 s future). Please clarify the intended scope and provide a demonstration for longer or variable durations.
  5. [Tables I-III] The tables contain typographical inconsistencies such as spaces within the numerical values (e.g., '5 .70' and '1 .97'). These should be corrected.
  6. [V.C] The paper says DeepMF-S1 'excels in both action similarity and trajectory similarity' based on Table I, but the differences among the four DeepMF variants are very small (e.g., KL 1.12 vs. 1.14) and no statistical assessment is provided. Please either add significance testing or soften the comparative claims.

Circularity Check

1 steps flagged · score 2.0 of 10

DeepMF's Bayesian factorization is a definitional decomposition with independent empirical support; the only author-overlapping citation is motivational and not load-bearing.

  1. other [Section I (Introduction), paragraph invoking previous work [9].]
    "In our previous work [9], we found that combining the two methods mentioned above, that is, converting a large number of normal scenes into safety-critical scenes in the simulator, can effectively test AD systems and conduct adversarial training. However, the previous work [9] still has many limitations, such as adversary selection relying on manual labels and risk scenarios being open-loop, which means it cannot respond to the reaction of A V in real time and is limited in generation time."

    This citation [9] (CAT) shares the author L. Zhang with the present paper and is the only author-overlapping reference. It is invoked solely to motivate the closed-loop formulation and to enumerate CAT's limitations; it supplies no fitted constant, no uniqueness theorem, and no probability factor that DeepMF's Eq. (5) reduces to. DeepMF's derivation is the exact chain rule of probability (Eqs. 2-5), its predictors are external architectures (DenseTNT, VectorNet, VGG16), and its headline numbers (93%/88% collision rates, naturalness metrics) are produced in MetaDrive against external baselines on 1,500 Waymo scenarios. The self-citation is therefore a minor, non-load-bearing reference, not a forced result.

full rationale

DeepMF's derivation chain is a definitional probability identity rather than a fitted prediction. Equations (2)-(5) expand E(Coll|X) by the chain rule: P(Coll, Y_AV, Y_OV|X) = P(Y_OV|X) * P(Y_AV|Y_OV, X) * P(Coll|Y_AV, Y_OV, X), and Algorithm 1 (line 13) maximizes this surrogate over sampled OV trajectories. No parameter is fitted to the evaluation metric: the opponent-scoring network is trained on transparent geometric pseudo-labels (trajectory-box overlap plus centroid distance, Section IV.B.1), and the marginal and conditional trajectory predictors are standard DenseTNT models over Waymo data. The target quantity (collision) and the evaluation quantity (collision rate in MetaDrive) name the same event, but the simulator outcome is emergent: it depends on the actual AV planner (replay or IDM), the closed-loop replanning cycle, and vehicle kinematics, so a high collision rate is not forced by construction. The only author-overlapping reference is [9] (CAT, co-authored by L. Zhang), used motivationally and to state CAT's limitations; it is not an unverified premise on which DeepMF's claims rest. The experiments are self-contained against external baselines (STRIVE, ART, BBO, BGA, BRS) on 1,500 Waymo scenarios with eight metrics. The skeptic's concern that the conditional AV predictor is trained on ground-truth OV futures but conditioned on sampled OV trajectories at inference (Section IV.C.2 and Algorithm 1, line 8) is a genuine distribution-shift risk, but it is not a circular reduction: the selection criterion could be wrong for reasons independent of its definition, and the measured collision rate is an emergent simulator outcome. The pseudo-label heuristic (Section IV.B.1) transparently defines 'risky' geometrically, so calling it self-referential is fair, yet the learned scorer is a real classifier and the closed-loop evaluation adds content beyond the label generator. Overall circularity is minor.

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

No new physical or conceptual entities are introduced; the adversarial score is a learned model output rather than a postulated force, particle, or conserved quantity. The central method rests on standard probabilistic identities plus domain assumptions about simulator fidelity and the transferability of models trained on natural logs to closed-loop simulation. Several hyperparameters are hand-chosen and not specified in the text.

