Pith. sign in

REVIEW 4 major objections 7 minor 1 cited by

Beyond Patterns: Harnessing Causal Logic for Autonomous Driving Trajectory Prediction

T0 review · 4 major / 7 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read This paper claims to beat prior trajectory predictors on five driving datasets by treating the road map and nearby agents as confounders and adjusting for them causally.

desk verdict A plausible trajectory predictor with competitive numbers, but the causal inference story is not derived—send to review, don't take the causal claims at face value. read the letter →

arxiv 2505.06856 v1 pith:QGXJGVQV submitted 2025-05-11 cs.AI cs.RO

Bonan Wang , Haicheng Liao , Chengyue Wang , Bin Rao , Yanchen Guan , Guyang Yu , Jiaxun Zhang , Songning Lai
show 2 more authors
Chengzhong Xu Zhenning Li
This is my paper · ORCID
classification cs.AIcs.RO
keywords trajectorypredictioncausalinferencebackdooradjustmentcounterfactualanalysisautonomousdrivingdiffusionmodeldomaingeneralizationspuriouscorrelation
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 paper tries to establish that causal inference, not just pattern matching, is what makes trajectory prediction generalize across driving scenes. The authors build a causal graph in which the target vehicle's past motion and the surrounding agents cause the future trajectory, while the road layout confounds everything, creating backdoor paths that a purely correlational model would learn as spurious shortcuts. They implement backdoor adjustment by generating $n$ alternative road layouts with a diffusion model and averaging predictions over them with equal weight $1/n$, then subtract a counterfactual prediction obtained by zeroing out the target's history to remove the remaining confounding influence of other agents. On ApolloScape, nuScenes, NGSIM, HighD, and MoCAD the resulting model reports lower error metrics than the compared baselines, stays accurate under added noise and dropped frames, and keeps its advantage when the causal module is grafted onto another predictor. If the causal attribution is correct, this is a route to predictors that trust the actual cause of a maneuver, such as a pedestrian entering a crosswalk, rather than a correlation learned from the training distribution.

What carries the argument

The load-bearing object is the causal graph $X \to Y$, $T \to Y$ with $S \to X$ and $S \to T$ creating backdoor paths, operationalized as a pair of adjustment equations: backdoor adjustment $\tilde{Y} = \sum_{i=1}^{n} g_\theta(X, S=s_i, T)P(s_i)$ with uniform weight $P(s_i)=1/n$, where the $n$ road layouts $s_i$ are produced by a diffusion model that noisifies and regenerates the encoded map token; and counterfactual adjustment $\tilde{Y}_c = \sum_{i=1}^{n} g_\theta(\mathrm{do}(X=X_c), S=s_i, T)P(s_i)$, where $X_c$ zeroes out the target agent's history. Subtracting the counterfactual from the factual, $Y = \tilde{Y} - \tilde{Y}_c$, is meant to cancel the non-causal contribution of the temporal and spatial context. Around this identity sit the spatial, BEV, and temporal encoders that tokenize each input, a targeted multi-view attention module fusing the tokens, and a cross-modal progressive-fusion decoder that refines an anchor query over several stages, alongside a dual-scale CNN branch, feeding a causal decoder that emits multimodal trajectory hypotheses. The diffusion-based sampling is what makes the stratification in the backdoor sum tractable: it generates the alternative road layouts that the adjustment averages over.

What would settle it

Keep the architecture identical but replace the diffusion-based backdoor module with the single observed road layout, so $n=1$ and no stratification occurs: if error metrics barely move, the adjustment term is not carrying the claimed benefit. The sharper test is synthetic: generate traffic scenes with a known confounder, train the causal model and a purely correlational baseline on them, and check whether the causal model recovers the true intervention effect, for instance whether it learns to predict a stop at a crosswalk when pedestrians are present, purely from the known causal structure.

