Pith. sign in

REVIEW 4 major objections 5 minor 122 references

Long-term Traffic Scene Prediction via Polynomial Representations in Autonomous Driving

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read Polynomial trajectory representations keep accuracy and generalize across traffic datasets.

desk verdict A serious dissertation with a real contribution in the empirical Bayes fitting analysis, but its central out-of-distribution claim is weakened by a non-neutral cross-dataset protocol that the thesis acknowledges but never fully neutralizes. read the letter →

arxiv 2608.03330 v1 pith:YGB6LWHZ submitted 2026-08-04 cs.AI

Yue Yao This is my paper
classification cs.AI
keywords polynomialtrajectoryrepresentationtrafficscenepredictionout-of-distributiongeneralizationBernsteinpolynomialsdiffusionmodelsautonomousdrivingempiricalBayesmulti-agent
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

This dissertation argues that representing traffic trajectories and maps with Bernstein polynomials, rather than raw time-sequences of points, gives prediction models a compact, smooth, and physically consistent foundation. It tries to establish that moderate-degree polynomials fit real motion with negligible bias, and that models built on them match near-state-of-the-art accuracy in-distribution while transferring substantially better to new datasets. The payoff, if true, is that a single representation choice improves efficiency and out-of-distribution robustness without sacrificing accuracy. The work also claims that standard in-distribution regression metrics can mask these generalization gains.

What carries the argument

The Bernstein polynomial parameterization of curves: a trajectory or map element is written as a weighted sum $\sum_{n=0}^N \phi_n(\tau) w_n$ of Bernstein basis functions, with control points $w_n$ as parameters. The thesis uses this end-to-end: a modified Kalman filter tracks the control points of an agent's history; total-least-squares fitting converts map centerlines into 3-degree polynomials; the prediction head outputs kinematic states at 0, 3, and 6 seconds and reconstructs a 6-degree trajectory by least squares; and a diffusion process denoises polynomial control-point vectors for scene generation. The representation's inherent continuity, convex-hull property, and low parameter count

What would settle it

Run the same cross-dataset protocol while varying history length for all models: train and test sequence baselines on the homogenized 5-second history and 4.1-second horizon, and also on Waymo's original 1.1-second history. If the sequence baselines match or beat the polynomial model on the 5-second version, the claimed representation advantage is falsified. A second check is to apply the same protocol to a third city or dataset pair not used in the thesis and compare out-of-distribution minADE.

Watch

Extended reading notes

Core claim

The central claim is that polynomial representations are not just a compression trick; they are a better inductive bias for traffic prediction. Using empirical Bayes estimation of prior covariances and observation noise on three large datasets, the thesis finds that optimal polynomial degrees stay moderate (about 5 to 7 for 5- to 8-second trajectories) and that fit error remains far below the displacement error of state-of-the-art predictors. A prediction model that represents agent history, map elements, and future trajectories all as Bernstein polynomials reaches near-state-of-the-art accuracy on Argoverse 2 and, according to the reported experiments, outperforms sequence-based baselines u

Load-bearing premise

The out-of-distribution advantage rests on a homogenization protocol that lengthens Waymo Open history from 1.1 seconds to 5 seconds and fixes a 4.1-second horizon; if this shifted task is not equally favorable to every model, the observed gap could come from task difficulty rather than from the polynomial representation.

Editorial extensions

If this is right

  • Moderate-degree polynomials (about 5 to 7) suffice for realistic 5- to 8-second trajectories, so choosing this representation need not cap prediction accuracy.
  • Models trained with polynomial inputs and outputs transfer across independently collected datasets with a smaller performance drop than sequence-based models.
  • The same polynomial backbone supports both marginal prediction and diffusion-based joint scene generation, unifying two common formulations of the task.
  • Evaluation that only uses in-distribution minADE and minFDE can overstate real-world robustness; out-of-distribution and plausibility metrics are needed to see the difference.
  • Compact polynomial inputs reduce memory and inference cost compared with dense sequence representations in attention-based architectures.

Reading between the lines

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

  • A fully neutral counterfactual, retraining sequence baselines on the same 5-second history and 4.1-second horizon protocol, would isolate whether the reported out-of-distribution advantage comes from the representation or from the changed task; the thesis does not provide that complete control.
  • The polynomial inductive bias is likely to help most in low-data or sensor-noise regimes, because the empirical Bayes fitting acts as a learned smoother; this is an extension the thesis only hints at.
  • Spline variants with adaptive knot placement could trade some of the simplicity of a single polynomial for local flexibility on highly curved roads, a direction the thesis explicitly leaves open.
  • A practical deployment test would use natively polynomial HD maps (for example OpenDRIVE) and measure on-device latency in dense scenes, where the reduced token count should lower the quadratic cost of self-attention.
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 / 5 minor

Summary. This dissertation argues that polynomial representations—specifically Bernstein polynomials—provide a compact, physically consistent, and generalizable foundation for traffic scene prediction. Chapter 3 presents an empirical Bayes framework to estimate observation noise, prior covariances, and optimal polynomial degrees for agent trajectories, concluding that moderate-degree polynomials fit real-world trajectories with error far below current prediction errors. Chapter 4 introduces the 'Everything Polynomial' (EP) prediction model, together with a homogenization protocol that aligns Argoverse 2 and Waymo Open data formats, and reports competitive in-distribution accuracy with improved out-of-distribution robustness relative to two sequence-based baselines. Chapter 5 extends EP with a diffusion-based generative model for multi-agent scene prediction, reporting improved plausibility and cross-dataset generalization. The central thesis is that the representation choice, rather than dataset-specific architecture tuning, drives efficiency, robustness, and generalization.

Significance. If the central OoD claim holds, the work would be a meaningful step toward a single trajectory/map representation that transfers across independently collected autonomous-driving datasets while reducing computational cost. The thesis has clear strengths: both implementation repositories are publicly available, the empirical Bayes analysis is principled and conducted on three large-scale datasets, the homogenization protocol is a serious attempt to enable fair cross-dataset evaluation, and the limitations are explicitly discussed. However, the OoD generalization claim is currently not uniquely attributable to polynomial representations because the homogenization protocol changes the prediction task in ways that may favor the proposed models independently of the representation. The comparison of fit error with minADE6 is also informative but not a direct bound. These issues are load-bearing for the main thesis; if addressed with matched-task counterfactuals and a cleaner isolation of representation error, the contribution would be solid.