free parameters (5)
  • Update cycle T = 1 s (DeepMF-S1), also 2 s, 4 s, open-loop
    Replanning frequency is a hand-chosen design parameter that directly controls the closed-loop vs. open-loop trade-off; no automated selection, and results in Table I depend on it.
  • Pseudo-label thresholds for opponent selection = not specified
    Section IV.B.1 defines positive/negative samples via trajectory bounding-box overlap and centroid distance less than the AV length, but gives no explicit threshold values or overlap criterion.
  • Focal loss hyperparameters alpha and gamma = not given
    Eq. (6) uses alpha_t and gamma to balance classes and focus on hard examples, but no values are reported in Section IV.B.2.
  • Trajectory sample counts N1, N2 = not given
    Algorithm 1 samples N1 opponent trajectories and N2 conditional AV trajectories, but the paper never states these counts; they are free choices that affect runtime and risk.
  • Collision criterion = bounding-box overlap at same timestamp
    Section IV.C.3 defines a collision as overlapping bounding boxes at the same time; the geometric margin and timestep discretization are not specified.
assumptions (4)
  • standard math Probability chain rule and Bayes' rule are valid for the scenario decomposition.
    Eqs. (2)-(4) decompose the collision probability; this is mathematically exact, though Eq. (5) misuses it as an optimization objective.
  • domain assumption Bounding-box overlap in the simulator is a faithful proxy for real traffic collisions.
    Section IV.C.3 and the efficiency metrics equate collision with geometric bounding-box overlap at a shared time, ignoring vehicle deformation, physics, and post-impact motion.
  • domain assumption Models trained on offline Waymo natural logs transfer to closed-loop replanning in MetaDrive.
    The opponent predictor and DenseTNT predictors are trained on natural driving data (Sec. IV.B/C) and then applied discretely in a closed loop in MetaDrive with IDM or replay planners, with no fine-tuning or distribution-shift correction.
  • domain assumption MetaDrive simulator faithfully represents Waymo traffic scenarios.
    Section V.A.2 imports Waymo scenarios into MetaDrive; no validation of how faithfully the simulator reproduces the original traffic dynamics is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DeepMF: Deep Motion Factorization for Closed-Loop Safety-Critical Driving Scenario Simulation." pith.science (2026). https://pith.science/paper/FAA2D3LH

@misc{pith2026241217487,
  author       = {Pith},
  title        = {Pith review of: DeepMF: Deep Motion Factorization for Closed-Loop Safety-Critical Driving Scenario Simulation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FAA2D3LH}},
  note         = {Machine review of arXiv:2412.17487}
}
read the original abstract

Safety-critical traffic scenarios are of great practical relevance to evaluating the robustness of autonomous driving (AD) systems. Given that these long-tail events are extremely rare in real-world traffic data, there is a growing body of work dedicated to the automatic traffic scenario generation. However, nearly all existing algorithms for generating safety-critical scenarios rely on snippets of previously recorded traffic events, transforming normal traffic flow into accident-prone situations directly. In other words, safety-critical traffic scenario generation is hindsight and not applicable to newly encountered and open-ended traffic events.In this paper, we propose the Deep Motion Factorization (DeepMF) framework, which extends static safety-critical driving scenario generation to closed-loop and interactive adversarial traffic simulation. DeepMF casts safety-critical traffic simulation as a Bayesian factorization that includes the assignment of hazardous traffic participants, the motion prediction of selected opponents, the reaction estimation of autonomous vehicle (AV) and the probability estimation of the accident occur. All the aforementioned terms are calculated using decoupled deep neural networks, with inputs limited to the current observation and historical states. Consequently, DeepMF can effectively and efficiently simulate safety-critical traffic scenarios at any triggered time and for any duration by maximizing the compounded posterior probability of traffic risk. Extensive experiments demonstrate that DeepMF excels in terms of risk management, flexibility, and diversity, showcasing outstanding performance in simulating a wide range of realistic, high-risk traffic scenarios.

Figures

Figures reproduced from arXiv: 2412.17487 by the authors.

Figure 1
Figure 1. Overwiew of DeepMF. The opponent predictor selects the best attacker from surrounding agents. The trajectory [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The process of DeepMF generating a challenging scenario begins with assigning adversarial scores to all surrounding [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Output of the opponent prediction module on real [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: DeepMF can efficiently generate highly aggressive, natural, and diverse opponent motions across various traffic [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: The qualitative evaluation results where various algorithms generate adversarial trajectories based on the same raw [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Eight types of accidents in safety-critical scenarios [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

22 extracted references · 14 canonical work pages

  1. [1]

    Lane-change detection based on vehicle-trajectory prediction,

    H. Woo, Y . Ji, H. Kono, Y . Tamura, Y . Kuroda, T. Sugano, Y . Ya- mamoto, A. Yamashita, and H. Asama, “Lane-change detection based on vehicle-trajectory prediction,” IEEE Robotics and Automation Let- ters, vol. 2, no. 2, pp. 1109–1116, 2017

    work page 2017

  2. [2]

    Large scale interactive motion forecasting for autonomous driving: The waymo open motion dataset,

    S. Ettinger, S. Cheng, B. Caine, C. Liu, H. Zhao, S. Pradhan, Y . Chai, B. Sapp, C. Qi, Y . Zhou, Z. Yang, A. Chouard, P. Sun, J. Ngiam, V . Vasudevan, A. McCauley, J. Shlens, and D. Anguelov, “Large scale interactive motion forecasting for autonomous driving: The waymo open motion dataset,” in 2021 IEEE/CVF International Conference on Computer Vision (IC...