Watch

Extended reading notes

Core claim

The paper's central claim is that decomposing the driving environment into a spatial component $S$ and a temporal component $T$, and then intervening on both through causal machinery, removes the spurious correlations that bind ordinary predictors to their training scenes. The model computes $\tilde{Y} = \sum_{i=1}^{n} g_\theta(X, S=s_i, T)P(s_i)$ with $P(s_i)=1/n$, which is backdoor adjustment over diffusion-generated road layouts, and the counterfactual $\tilde{Y}_c = \sum_{i=1}^{n} g_\theta(\mathrm{do}(X=X_c), S=s_i, T)P(s_i)$, in which the target's history is replaced by a counterfactual value; the final prediction is the difference $Y = \tilde{Y} - \tilde{Y}_c$. Trained in two stages, with diffusion loss first and then the full model, the predictor reports lower WSADE and WSFDE on ApolloScape, minADE and FDE on nuScenes, and RMSE on NGSIM, HighD, and MoCAD than the compared baselines, keeps 0.28M parameters with 57ms inference on nuScenes, and stays ahead under added noise and frame dropout. The causal module also plugs into the PGP baseline and improves its turning predictions, while ablating the causal components degrades performance the most.

Load-bearing premise

The load-bearing premise is that the road layout $S$ really is a confounder of the target's future trajectory and that averaging over diffusion-generated layouts, each weighted equally at $1/n$, faithfully implements backdoor adjustment; if the generated layouts do not track the true distribution of road layouts, or if an unmodeled confounder such as traffic-signal phase or driver identity is the real driver of behavior, the causal claims do not follow even if the accuracy numbers reproduce.

Editorial extensions

If this is right

  • If the causal decomposition is correct, predictors can ignore correlations that do not survive intervention, so accuracy on rare and unseen scenes, including crosswalks, intersections, and unfamiliar road layouts, should improve without new labeled data.
  • Because the causal module also improves the PGP baseline when plugged in, existing predictors could adopt the module without a full redesign, making the method an upgrade rather than a replacement.
  • The robustness results imply that under sensor noise and missing frames, modeled as curvature-scaled Gaussian noise and random frame dropout, prediction quality degrades more slowly than for the compared baselines, which matters for real perception stacks.
  • The efficiency figures of 0.28M parameters and 57ms per sample on nuScenes imply the causal machinery does not add a latency barrier to real-time use.

Reading between the lines

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

  • The uniform prior $P(s_i)=1/n$ is the most stress-testable design choice: a prior learned from road-layout statistics, or one conditioned on traffic rules, could sharpen the backdoor sum if the diffusion samples are informative about the true confounder distribution.
  • The same backdoor-plus-counterfactual subtraction could transfer to other spatiotemporal forecasting problems, such as pedestrian intent or vessel and drone motion, wherever a static scene layout confounds an agent's observed behavior.
  • A direct test of whether the causal story rather than the extra parameters carries the gain: keep the architecture fixed and set $n=1$ so the backdoor sum collapses to the single observed layout; if metrics barely move, the adjustment is not the active ingredient.
  • The combination rule $Y = \tilde{Y} - \tilde{Y}_c$ treats the causal effect as additive; whether a ratio or a learned gating of factual and counterfactual terms would serve better is a variant the paper does not test.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 7 minor

Summary. The paper proposes a trajectory prediction framework for autonomous driving that combines spatial, BEV, and temporal encoders with a diffusion-based backdoor adjustment module, counterfactual analysis via zeroed historical trajectories, and a cross-modal progressive fusion decoder. The authors claim that this causal inference paradigm removes spurious correlations, improves robustness and generalization, and achieves state-of-the-art accuracy on five real-world datasets (ApolloScape, nuScenes, NGSIM, HighD, MoCAD), while also reporting efficiency and plug-and-play validation results.

Significance. If the causal identification claims were valid, the paper would offer a principled way to reduce confounder bias in trajectory prediction and would strengthen the case for causal methods in autonomous driving. The empirical contribution is substantial: evaluation across five datasets with multiple metrics, robustness to noise and missing frames, a domain generalization study, a parameter-efficient real-time model, and a plug-and-play integration into an existing baseline. However, the causal claims are not established by the derivations or experiments as written; the central contribution is currently an architecture with causal labels rather than a demonstrated causal estimator.