major comments (4)
  1. [§4.3, Table 4.2 and §4.6.5] The OoD generalization claim is confounded by the homogenization protocol. WO's original task uses 1.1 s history and an 8 s horizon; the homogenized task uses 5 s history and a 4.1 s horizon. A longer history gives the polynomial Kalman smoother in §4.4.1 more observations to estimate control points, while a shorter horizon favors low-degree smooth outputs; both can benefit EP independently of any representational advantage. The complexity analysis in §4.6.3 measures target-agent deviation from constant velocity but does not provide a matched-history/horizon counterfactual. Without an experiment in which all models are trained and evaluated under identical native—or at least symmetric—task specifications, the observed OoD advantage cannot be uniquely attributed to polynomial representations.
  2. [§3.6.6, Fig. 3.11] The comparison of polynomial AFE with minADE6 of SotA predictors is suggestive but not a direct apples-to-apples bound. AFE is a per-trajectory approximation error computed with access to the full future and includes observation noise, while minADE6 is a best-of-six prediction error against noisy labels. The conclusion that 'the bias introduced by polynomial representation does not fundamentally limit prediction' would be strengthened by isolating representation error from observation noise, for example by validating on synthetic trajectories with known ground truth or by comparing the AFE of polynomial fitting with the fitting error of the sequence-based representations actually used by the benchmark models.
  3. [§3.4.2 and §3.5.2] The empirical Bayes estimates rely on a Gaussian prior and observation noise model (Eqs. 3.10–3.11), a quadratic radial variance assumption (Eq. 3.23), and manually selected RTS outlier thresholds (2 m and agent-type acceleration bounds). These choices are not validated against ground-truth sensor noise or an independent calibration set. Since the estimated priors are reused in Chapter 4 (Sections 4.4.1 and 4.4.2) to initialize the Kalman filter and observation noise, misspecification could propagate. The limitation is acknowledged in §3.7, but a sensitivity analysis over the thresholds and noise-model alternatives would materially increase confidence in the degree selection and downstream results.
  4. [§4.6.1, Table 4.4] The training-horizon asymmetry between the two OoD configurations makes the 'delta' metrics in Table 4.7 difficult to interpret. In A2*→WO*, models are trained with a 6 s horizon but evaluated at 4.1 s, whereas in WO*→A2* models are trained with exactly the 4.1 s evaluation horizon. This asymmetry can affect ID and OoD performance independently of representation. Reporting results in which every model is trained at the horizon at which it is evaluated, or explicitly ablating the training horizon, would remove this additional confound.
minor comments (5)
  1. [Eq. (2.20)] 'simplfied' should be 'simplified'; also, the symbol θ is used both for the diffusion-model parameters and later for observation-noise parameters, which is confusing across chapters.
  2. [§2.1.3] In the description of minSFDE, 'focus on the the final timestep' contains a duplicated article; the surrounding notation could also make explicit that T is the final timestep index.
  3. [Table 4.1] The entry 'EP-noAugQCNet-noAug' is missing a line break or space; the table would be clearer with separate rows or a clear delimiter.
  4. [§4.4.1] The process-noise covariance Qt is described as 'empirically defined' with descending values (0.3, 0.2, 0.1) and a just-in-time explanation for the ordering. A short ablation or a reference to a principled tuning procedure would improve reproducibility.
  5. [§3.4.4, Eq. (3.31)] The BIC formula uses T for trajectory timesteps, but in Chapter 3 T is also used for total discrete timesteps; with the continuous-time counterpart T′ introduced earlier, the notation should be kept consistent to avoid ambiguity.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the thesis's predictions are measured on held-out data and the polynomial representation is benchmarked against external models.

full rationale

The central derivation chain is empirical rather than definitional. The polynomial degree and observation-noise priors in Chapter 3 are estimated on training splits via empirical Bayes / AIC, and the same fixed values are then used as hyperparameters in Chapter 4's model; the reported minADE/minFDE and OoD metrics are evaluated on validation/test splits, so no fitted parameter is renamed as a prediction. The self-citations (ITSC 2023, IROS 2024, RA-L 2025) point to the author's own published versions of the same work, but the dissertation re-derives the analysis and the implementations are publicly available; no load-bearing uniqueness theorem or ansatz is imported solely through self-citation. The acknowledged homogenization shift (Section 4.3, Table 4.2; discussed in Section 4.6.3) changes WO's history length and prediction horizon and may alter task difficulty, but this is a potential confound for the OoD generalization claim rather than a circular reduction: the model outputs are not constructed from the evaluation target. Likewise, the in-sample fit error comparison with SotA displacement errors is a statistical comparison issue, not a case where the result equals an input by construction. The dissertation's stated limitations (Sections 3.7, 4.6.3, 5.4.8, 6.2) are candid and do not reveal a circular step. Overall, the paper is self-contained against external benchmarks and merits a circularity score of 0.

Assumptions & free parameters 11 free parameters · 5 assumptions · 0 invented entities

The central claims rest on free parameters fit to the training data (polynomial degrees, observation noise, outlier thresholds) and on domain assumptions about Gaussian noise and the fairness of the homogenized OoD protocol. No new physical entities are introduced.

free parameters (11)
  • polynomial degree (history) = 5
    Chosen via AIC analysis in Chapter 3 across datasets and agent types; used for all history trajectories including ego despite AIC suggesting 6 for ego.
  • polynomial degree (map) = 3
    Matched to OpenDRIVE standard and used for lane centerlines and crosswalks after TLS fitting.
  • polynomial degree (prediction) = 6
    One degree higher than the AIC recommendation for 5s history, chosen to capture more complex motion patterns in 6s trajectories.
  • observation noise parameters θne = σψ ~ 1e-3 to 3e-4, σc ~ 0.017-0.161, b0/b1/b2 not fully listed
    Estimated by empirical Bayes per dataset (Table 3.5); used as fixed constants in the Kalman filter in Chapter 4.
  • observation noise parameters θego = σdiag ~ 0.008-0.024, σcov ~ 2e-4 to -1e-7
    Estimated by empirical Bayes; used for ego trajectory noise.
  • RTS smoother observation noise Rt = diag(0.1,0.1) etc. in Table 3.1
    Hand-selected values for outlier detection preprocessing; affect which trajectories remain in the analysis.
  • RTS smoother outlier thresholds = 2m deviation, acceleration bounds [-10,6], [-4,2], [-3,2] m/s2
    Empirically tuned thresholds in Section 3.5.2 that filter out physically implausible trajectories.
  • Kalman filter process noise Qt = diag(0,0,0,0.3,0.2,0.1)⊗I2
    Ad hoc choice for the polynomial control point tracking model in Section 4.4.1.
  • Kalman filter prior covariance ΣP = diag(2e5 * ones)
    Uninformative prior with high variance used in the transition model.
  • TLS map fitting tolerance = 0.1m average fit error
    Threshold for recursively splitting lane segments during polynomial map fitting (Section 4.4.2).
  • max agents / max map elements = 50 / 150
    Fixed input sizes for batching; limits may exclude relevant context in dense scenes.
assumptions (5)
  • domain assumption Gaussian prior and observation noise distributions
    Used throughout Chapter 3 empirical Bayes; the thesis itself notes in Section 3.7 that this may not capture the full complexity of real-world distributions.
  • ad hoc to paper Observation noise for non-ego agents is polar with quadratic radial variance (Eq. 3.23)
    This specific functional form is justified by LiDAR intuition but is not validated against an independent sensor model.
  • domain assumption AIC/BIC select the optimal polynomial degree for prediction
    Information criteria measure fitting quality, not necessarily downstream prediction performance. The thesis uses these to pick degrees and then builds prediction models on top.
  • ad hoc to paper Homogenized datasets (5s history, 4.1s horizon) provide a fair OoD comparison
    The protocol changes WO's task definition; the paper's own complexity analysis (Section 4.6.3) does not fully rule out that the observed OoD gains are due to the shifted task rather than the representation.
  • domain assumption Representation error is upper bounded by fit error and is negligible when fit error << displacement error
    The decomposition in Figure 3.1 assumes the three error components add and that the empirical Bayes observation noise estimate is accurate.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Long-term Traffic Scene Prediction via Polynomial Representations in Autonomous Driving." pith.science (2026). https://pith.science/paper/YGB6LWHZ

@misc{pith2026260803330,
  author       = {Pith},
  title        = {Pith review of: Long-term Traffic Scene Prediction via Polynomial Representations in Autonomous Driving},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YGB6LWHZ}},
  note         = {Machine review of arXiv:2608.03330}
}
read the original abstract

This thesis addresses fundamental challenges in traffic scene prediction for autonomous driving by introducing robust and computationally efficient models based on polynomial representations. While conventional sequence-based representations often struggle with noise and generalization, this work demonstrates that polynomial representations offer significant advantages in computational efficiency, generalization, and prediction plausibility. Through theoretical analysis and empirical validation, this thesis demonstrates that moderate-degree polynomials capture real-world motion dynamics with high fidelity without constraining predictive performance. Building on this foundation, a prediction model representing both trajectories and map geometry with polynomial representations achieves near state-of-the-art accuracy on standard benchmarks while substantially improving generalization under distribution shift. Extending this concept, a diffusion- based generative framework enables multi-agent scene generation, producing traffic continuations that are more plausible and kinematically consistent than those generated by conventional baselines. Evaluations on the Argoverse 2 and Waymo Open datasets confirm that polynomial representations reduce computational cost, enhance cross-dataset generalization, and yield smoother trajectories and higher behavioral plausibility. The findings reveal that standard in-distribution evaluation and regression-based metrics may fail to reflect true model generalization and prediction plausibility. By providing theoretical justification and empirical validation, this dissertation estab- lishes polynomial trajectory representations as an efficient, expressive, and generalizable foundation for traffic scene prediction in safety critical autonomous driving.