    arXiv 2021

  3. [3]

    nuscenes: A multimodal dataset for autonomous driving,

    H. Caesar, V . Bankiti, A. H. Lang, S. V ora, V . E. Liong, Q. Xu, A. Krishnan, Y . Pan, G. Baldan, and O. Beijbom, “nuscenes: A multimodal dataset for autonomous driving,” in 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , Jun 2020. [Online]. Available: http://dx.doi.org/10.1109/cvpr42600. 2020.01164

    work page arXiv 2020

  4. [4]

    Metadrive: Composing diverse driving scenarios for generalizable reinforcement learning,

    Q. Li, Z. Peng, L. Feng, Q. Zhang, Z. Xue, and B. Zhou, “Metadrive: Composing diverse driving scenarios for generalizable reinforcement learning,” IEEE Transactions on Pattern Analysis and Machine Intelligence , p. 1–14, Jan 2022. [Online]. Available: http://dx.doi.org/10.1109/tpami.2022.3190471

    arXiv 2022

  5. [5]

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

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

    work page 2022

  6. [6]

    Learning to simulate,

    N. Ruiz, S. Schulter, and M. Chandraker, “Learning to simulate,” International Conference on Learning Representations,International Conference on Learning Representations , Sep 2018

    work page 2018

  7. [7]

    Road vehicles—safety of the intended functionality,

    C. ISO21448, “Road vehicles—safety of the intended functionality,” 2022

    work page 2022

  8. [8]

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

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

    work page 2023

Show all 22 references
  1. [9]

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

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

    2023

  2. [10]

    Data-driven traffic simulation: A comprehensive review,

    D. Chen, M. Zhu, H. Yang, X. Wang, and Y . Wang, “Data-driven traffic simulation: A comprehensive review,” IEEE Transactions on Intelligent Vehicles, 2024

    2024

  3. [11]

    Augmented lidar simulator for autonomous driving,

    J. Fang, D. Zhou, F. Yan, T. Zhao, F. Zhang, Y . Ma, L. Wang, and R. Yang, “Augmented lidar simulator for autonomous driving,” IEEE Robotics and Automation Letters , vol. 5, no. 2, pp. 1931–1938, 2020

    1931

  4. [12]

    Extracting traffic primitives directly from nat- uralistically logged data for self-driving applications,

    W. Wang and D. Zhao, “Extracting traffic primitives directly from nat- uralistically logged data for self-driving applications,” IEEE Robotics and Automation Letters , vol. 3, no. 2, pp. 1223–1229, 2018

    2018

  5. [13]

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

    J. Wang, A. Pun, J. Tu, S. Manivasagam, A. Sadat, S. Casas, M. Ren, and R. Urtasun, “Advsim: Generating safety-critical scenarios for self- driving vehicles,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2021, pp. 9909–9918

    2021

  6. [14]

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

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

    2023

  7. [15]

    On adver- sarial robustness of trajectory prediction for autonomous vehicles,

    Q. Zhang, S. Hu, J. Sun, Q. A. Chen, and Z. M. Mao, “On adver- sarial robustness of trajectory prediction for autonomous vehicles,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 15 159–15 168

    2022

  8. [16]

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

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

    2022

  9. [17]

    Vectornet: Encoding hd maps and agent dynamics from vectorized representation,

    J. Gao, C. Sun, H. Zhao, Y . Shen, D. Anguelov, C. Li, and C. Schmid, “Vectornet: Encoding hd maps and agent dynamics from vectorized representation,” in Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition , 2020, pp. 11 525–11 533

    2020

  10. [18]

    Very deep convolutional networks for large-scale image recognition,

    K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” arXiv preprint arXiv:1409.1556 , 2014

    2014 arXiv

  11. [19]

    Densetnt: End-to-end trajectory pre- diction from dense goal sets,

    J. Gu, C. Sun, and H. Zhao, “Densetnt: End-to-end trajectory pre- diction from dense goal sets,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2021, pp. 15 303– 15 312

    2021

  12. [20]

    A tutorial on bayesian optimization,

    P. I. Frazier, “A tutorial on bayesian optimization,” arXiv preprint arXiv:1807.02811, 2018

    2018 arXiv

  13. [21]

    Genetic algorithm-a literature review,

    A. Lambora, K. Gupta, and K. Chopra, “Genetic algorithm-a literature review,” in 2019 international conference on machine learning, big data, cloud and parallel computing (COMITCon) . IEEE, 2019, pp. 380–384

    2019

  14. [22]

    A review of random search methods,

    S. Andrad ´ottir, “A review of random search methods,” Handbook of simulation optimization, pp. 277–292, 2014

    2014

Pith tools

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