major comments (4)
  1. [Section 3.2, Eq. (1)] The backdoor adjustment formula in Eq. (1) is not derived from the stated causal graph. For Eq. (1) to be a valid backdoor adjustment, g_theta must estimate P(Y|X,T,S=s_i), the s_i must be sampled from the true confounder distribution P(S), and P(s_i) must be the corresponding probability masses. The paper instead sets P(s_i)=1/n on a 'maximum entropy' principle without justifying uniform weights over diffusion-generated tokens, and it does not show that the diffusion samples approximate any meaningful stratum distribution over S. As written, Eq. (1) is a model definition, not an identifiable causal estimand.
  2. [Section 3.2, Eqs. (2) and (3)] The counterfactual analysis in Eq. (2) is not Pearl counterfactual inference. Setting X_c to zero vectors in the progressive fusion module does not specify an abduction step, an intervention do(X=X_c), or a prediction step on a modified structural model. Consequently, the composite output Y = Y_tilde - Y_tilde_c is not a standard causal estimand, and no identification argument is provided for this subtraction. The counterfactual branch is a heuristic architectural choice rather than a causal operation.
  3. [Section 3.2, Figure 1(c)] The causal graph labels T as a confounder, but Figure 1(c) does not draw an edge T -> X. A confounder of the effect of X on Y requires a backdoor path with an arrow into X. Either the graph is mis-specified or T is not a confounder, which undermines the motivation for using counterfactual subtraction to 'eliminate the confounding effects of temporal agent data T.' Additionally, the claim that the diffusion-based backdoor adjustment severs S -> X and S -> T is contradicted by Eqs. (5)-(7), where the BEV encoder still injects spatial information S through B^h and the spatial token S^{h,i} is used directly in attention.
  4. [Section 4.3, Tables 6 and 7] The ablation study does not isolate the causal adjustment. Method D removes all causal modules and simultaneously replaces the decoder with a simple GRU, so the performance drop could be due to the diffusion augmentation, the attention modules, or the decoder architecture rather than the causal operations. Moreover, Tables 1-4 and 7 report no error bars, confidence intervals, or repeated-trial statistics, so the claimed improvements (for example, a 1.84% WSADE gain on ApolloScape) are not shown to be statistically reliable.
minor comments (7)
  1. [Section 3.2] The heading 'Casual Inference' should be 'Causal Inference'.
  2. [Section 1] There is an inconsistent spacing in 'A Vs' and 'A V'; please use a consistent notation such as 'AVs' or 'autonomous vehicles'.
  3. [Table 1 caption] The caption states 'Bold and underlined values represent the best and second-best performance,' but the table does not visibly use underlining.
  4. [Table 3] The same BAT baseline appears with different reference years ('Liao et al., 2024c' and 'Liao et al., 2024d') in the HighD and MoCAD rows; please verify the citations.
  5. [Section 3.4, Eq. (10)] The definition of L0 is described as 'varies by dataset,' but the precise form for each dataset is not given; please specify the exact formula used for each metric.
  6. [Figure 3] The subfigures in Figure 3 lack axis labels and error bars; please add these so the minADE5 values and variability can be assessed.
  7. [Section 4.2] There is a typo in 'these results validate the prediction accurancy of our model'; 'accurancy' should be 'accuracy'.

Circularity Check

1 steps flagged · score 4.0 of 10

Causal output is defined as Y = Ytilde - Ytilde_c, making the causal claim self-definitional; the empirical benchmark results are independent.

  1. self definitional [Section 3.2 (Casual Inference), final equation after Eq. (2); implemented in Section 3.3 (Cross-modal Progressive Fusion, Causal Decoder)]
    "We combine backdoor adjustment and counterfactual analysis together in a comprehensive manner to derive:Y = ˜Y− ˜Yc. ... To reduce the impact of confounding variables T in the trajectory prediction system, we substitute the historical trajectories with zero vectors and rerun the progressive fusion module to generate counterfactual anchor Qi c."