Figures

Figures reproduced from arXiv: 2608.03330 by the authors.

Figure 1
Figure 1. A typical modular architecture of an autonomous driving function stack [PITH_FULL_IMAGE:figures/full_fig_p025_1.png] view at source ↗
Figure 2
Figure 2. Marginal and joint prediction [PITH_FULL_IMAGE:figures/full_fig_p036_2.png] view at source ↗
Figure 2
Figure 2. Minimum average displacement error (minADE) and minimum final dis [PITH_FULL_IMAGE:figures/full_fig_p037_2.png] view at source ↗
Figures from the paper (39 more)
Figure 2
Figure 2. Figure 2: Data representations for agent history trajectory and map [PITH_FULL_IMAGE:figures/full_fig_p041_2.png]
Figure 2
Figure 2. Figure 2: Feature extraction in convolutional layers [PITH_FULL_IMAGE:figures/full_fig_p046_2.png]
Figure 2
Figure 2. Figure 2: Scaled dot-product attention and multi-head attention [ [PITH_FULL_IMAGE:figures/full_fig_p048_2.png]
Figure 2
Figure 2. Figure 2: Architecture of attention block [ [PITH_FULL_IMAGE:figures/full_fig_p050_2.png]
Figure 2
Figure 2. Figure 2: Forward diffusion and reverse denoising [PITH_FULL_IMAGE:figures/full_fig_p053_2.png]
Figure 3
Figure 3. Figure 3: Components of displacement error [PITH_FULL_IMAGE:figures/full_fig_p064_3.png]
Figure 3
Figure 3. Figure 3: An example of a 5-degree polynomial with 50 uniformly sampled points. [PITH_FULL_IMAGE:figures/full_fig_p068_3.png]
Figure 3
Figure 3. Figure 3: Visualization of 100 translated trajectories from the Argoverse 1 dataset [PITH_FULL_IMAGE:figures/full_fig_p070_3.png]
Figure 3
Figure 3. Figure 3: Measurement in the polar coordinate [PITH_FULL_IMAGE:figures/full_fig_p075_3.png]
Figure 3
Figure 3. Figure 3 [PITH_FULL_IMAGE:figures/full_fig_p077_3.png]
Figure 3
Figure 3. Figure 3: Tracking outlier in A1 [ [PITH_FULL_IMAGE:figures/full_fig_p080_3.png]
Figure 3
Figure 3. Figure 3 [PITH_FULL_IMAGE:figures/full_fig_p088_3.png]
Figure 3
Figure 3. Figure 3: Optimal polynomial degree based on AIC and BIC [PITH_FULL_IMAGE:figures/full_fig_p091_3.png]
Figure 3
Figure 3. Figure 3: The fit error of models for cyclist (left) and pedestrian (right) trajectories [PITH_FULL_IMAGE:figures/full_fig_p092_3.png]
Figure 3
Figure 3. Figure 3: Comparison of the average fit error (AFE) of polynomial representations at [PITH_FULL_IMAGE:figures/full_fig_p093_3.png]
Figure 4
Figure 4. Figure 4: Kernel density plot of the maximum absolute curvature and length for [PITH_FULL_IMAGE:figures/full_fig_p099_4.png]
Figure 4
Figure 4. Figure 4: The two augmentation strategies for non-target agent data employed in [PITH_FULL_IMAGE:figures/full_fig_p103_4.png]
Figure 4
Figure 4. Figure 4: Example of predicted control points based on the transition model in [PITH_FULL_IMAGE:figures/full_fig_p107_4.png]
Figure 4
Figure 4. Figure 4: Accumulated arc length introduces errors for curved map elements. [PITH_FULL_IMAGE:figures/full_fig_p111_4.png]
Figure 4
Figure 4. Figure 4: One traffic scene in A2 represented with polynomials. Only boundaries of [PITH_FULL_IMAGE:figures/full_fig_p114_4.png]
Figure 4
Figure 4. Figure 4: Our proposed model pipeline [PITH_FULL_IMAGE:figures/full_fig_p115_4.png]
Figure 4
Figure 4. Figure 4: Histogram of nearby agents and map elements within a 100-meter range of [PITH_FULL_IMAGE:figures/full_fig_p117_4.png]
Figure 4
Figure 4. Figure 4: Encoder architecture shared by EP variants. The inputs are defined [PITH_FULL_IMAGE:figures/full_fig_p119_4.png]
Figure 4
Figure 4. Figure 4: Decoder architectures of EP-F, EP-Q and EP-noAug. [PITH_FULL_IMAGE:figures/full_fig_p121_4.png]
Figure 4
Figure 4. Figure 4: ID performance of benchmarks on the A2 test set. [PITH_FULL_IMAGE:figures/full_fig_p127_4.png]
Figure 4
Figure 4. Figure 4: Kernel density plot of normalized longitudinal and lateral distances with [PITH_FULL_IMAGE:figures/full_fig_p128_4.png]
Figure 4
Figure 4. Figure 4: The OoD testing results of FMAE, QCNet, EP and their variants. The [PITH_FULL_IMAGE:figures/full_fig_p129_4.png]
Figure 5
Figure 5. Figure 5: Two limitations of regression-based metrics. [PITH_FULL_IMAGE:figures/full_fig_p143_5.png]
Figure 5
Figure 5. Figure 5: Overview of EP-Diffuser for traffic scene generation. For clear presentation, [PITH_FULL_IMAGE:figures/full_fig_p149_5.png]
Figure 5
Figure 5. Figure 5: An example of a traffic scene graph [PITH_FULL_IMAGE:figures/full_fig_p151_5.png]
Figure 5
Figure 5. Figure 5: The edge features between source and target nodes at the reference timestep, [PITH_FULL_IMAGE:figures/full_fig_p154_5.png]
Figure 5
Figure 5. Figure 5: Illustration of the EP-Diffuser denoising process on Argoverse 2. [PITH_FULL_IMAGE:figures/full_fig_p156_5.png]
Figure 5
Figure 5. Figure 5: Encoder architecture of EP-Diffuser [PITH_FULL_IMAGE:figures/full_fig_p157_5.png]
Figure 5
Figure 5. Figure 5: Denoiser architecture of EP-Diffuser [PITH_FULL_IMAGE:figures/full_fig_p159_5.png]
Figure 5
Figure 5. Figure 5: Impact of post-processing on realism score, accuracy (minSADE), and [PITH_FULL_IMAGE:figures/full_fig_p163_5.png]
Figure 5
Figure 5. Figure 5: Observed sample and generated traffic scene samples from EP-Diffuser for [PITH_FULL_IMAGE:figures/full_fig_p165_5.png]
Figure 5
Figure 5. Figure 5: Predicted heading, longitudinal velocity, acceleration, and jerk for a vehicle’s [PITH_FULL_IMAGE:figures/full_fig_p166_5.png]
Figure 5
Figure 5. Figure 5: Realism meta score on A2 R20P and inference time for diffusion models [PITH_FULL_IMAGE:figures/full_fig_p167_5.png]
Figure 5
Figure 5. Figure 5: Examples of failure cases from EP-Diffuser: (left) agent collision where [PITH_FULL_IMAGE:figures/full_fig_p171_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