    The prediction Y is asserted to be the result of causal reasoning, but the paper defines it as the difference of two network outputs: the factual composite token and the counterfactual composite token obtained by zeroing the historical trajectory. No do-calculus or abduction step shows that this difference equals a causal effect; the equality Y = Ytilde - Ytilde_c is true by construction of the model architecture. Consequently the causal conclusion is self-definitional: the estimator is the subtraction, so any claim that 'causal inference yields Y' reduces to the naming of the subtraction rather than to an independently identified causal quantity.

full rationale

The empirical trajectory-prediction claims are not circular: Tables 1-3 report held-out test metrics under standard losses, and no parameter is fitted to the reported quantity and then renamed a prediction. The many self-citations are to the authors' own baselines and datasets (e.g., HLTP++, BAT, NEST) and are not used to justify the causal identification, so they are not load-bearing circularity. The uniform P(s_i)=1/n in Eq. (1) is an unverified modeling assumption (external citation to Ge et al.), which is a correctness/identification risk rather than a circular step. The one genuine circularity is the causal claim itself: Eq. (2) is implemented as zeroing the history in the same progressive-fusion network, and the final output is defined as Ytilde - Ytilde_c. Thus the 'causal effect' equals the model definition by construction, not a result derived from Pearl-style identification. This partial circularity affects the causal framing but not the standalone empirical accuracy results, so the score is moderate.

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

The central causal claim depends on an assumed causal graph, uniform backdoor strata, a heuristic counterfactual, and the diffusion model approximating P(S). These are not independently validated, and no code or detailed hyperparameters are provided.

free parameters (4)
  • n (number of backdoor strata)
    The number of environmental categories in Eq (1) is assumed; P(s_i)=1/n is set by 'maximum entropy' without justification or sensitivity analysis.
  • diffusion steps m
    The number of diffusion steps in the backdoor adjustment module is not specified.
  • progressive fusion iterations T_rec
    The number of refinement iterations in the progressive query module is not specified.
  • noise magnitude alpha in robustness test = 8, 16 (test)
    The robustness augmentation in Eq (11) uses alpha=8,16 at test time; these are chosen ad hoc and are not part of the central method.
assumptions (4)
  • domain assumption The causal graph S -> X, S -> T, X -> Y, T -> Y correctly represents the data-generating process.
    Stated in Sec 3.2 and Figure 1(c); no test of faithfulness or graph fit is provided.
  • ad hoc to paper Backdoor adjustment over S with uniform P(s_i)=1/n identifies the causal effect of (X,T) on Y.
    Eq (1) assumes that stratifying by S and averaging uniformly removes confounding; this is not a standard backdoor adjustment because P(S) is unknown and set to uniform.
  • ad hoc to paper Counterfactual replacement X_c=0 corresponds to a valid intervention do(X=X_c).
    Zeroing the history trajectory is not derived from the causal model; it is a heuristic to 'isolate' T.
  • ad hoc to paper Generated samples from the diffusion model approximate the confounder distribution P(S).
    Sec 3.3 'Diffusion-based Backdoor Adjustment' assumes the denoised spatial maps form a stratification of S; no theoretical guarantee is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Beyond Patterns: Harnessing Causal Logic for Autonomous Driving Trajectory Prediction." pith.science (2026). https://pith.science/paper/QGXJGVQV

@misc{pith2026250506856,
  author       = {Pith},
  title        = {Pith review of: Beyond Patterns: Harnessing Causal Logic for Autonomous Driving Trajectory Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QGXJGVQV}},
  note         = {Machine review of arXiv:2505.06856}
}
read the original abstract

Accurate trajectory prediction has long been a major challenge for autonomous driving (AD). Traditional data-driven models predominantly rely on statistical correlations, often overlooking the causal relationships that govern traffic behavior. In this paper, we introduce a novel trajectory prediction framework that leverages causal inference to enhance predictive robustness, generalization, and accuracy. By decomposing the environment into spatial and temporal components, our approach identifies and mitigates spurious correlations, uncovering genuine causal relationships. We also employ a progressive fusion strategy to integrate multimodal information, simulating human-like reasoning processes and enabling real-time inference. Evaluations on five real-world datasets--ApolloScape, nuScenes, NGSIM, HighD, and MoCAD--demonstrate our model's superiority over existing state-of-the-art (SOTA) methods, with improvements in key metrics such as RMSE and FDE. Our findings highlight the potential of causal reasoning to transform trajectory prediction, paving the way for robust AD systems.

Figures

Figures reproduced from arXiv: 2505.06856 by the authors.