122 extracted references · 64 canonical work pages

  1. [1]

    Jerk within the context of science and engineering—a systematic review,

    H. Hayati, D. Eager, A.-M. Pendrill, and H. Alberg, “Jerk within the context of science and engineering—a systematic review,”Vibration, vol. 3, no. 4, pp. 371– 409, 2020

    2020

  2. [2]

    Self-Driving like a Human driver instead of a Robocar: Personalized comfortable driving experience for autonomous vehicles

    I. Bae, J. Moon, J. Jhung, H. Suk, T. Kim, H. Park, J. Cha, J. Kim, D. Kim, and S. Kim, “Self-driving like a human driver instead of a robocar: Personal- ized comfortable driving experience for autonomous vehicles,”arXiv preprint arXiv:2001.03908, 2020

    work page Pith review arXiv 2001

  3. [3]

    Understanding and modeling the human driver,

    C. C. Macadam, “Understanding and modeling the human driver,”Vehicle system dynamics, vol. 40, no. 1-3, pp. 101–134, 2003

    2003

  4. [4]

    Bicycle braking performance testing and analysis,

    N. Famiglietti, B. Nguyen, E. Fatzinger, and J. Landerville, “Bicycle braking performance testing and analysis,”SAE International Journal of Advances and Current Practices in Mobility, vol. 2, pp. 3384–3397, 2020

    2020

  5. [5]

    Planning-oriented autonomous driving,

    Y. Hu, J. Yang, L. Chen, K. Li, C. Sima, X. Zhu, S. Chai, S. Du, T. Lin, W. Wang, et al., “Planning-oriented autonomous driving,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 17853–17862, 2023

    2023

  6. [6]

    Vad: Vectorized scene representation for efficient autonomous driving,

    B. Jiang, S. Chen, Q. Xu, B. Liao, J. Chen, H. Zhou, Q. Zhang, W. Liu, C. Huang, and X. Wang, “Vad: Vectorized scene representation for efficient autonomous driving,” inIEEE/CVF International Conference on Computer Vision (ICCV), pp. 8340–8350, 2023

    2023

  7. [7]

    Sparsedrive: End-to- end autonomous driving via sparse scene representation,

    W. Sun, X. Lin, Y. Shi, C. Zhang, H. Wu, and S. Zheng, “Sparsedrive: End-to- end autonomous driving via sparse scene representation,” inIEEE International Conference on Robotics and Automation (ICRA), pp. 8795–8801, 2025

    2025

  8. [8]

    Unfreezing the robot: Navigation in dense, inter- acting crowds,

    P. Trautman and A. Krause, “Unfreezing the robot: Navigation in dense, inter- acting crowds,” inIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 797–803, 2010

    2010

Show all 122 references
  1. [9]

    The integration of prediction and planning in deep learning automated driving systems: A review,

    S. Hagedorn, M. Hallgarten, M. Stoll, and A. P. Condurache, “The integration of prediction and planning in deep learning automated driving systems: A review,” IEEE Transactions on Intelligent Vehicles, vol. 10, no. 5, pp. 3626–3643, 2024. 158Bibliography

    2024

  2. [10]

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

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

    1999

  3. [11]

    Enhanced intelligent driver model to access the impact of driving strategies on traffic capacity,

    A. Kesting, M. Treiber, and D. Helbing, “Enhanced intelligent driver model to access the impact of driving strategies on traffic capacity,”Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, vol. 368, no. 1928, pp. 4585–4605, 2010

    1928

  4. [12]

    Philipp,Perception and Prediction of Urban Traffic Scenarios for Autonomous Driving

    A. Philipp,Perception and Prediction of Urban Traffic Scenarios for Autonomous Driving. PhD thesis, Freie Universitaet Berlin, 2021

    2021

  5. [13]

    Social LSTM: Human Trajectory Prediction in Crowded Spaces,

    A. Alahi, K. Goel, V. Ramanathan, A. Robicquet, L. Fei-Fei, and S. Savarese, “Social LSTM: Human Trajectory Prediction in Crowded Spaces,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 961–971, 2016

    2016

  6. [14]

    Learning lane graph representations for motion forecasting,

    M. Liang, B. Yang, R. Hu, Y. Chen, R. Liao, S. Feng, and R. Urtasun, “Learning lane graph representations for motion forecasting,” inEuropean Conference on Computer Vision (ECCV), pp. 541–556, 2020

    2020

  7. [15]

    Hivt: Hierarchicalvectortransformer for multi-agent motion prediction,

    Z.Zhou, L.Ye, J.Wang, K.Wu, andK.Lu, “Hivt: Hierarchicalvectortransformer for multi-agent motion prediction,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 8823–8833, 2022

    2022

  8. [16]

    nuscenes: A multimodal dataset for autonomous driving,

    H. Caesar, V. Bankiti, A. H. Lang, S. Vora, V. E. Liong, Q. Xu, A. Krishnan, Y. Pan, G. Baldan, and O. Beijbom, “nuscenes: A multimodal dataset for autonomous driving,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 11621–11631, 2020

    2020

  9. [17]

    Argoverse 2: Next generation datasets for self-driving perception and forecasting,

    B. Wilson, W. Qi, T. Agarwal, J. Lambert, J. Singh, S. Khandelwal, B. Pan, R. Kumar, A. Hartnett, J. K. Pontes,et al., “Argoverse 2: Next generation datasets for self-driving perception and forecasting,” inConference on Neural Information Processing Systems (NeurIPS) Datasets ...

    2021

  10. [18]

    Large scale interactive motion forecasting for au- tonomous driving: The waymo open motion dataset,

    S. Ettinger, S. Cheng, B. Caine, C. Liu, H. Zhao, S. Pradhan, Y. Chai, B. Sapp, C. R. Qi, Y. Zhou,et al., “Large scale interactive motion forecasting for au- tonomous driving: The waymo open motion dataset,” inIEEE/CVF International Conference on Computer Vision (ICCV), pp. 97...

    2021

  11. [19]

    Interaction dataset: An international, adversarial and cooperative motion dataset in interactive driving scenarios with semantic maps,

    W. Zhan, L. Sun, D. Wang, H. Shi, A. Clausse, M. Naumann, J. Kummerle, H. Konigshof, C. Stiller, A. de La Fortelle,et al., “Interaction dataset: An international, adversarial and cooperative motion dataset in interactive driving scenarios with semantic maps,”arXiv preprint arX...

    1910 arXiv

  12. [20]

    The highd dataset: A drone dataset of naturalistic vehicle trajectories on german highways for validation of highly automated driving systems,

    R. Krajewski, J. Bock, L. Kloeker, and L. Eckstein, “The highd dataset: A drone dataset of naturalistic vehicle trajectories on german highways for validation of highly automated driving systems,” inIEEE International Conference on Intelligent Transportation Systems (ITSC), pp...

    2018

  13. [21]

    Vehicle trajectory prediction works, but not everywhere,

    M. Bahari, S. Saadatnejad, A. Rahimi, M. Shaverdikondori, A. H. Shahidzadeh, S.-M. Moosavi-Dezfooli, and A. Alahi, “Vehicle trajectory prediction works, but not everywhere,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 17123–17133, 2022

    2022

  14. [22]

    Unitraj: A unified framework for scalable vehicle trajectory prediction,

    L. Feng, M. Bahari, K. M. B. Amor, É. Zablocki, M. Cord, and A. Alahi, “Unitraj: A unified framework for scalable vehicle trajectory prediction,” in European Conference on Computer Vision (ECCV), pp. 106–123, Springer, 2024

    2024

  15. [23]

    An empirical bayes analysis of object tra- jectory representation models,

    Y. Yao, D. Goehring, and J. Reichardt, “An empirical bayes analysis of object tra- jectory representation models,” inIEEE International Conference on Intelligent Transportation Systems (ITSC), pp. 902–909, 2023

    2023

  16. [24]

    Trajectron++: Dynamically-Feasible Trajectory Forecasting with Heterogeneous Data,

    T. Salzmann, B. Ivanovic, P. Chakravarty, and M. Pavone, “Trajectron++: Dynamically-Feasible Trajectory Forecasting with Heterogeneous Data,” inEu- ropean Conference on Computer Vision (ECCV), pp. 683–700, 2020

    2020

  17. [25]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” inConference on Neural Information Processing Systems (NeurIPS), vol. 30, 2017

    2017

  18. [26]

    Improving out- of-distribution generalization of trajectory prediction for autonomous driving via polynomial representations,

    Y. Yao, S. Yan, D. Goehring, W. Burgard, and J. Reichardt, “Improving out- of-distribution generalization of trajectory prediction for autonomous driving via polynomial representations,” inIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 488–495,...

    2024

  19. [27]

    Ep-diffuser: An efficient diffusion model for traffic scene generation and prediction via polynomial repre- sentations,

    Y. Yao, M.-K. Bouzidi, D. Goehring, and J. Reichardt, “Ep-diffuser: An efficient diffusion model for traffic scene generation and prediction via polynomial repre- sentations,”IEEE Robotics and Automation Letters, vol. 10, no. 9, pp. 9478–9485, 2025

    2025

  20. [28]

    Argoverse: 3d tracking and forecasting with rich maps,

    M.-F. Chang, J. Lambert, P. Sangkloy, J. Singh, S. Bak, A. Hartnett, D. Wang, P. Carr, S. Lucey, D. Ramanan, and J. Hays, “Argoverse: 3d tracking and forecasting with rich maps,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019

    2019

  21. [29]

    Shifts: A datasetofrealdistributionalshiftacrossmultiplelarge-scaletasks,

    A. Malinin, N. Band, G. Chesnokov, Y. Gal, M. J. Gales, A. Noskov, A. Ploskonosov, L. Prokhorenkova, I. Provilkov, V. Raina,et al., “Shifts: A datasetofrealdistributionalshiftacrossmultiplelarge-scaletasks,”arXiv preprint arXiv:2107.07455, 2021

    2021 arXiv

  22. [30]

    nuplan: A closed-loop ml-based planning benchmark for autonomous vehicles,

    H. Caesar, J. Kabzan, K. S. Tan, W. K. Fong, E. Wolff, A. Lang, L. Fletcher, O. Beijbom, and S. Omari, “nuplan: A closed-loop ml-based planning benchmark for autonomous vehicles,”arXiv preprint arXiv:2106.11810, 2021

    2021 arXiv

  23. [31]

    Continuous driver intention recogni- tionwithhiddenmarkovmodels,

    H. Berndt, J. Emmert, and K. Dietmayer, “Continuous driver intention recogni- tionwithhiddenmarkovmodels,” inIEEE International Conference on Intelligent Transportation Systems (ITSC), pp. 1189–1194, 2008. 160Bibliography

    2008

  24. [32]

    A probabilistic model for estimating driver behaviors and vehicle trajectories in traffic environments,

    T. Gindele, S. Brechtel, and R. Dillmann, “A probabilistic model for estimating driver behaviors and vehicle trajectories in traffic environments,” inIEEE Interna- tional Conference on Intelligent Transportation Systems (ITSC), pp. 1625–1631, 2010

    2010

  25. [33]

    A survey on motion prediction and risk assessment for intelligent vehicles,

    S. Lefèvre, D. Vasquez, and C. Laugier, “A survey on motion prediction and risk assessment for intelligent vehicles,”ROBOMECH journal, vol. 1, no. 1, p. 1, 2014

    2014

  26. [34]

    A survey on trajectory-prediction methods for autonomous driving,

    Y. Huang, J. Du, Z. Yang, Z. Zhou, L. Zhang, and H. Chen, “A survey on trajectory-prediction methods for autonomous driving,”IEEE transactions on intelligent vehicles, vol. 7, no. 3, pp. 652–674, 2022

    2022

  27. [35]

    Scenario understanding and motion prediction for autonomous vehicles—review and comparison,

    P. Karle, M. Geisslinger, J. Betz, and M. Lienkamp, “Scenario understanding and motion prediction for autonomous vehicles—review and comparison,”IEEE Transactions on Intelligent Transportation Systems, vol. 23, no. 10, pp. 16962– 16982, 2022

    2022

  28. [36]

    Covernet: Multimodal behavior prediction using trajectory sets,

    T. Phan-Minh, E. C. Grigore, F. A. Boulton, O. Beijbom, and E. M. Wolff, “Covernet: Multimodal behavior prediction using trajectory sets,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 14074– 14083, 2020

    2020

  29. [37]

    Trajectorycnn: a new spatio-temporal feature learning network for human motion prediction,

    X. Liu, J. Yin, J. Liu, P. Ding, J. Liu, and H. Liu, “Trajectorycnn: a new spatio-temporal feature learning network for human motion prediction,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 31, no. 6, pp. 2133–2146, 2020

    2020

  30. [38]

    Imagenet classification with deep convolutional neural networks,

    A. Krizhevsky, I. Sutskever, and G. E. Hinton, “Imagenet classification with deep convolutional neural networks,” inConference on Neural Information Processing Systems (NeurIPS), vol. 25, pp. 1097–1105, 2012

    2012

  31. [39]

    Harmonic networks: Deep translation and rotation equivariance,

    D. E. Worrall, S. J. Garbin, D. Turmukhambetov, and G. J. Brostow, “Harmonic networks: Deep translation and rotation equivariance,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 5028–5037, 2017

    2017

  32. [40]

    Rotation equivariant vector field networks,

    D. Marcos, M. Volpi, N. Komodakis, and D. Tuia, “Rotation equivariant vector field networks,” inIEEE/CVF International Conference on Computer Vision (ICCV), pp. 5048–5057, 2017

    2017

  33. [41]

    Query-centric trajectory predic- tion,

    Z. Zhou, J. Wang, Y.-H. Li, and Y.-K. Huang, “Query-centric trajectory predic- tion,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 17863–17873, 2023

    2023

  34. [42]

    Forecast-mae: Self-supervised pre-training for motion forecasting with masked autoencoders,

    J. Cheng, X. Mei, and M. Liu, “Forecast-mae: Self-supervised pre-training for motion forecasting with masked autoencoders,” inIEEE/CVF International Conference on Computer Vision (ICCV), pp. 8679–8689, 2023

    2023

  35. [43]

    Motionlm: Multi-agent motion forecasting as language modeling,

    A. Seff, B. Cera, D. Chen, M. Ng, A. Zhou, N. Nayakanti, K. S. Refaat, R. Al- Rfou, and B. Sapp, “Motionlm: Multi-agent motion forecasting as language modeling,” inIEEE/CVF International Conference on Computer Vision (ICCV), pp. 8579–8590, 2023. Bibliography161

    2023

  36. [44]

    Mo- tiondiffuser: Controllable multi-agent motion prediction using diffusion,

    C. Jiang, A. Cornman, C. Park, B. Sapp, Y. Zhou, D. Anguelov,et al., “Mo- tiondiffuser: Controllable multi-agent motion prediction using diffusion,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 9644–9653, 2023

    2023

  37. [45]

    Optimizing diffusion models for joint trajectory prediction and controllable generation,

    Y. Wang, C. Tang, L. Sun, S. Rossi, Y. Xie, C. Peng, T. Hannagan, S. Sabatini, N. Poerio, M. Tomizuka,et al., “Optimizing diffusion models for joint trajectory prediction and controllable generation,” inEuropean Conference on Computer Vision (ECCV), pp. 324–341, 2025

    2025

  38. [46]

    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 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 11525–11533, 2020

    2020

  39. [47]

    Uncertainty-aware driver trajectory prediction at urban intersections,

    X. Huang, S. G. McGill, B. C. Williams, L. Fletcher, and G. Rosman, “Uncertainty-aware driver trajectory prediction at urban intersections,” inIEEE International Conference on Robotics and Automation (ICRA), pp. 9718–9724, 2019

    2019

  40. [48]

    Plop: Probabilistic poly- nomial objects trajectory planning for autonomous driving,

    T. Buhet, E. Wirbel, A. Bursuc, and X. Perrotton, “Plop: Probabilistic poly- nomial objects trajectory planning for autonomous driving,”arXiv preprint arXiv:2003.08744, 2020

    2003 arXiv

  41. [49]

    Temporally-continuous probabilistic prediction using polynomial trajectory pa- rameterization,

    Z. Su, C. Wang, H. Cui, N. Djuric, C. Vallespi-Gonzalez, and D. Bradley, “Temporally-continuous probabilistic prediction using polynomial trajectory pa- rameterization,” inIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 3837–3843, IEEE, 2021

    2021

  42. [50]

    You mostly walk alone: Analyzing feature attribution in trajectory prediction,

    O. Makansi, J. Von Kügelgen, F. Locatello, P. Gehler, D. Janzing, T. Brox, and B. Schölkopf, “You mostly walk alone: Analyzing feature attribution in trajectory prediction,”arXiv preprint arXiv:2110.05304, 2021

    2021 arXiv

  43. [51]

    Multipath: Multiple probabilistic anchor trajectory hypotheses for behavior prediction,

    Y. Chai, B. Sapp, M. Bansal, and D. Anguelov, “Multipath: Multiple probabilistic anchor trajectory hypotheses for behavior prediction,” inConference on Robot Learning (CoRL), 2019

    2019

  44. [52]

    Neural machine translation by jointly learning to align and translate,

    D. Bahdanau, K. Cho, and Y. Bengio, “Neural machine translation by jointly learning to align and translate,” inInternational Conference on Learning Repre- sentations (ICLR), 2015

    2015

  45. [53]

    Long short-term memory,

    S. Hochreiter and J. Schmidhuber, “Long short-term memory,”Neural computa- tion, vol. 9, no. 8, pp. 1735–1780, 1997

    1997

  46. [54]

    Mtr++: Multi-agent motion prediction with symmetric scene modeling and guided intention querying,

    S. Shi, L. Jiang, D. Dai, and B. Schiele, “Mtr++: Multi-agent motion prediction with symmetric scene modeling and guided intention querying,”IEEE Transac- tions on Pattern Analysis and Machine Intelligence, vol. 46, no. 5, pp. 3955–3971, 2024

    2024

  47. [55]

    Way- former: Motion forecasting via simple & efficient attention networks,

    N. Nayakanti, R. Al-Rfou, A. Zhou, K. Goel, K. S. Refaat, and B. Sapp, “Way- former: Motion forecasting via simple & efficient attention networks,”arXiv preprint arXiv:2207.05844, 2022. 162Bibliography

    2022 arXiv

  48. [56]

    Deep residual learning for image recogni- tion,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recogni- tion,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 770–778, 2016

    2016

  49. [57]

    Layer normalization,

    J. L. Ba, J. R. Kiros, and G. E. Hinton, “Layer normalization,”arXiv preprint arXiv:1607.06450, 2016

    2016 arXiv

  50. [58]

    Training data-efficient image transformers & distillation through attention,

    H. Touvron, M. Cord, M. Douze, F. Massa, A. Sablayrolles, and H. Jégou, “Training data-efficient image transformers & distillation through attention,” in International Conference on Machine Learning (ICML), pp. 10347–10357, 2021

    2021

  51. [59]

    Trans- formers in vision: A survey,

    S. Khan, M. Naseer, M. Hayat, S. W. Zamir, F. S. Khan, and M. Shah, “Trans- formers in vision: A survey,”ACM computing surveys (CSUR), vol. 54, no. 10s, pp. 1–41, 2022

    2022

  52. [60]

    Social gan: Socially acceptable trajectories with generative adversarial networks,

    A. Gupta, J. Johnson, L. Fei-Fei, S. Savarese, and A. Alahi, “Social gan: Socially acceptable trajectories with generative adversarial networks,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 2255– 2264, 2018

    2018

  53. [61]

    Desire: Distant future prediction in dynamic scenes with interacting agents,

    N. Lee, W. Choi, P. Vernaza, C. B. Choy, P. H. Torr, and M. Chandraker, “Desire: Distant future prediction in dynamic scenes with interacting agents,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 336–345, 2017

    2017

  54. [62]

    Sophie: An attentive gan for predicting paths compliant to social and physical constraints,

    A. Sadeghian, V. Kosaraju, A. Sadeghian, N. Hirose, H. Rezatofighi, and S. Savarese, “Sophie: An attentive gan for predicting paths compliant to social and physical constraints,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 1349–1358, 2019

    2019

  55. [63]

    Vehicle trajectory prediction at intersections using interaction based generative adversarial networks,

    D. Roy, T. Ishizaka, C. K. Mohan, and A. Fukuda, “Vehicle trajectory prediction at intersections using interaction based generative adversarial networks,” in IEEE International Conference on Intelligent Transportation Systems (ITSC), pp. 2318–2323, 2019

    2019

  56. [64]

    Understanding posterior collapse in generative latent variable models,

    J. Lucas, G. Tucker, R. Grosse, and M. Norouzi, “Understanding posterior collapse in generative latent variable models,” inInternational Conference on Learning Representations (ICLR) Workshop DeepGenStruct, 2019

    2019

  57. [65]

    Improved techniques for training gans,

    T. Salimans, I. Goodfellow, W. Zaremba, V. Cheung, A. Radford, and X. Chen, “Improved techniques for training gans,” inConference on Neural Information Processing Systems (NeurIPS), vol. 29, 2016

    2016

  58. [66]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” inCon- ference on Neural Information Processing Systems (NeurIPS), vol. 33, pp. 6840– 6851, 2020

    2020

  59. [67]

    Denoising diffusion implicit models,

    J. Song, C. Meng, and S. Ermon, “Denoising diffusion implicit models,” in International Conference on Learning Representations (ICLR), 2021. Bibliography163

    2021

  60. [68]

    Language models are few-shot learners,

    T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, P. Dhariwal, A. Nee- lakantan, P. Shyam, G. Sastry, A. Askell,et al., “Language models are few-shot learners,” inConference on Neural Information Processing Systems (NeurIPS), vol. 33, pp. 1877–1901, 2020

    1901

  61. [69]

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

    W. Wu, X. Feng, Z. Gao, and Y. Kan, “Smart: scalable multi-agent real- time motion generation via next-token prediction,” inConference on Neural Information Processing Systems (NeurIPS), vol. 37, pp. 114048–114071, 2024

    2024

  62. [70]

    Behaviorgpt: Smart agent simulation for autonomous driving with next-patch prediction,

    Z. Zhou, H. Haibo, X. Chen, J. Wang, N. Guan, K. Wu, Y.-H. Li, Y.-K. Huang, and C. J. Xue, “Behaviorgpt: Smart agent simulation for autonomous driving with next-patch prediction,” inConference on Neural Information Processing Systems (NeurIPS), vol. 37, pp. 79597–79617, 2024

    2024

  63. [71]

    Amp: Autore- gressive motion prediction revisited with next token prediction for autonomous driving,

    X. Jia, S. Shi, Z. Chen, L. Jiang, W. Liao, T. He, and J. Yan, “Amp: Autore- gressive motion prediction revisited with next token prediction for autonomous driving,”arXiv preprint arXiv:2403.13331, 2024

    2024 arXiv

  64. [72]

    Closed-loop supervised fine-tuning of tokenized traffic models,

    Z. Zhang, P. Karkus, M. Igl, W. Ding, Y. Chen, B. Ivanovic, and M. Pavone, “Closed-loop supervised fine-tuning of tokenized traffic models,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 5422–5432, 2025

    2025

  65. [73]

    Deep un- supervised learning using nonequilibrium thermodynamics,

    J. Sohl-Dickstein, E. Weiss, N. Maheswaranathan, and S. Ganguli, “Deep un- supervised learning using nonequilibrium thermodynamics,” inInternational Conference on Machine Learning (ICML), pp. 2256–2265, pmlr, 2015

    2015

  66. [74]

    High-resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High-resolution image synthesis with latent diffusion models,” inIEEE/CVF International Conference on Computer Vision (ICCV), pp. 10684–10695, 2022

    2022

  67. [75]

    Gaia-1: A generative world model for autonomous driving,

    A. Hu, L. Russell, H. Yeo, Z. Murez, G. Fedoseev, A. Kendall, J. Shotton, and G. Corrado, “Gaia-1: A generative world model for autonomous driving,”arXiv preprint arXiv:2309.17080, 2023

    2023 arXiv

  68. [76]

    Leapfrog diffusion model for stochastic trajectory prediction,

    W. Mao, C. Xu, Q. Zhu, S. Chen, and Y. Wang, “Leapfrog diffusion model for stochastic trajectory prediction,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 5517–5526, 2023

    2023

  69. [77]

    Dice: Diverse diffusion model with scoring for trajectory prediction,

    Y. Choi, R. C. Mercurius, S. M. A. Shabestary, and A. Rasouli, “Dice: Diverse diffusion model with scoring for trajectory prediction,” inIEEE Intelligent Vehicles Symposium (IV), pp. 3023–3029, IEEE, 2024

    2024

  70. [78]

    On information and sufficiency,

    S. Kullback and R. A. Leibler, “On information and sufficiency,”The annals of mathematical statistics, vol. 22, no. 1, pp. 79–86, 1951

    1951

  71. [79]

    Trajectories as markov-states for long term traffic scene prediction,

    J. Reichardt, “Trajectories as markov-states for long term traffic scene prediction,” in14-th UniDAS FAS-Workshop, p. 14, 2022

    2022

  72. [80]

    K. P. Murphy,Machine learning: a probabilistic perspective. MIT press, 2012. 164Bibliography

    2012

  73. [81]

    Efron,Large-scale inference: empirical Bayes methods for estimation, testing, and prediction, vol

    B. Efron,Large-scale inference: empirical Bayes methods for estimation, testing, and prediction, vol. 1. Cambridge University Press, 2012

    2012

  74. [82]

    Performance analysis of 10 models of 3d lidars for automated driving,

    J. Lambert, A. Carballo, A. M. Cano, P. Narksri, D. Wong, E. Takeuchi, and K. Takeda, “Performance analysis of 10 models of 3d lidars for automated driving,” IEEE Access, vol. 8, pp. 131699–131722, 2020

    2020

  75. [83]

    Perception characteristics distance: Measuring stability and robustness of perception system in dynamic conditions under a certain decision rule,

    B. Jiang, L. Shi, Z. Lin, L. Stowe, and F. Guo, “Perception characteristics distance: Measuring stability and robustness of perception system in dynamic conditions under a certain decision rule,”arXiv preprint arXiv:2506.09217, 2025

    2025

  76. [84]

    Kämpchen,Feature-level fusion of laser scanner and video data for advanced driver assistance systems

    N. Kämpchen,Feature-level fusion of laser scanner and video data for advanced driver assistance systems. PhD thesis, Universität Ulm, 2007

    2007

  77. [85]

    Tensorflow distributions,

    J. V. Dillon, I. Langmore, D. Tran, E. Brevdo, S. Vasudevan, D. Moore, B. Patton, A. Alemi, M. Hoffman, and R. A. Saurous, “Tensorflow distributions,”arXiv preprint arXiv:1711.10604, 2017

    2017 arXiv

  78. [86]

    Information theory and an extension of the maximum likelihood principle,

    H. Akaike, “Information theory and an extension of the maximum likelihood principle,” inSelected papers of hirotugu akaike, pp. 199–213, Springer, 1998

    1998

  79. [87]

    Estimating the dimension of a model,

    G. Schwarz, “Estimating the dimension of a model,”The annals of statistics, pp. 461–464, 1978

    1978

  80. [88]

    Maximum likelihood estimates of linear dynamic systems,

    H. E. Rauch, F. Tung, and C. T. Striebel, “Maximum likelihood estimates of linear dynamic systems,”AIAA journal, vol. 3, no. 8, pp. 1445–1450, 1965

    1965

  81. [89]

    A new approach to linear filtering and prediction problems,

    R. E. Kalman, “A new approach to linear filtering and prediction problems,” Journal of Basic Engineering, vol. 82, no. 1, pp. 35–45, 1960

    1960

  82. [90]

    An introduction to gaussian processes for the kalman filter expert,

    S. Reece and S. Roberts, “An introduction to gaussian processes for the kalman filter expert,” inInternational Conference on Information Fusion, pp. 1–9, 2010

    2010

  83. [91]

    Acceleration-deceleration behaviour of various vehicle types,

    P. S. Bokare and A. K. Maurya, “Acceleration-deceleration behaviour of various vehicle types,”Transportation research procedia, vol. 25, pp. 4733–4749, 2017

    2017

  84. [92]

    Pedestrian acceleration and speeds,

    J. Zębala, P. Ciępka, and A. Reza, “Pedestrian acceleration and speeds,”Problems of Forensic Sciences, vol. 91, pp. 227–234, 2012

    2012

  85. [93]

    Adam: A method for stochastic optimization,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,”arXiv preprint arXiv:1412.6980, 2014

    2014 arXiv

  86. [94]

    Multipath++: Efficient information fusion and trajectory aggregation for behavior prediction,

    B. Varadarajan, A. Hefny, A. Srivastava, K. S. Refaat, N. Nayakanti, A. Cornman, K. Chen, B. Douillard, C. P. Lam, D. Anguelov,et al., “Multipath++: Efficient information fusion and trajectory aggregation for behavior prediction,” inIEEE International Conference on Robotics an...

    2022

  87. [95]

    A systematic study of the class imbalance problem in convolutional neural networks,

    M. Buda, A. Maki, and M. A. Mazurowski, “A systematic study of the class imbalance problem in convolutional neural networks,”Neural networks, vol. 106, pp. 249–259, 2018. Bibliography165

    2018

  88. [96]

    Interactive multi-modal motion planning with branch model predictive control,

    Y. Chen, U. Rosolia, W. Ubellacker, N. Csomay-Shanklin, and A. D. Ames, “Interactive multi-modal motion planning with branch model predictive control,” IEEE Robotics and Automation Letters, vol. 7, no. 2, pp. 5365–5372, 2022

    2022

  89. [97]

    ASAM OpenDRIVE

    Association for Standardization of Automation and Measuring Systems (ASAM), “ASAM OpenDRIVE.” https://www.asam.net/standards/detail/opendrive/, 2023

    2023

  90. [98]

    Total least squares fitting of bézier and b-spline curves to ordered data,

    C. F. Borges and T. Pastva, “Total least squares fitting of bézier and b-spline curves to ordered data,”Computer Aided Geometric Design, vol. 19, no. 4, pp. 275–289, 2002

    2002

  91. [99]

    G. H. Golub and C. F. Van Loan,Matrix Computations. The Johns Hopkins University Press, 3rd ed., 1996

    1996

  92. [100]

    borgespastva.m

    MathWorks, “borgespastva.m.” https://www.mathworks.com/matlabcentral/ fileexchange/46406-borgespastva-m, 2023

    2023

  93. [101]

    Efficient estimation of word representations in vector space,

    T. Mikolov, K. Chen, G. Corrado, and J. Dean, “Efficient estimation of word representations in vector space,”arXiv preprint arXiv:1301.3781, 2013

    2013 arXiv

  94. [102]

    On layer normalization in the transformer architecture,

    R. Xiong, Y. Yang, D. He, K. Zheng, S. Zheng, C. Xing, H. Zhang, Y. Lan, L. Wang, and T. Liu, “On layer normalization in the transformer architecture,” inInternational Conference on Machine Learning (ICML), pp. 10524–10533, 2020

    2020

  95. [103]

    Sgdr: Stochastic gradient descent with warm restarts,

    I. Loshchilov and F. Hutter, “Sgdr: Stochastic gradient descent with warm restarts,”arXiv preprint arXiv:1608.03983, 2016

    2016 arXiv

  96. [104]

    From marginal to joint predictions: Evaluating scene-consistent trajectory prediction approaches for automated driving,

    F. Konstantinidis, A. D. Guerreiro, R. Trumpp, M. Sackmann, U. Hofmann, M. Caccamo, and C. Stiller, “From marginal to joint predictions: Evaluating scene-consistent trajectory prediction approaches for automated driving,”arXiv preprint arXiv:2507.05254, 2025

    2025 arXiv

  97. [105]

    Thomas: Trajectory heatmap output with learned multi-agent sampling,

    T. Gilles, S. Sabatini, D. Tsishkou, B. Stanciulescu, and F. Moutarde, “Thomas: Trajectory heatmap output with learned multi-agent sampling,” inInternational Conference on Learning Representations (ICLR), 2022

    2022

  98. [106]

    M2i: From factored marginal trajectory prediction to interactive prediction,

    Q. Sun, X. Huang, J. Gu, B. C. Williams, and H. Zhao, “M2i: From factored marginal trajectory prediction to interactive prediction,” inIEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR), pp. 6543–6552, 2022

    2022

  99. [107]

    Qcnext: A next- generation framework for joint multi-agent trajectory prediction,

    Z. Zhou, Z. Wen, J. Wang, Y.-H. Li, and Y.-K. Huang, “Qcnext: A next- generation framework for joint multi-agent trajectory prediction,”arXiv preprint arXiv:2306.10508, 2023

    2023 arXiv

  100. [108]

    Motion planning under uncertainty: Integrating learning-based multi-modal predictors into branch model predictive control,

    M.-K. Bouzidi, B. Derajic, D. Goehring, and J. Reichardt, “Motion planning under uncertainty: Integrating learning-based multi-modal predictors into branch model predictive control,” inIEEE International Conference on Intelligent Trans- portation Systems (ITSC), pp. 2592–2598,...

    2024

  101. [109]

    Racp: Risk-aware contingency planning with multi-modal predictions,

    K. A. Mustafa, D. J. Ornia, J. Kober, and J. Alonso-Mora, “Racp: Risk-aware contingency planning with multi-modal predictions,”IEEE Transactions on Intelligent Vehicles, 2024

    2024

  102. [110]

    Closing the loop: Motion prediction models beyond open-loop benchmarks,

    M.-K. Bouzidi, C. Schlauch, N. Scheuerer, Y. Yao, N. Klein, D. Göhring,et al., “Closing the loop: Motion prediction models beyond open-loop benchmarks,” in IEEE International Conference on Intelligent Transportation Systems (ITSC), 2025

    2025

  103. [111]

    Annealed winner- takes-all for motion forecasting,

    Y. Xu, V. Letzelter, M. Chen, É. Zablocki, and M. Cord, “Annealed winner- takes-all for motion forecasting,” inIEEE International Conference on Robotics and Automation (ICRA), 2025

    2025

  104. [112]

    Nashformer: Leveraging local nash equilibria for semantically diverse trajectory prediction,

    J. Lidard, O. So, Y. Zhang, J. DeCastro, X. Cui, X. Huang,et al., “Nashformer: Leveraging local nash equilibria for semantically diverse trajectory prediction,” arXiv preprint arXiv:2305.17600, 2023

    2023 arXiv

  105. [113]

    Diversitygan: Diversity-aware vehicle motion prediction via latent semantic sampling,

    X. Huang, S. G. McGill, J. A. DeCastro, L. Fletcher, J. J. Leonard, B. C. Williams,et al., “Diversitygan: Diversity-aware vehicle motion prediction via latent semantic sampling,”IEEE Robotics and Automation Letters, vol. 5, no. 4, pp. 5089–5096, 2020

    2020

  106. [114]

    Diverse sampling for normalizing flow based trajectory forecasting,

    Y. J. Ma, J. P. Inala, D. Jayaraman, and O. Bastani, “Diverse sampling for normalizing flow based trajectory forecasting,”arXiv preprint arXiv:2011.15084, 2020

    2011 arXiv

  107. [115]

    Under- standing and mitigating copying in diffusion models,

    G. Somepalli, V. Singla, M. Goldblum, J. Geiping, and T. Goldstein, “Under- standing and mitigating copying in diffusion models,” inConference on Neural Information Processing Systems (NeurIPS), vol. 36, pp. 47783–47803, 2023

    2023

  108. [116]

    The waymo open sim agents challenge,

    N. Montali, J. Lambert, P. Mougin, A. Kuefler, N. Rhinehart, M. Li, C. Gulino, T. Emrich, Z. Yang, S. Whiteson,et al., “The waymo open sim agents challenge,” inConference on Neural Information Processing Systems (NeurIPS), vol. 36, 2024

    2024

  109. [117]

    Jfp: Joint future prediction with interactive multi-agent modeling for autonomous driving,

    W. Luo, C. Park, A. Cornman, B. Sapp, and D. Anguelov, “Jfp: Joint future prediction with interactive multi-agent modeling for autonomous driving,” in Conference on Robot Learning (CoRL), pp. 1457–1467, 2023

    2023

  110. [118]

    Fast graph representation learning with pytorch geometric,

    M. Fey and J. E. Lenssen, “Fast graph representation learning with pytorch geometric,” inInternational Conference on Learning Representations (ICLR) Workshop, 2019

    2019

  111. [119]

    Gorela: Go relative for viewpoint-invariant motion forecasting,

    A. Cui, S. Casas, K. Wong, S. Suo, and R. Urtasun, “Gorela: Go relative for viewpoint-invariant motion forecasting,”arXiv preprint arXiv:2211.02545, 2022

    2022 arXiv

  112. [120]

    Decoupled weight decay regularization,

    I. Loshchilov and F. Hutter, “Decoupled weight decay regularization,”arXiv preprint arXiv:1711.05101, 2017

    2017 arXiv

  113. [121]

    Dsdnet: Deep structured self-driving network,

    W. Zeng, S. Wang, R. Liao, Y. Chen, B. Yang, and R. Urtasun, “Dsdnet: Deep structured self-driving network,” inEuropean Conference on Computer Vision (ECCV), pp. 156–172, 2020. Bibliography167

    2020

  114. [122]

    Behaviorally diverse traffic simulation via reinforcement learning,

    S. Shiroshita, S. Maruyama, D. Nishiyama, M. Y. Castro, K. Hamzaoui, G. Ros- man,et al., “Behaviorally diverse traffic simulation via reinforcement learning,” in IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 2103–2110, 2020. Appendix A Example...

    2020

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.