Figure 1
Figure 1. Illustration of causal relationships in traffic scenarios. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overall framework of our two-stage model. The first stage involves token extraction with spatial, BEV, and temporal encoders [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Exploration of the domain generalization ability for dif [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Comparative analysis of the impact of causal inference on our model and others across challenging scenes. For clarity, the number [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. AMD: Adaptive Momentum and Decoupled Contrastive Learning Framework for Robust Long-Tail Trajectory Prediction

    cs.CV 2025-07 conditional novelty 5.0 of 10

    AMD combines momentum and decoupled contrastive learning, trajectory augmentations, and online clustering to improve prediction on rare driving scenarios.

Reference graph

Works this paper leans on

33 extracted references · 32 canonical work pages · cited by 1 Pith paper

  1. [1]

    Generative causal representation learning for out-of-distribution mo- tion forecasting

    [Bagi et al., 2023] Shayan Shirahmad Gale Bagi, Zahra Gharaee, Oliver Schulte, and Mark Crowley. Generative causal representation learning for out-of-distribution mo- tion forecasting. In ICML, pages 31596–31612. PMLR,

    work page 2023

  2. [8]

    Convolutional social pooling for vehicle trajec- tory prediction

    [Deo and Trivedi, 2018] Nachiket Deo and Mohan M Trivedi. Convolutional social pooling for vehicle trajec- tory prediction. In Proceedings of the IEEE conference on computer vision and pattern recognition workshops , pages 1468–1476,

    work page 2018

  3. [10]

    Tpnet: Trajectory proposal network for motion prediction,

    [Fang et al., 2021] Liangji Fang, Qinhong Jiang, Jianping Shi, and Bolei Zhou. Tpnet: Trajectory proposal network for motion prediction,

    work page 2021

  4. [12]

    World models for autonomous driv- ing: An initial survey

    [Guan et al., 2024] Yanchen Guan, Haicheng Liao, Zhen- ning Li, Jia Hu, Runze Yuan, Yunjian Li, Guohui Zhang, and Chengzhong Xu. World models for autonomous driv- ing: An initial survey. IEEE Transactions on Intelligent Vehicles,

    work page 2024

  5. [13]

    The apolloscape dataset for au- tonomous driving

    [Huang et al., 2018] Xinyu Huang, Xinjing Cheng, Qichuan Geng, Binbin Cao, Dingfu Zhou, Peng Wang, Yuanqing Lin, and Ruigang Yang. The apolloscape dataset for au- tonomous driving. In CVPR workshops, pages 954–960,

    work page 2018

  6. [14]

    Lapred: Lane-aware prediction of multi-modal future trajectories of dynamic agents,

    [Kim et al., 2021] ByeoungDo Kim, Seong Hyeon Park, Seokhwan Lee, Elbek Khoshimjonov, Dongsuk Kum, Jun- soo Kim, Jeong Soo Kim, and Jun Won Choi. Lapred: Lane-aware prediction of multi-modal future trajectories of dynamic agents,

    work page 2021

  7. [15]

    The highd dataset: A drone dataset of naturalistic vehicle trajectories on ger- man highways for validation of highly automated driving systems

    [Krajewski et al., 2018] Robert Krajewski, Julian Bock, Laurent Kloeker, and Lutz Eckstein. The highd dataset: A drone dataset of naturalistic vehicle trajectories on ger- man highways for validation of highly automated driving systems. In 2018 21st International Conference on Intel- ligent Transportation Systems (ITSC) , pages 2118–2125,

    work page 2018

  8. [16]

    Causal inference

    [Kuang et al., 2020] Kun Kuang, Lian Li, Zhi Geng, Lei Xu, Kun Zhang, Beishui Liao, Huaxin Huang, Peng Ding, Wang Miao, and Zhichao Jiang. Causal inference. En- gineering, 6(3):253–263,

    work page 2020

Show all 33 references
  1. [17]

    Traj-llm: A new exploration for empowering trajectory prediction with pre-trained large language models

    [Lan et al., 2024] Zhengxing Lan, Lingshan Liu, Bo Fan, Yisheng Lv, Yilong Ren, and Zhiyong Cui. Traj-llm: A new exploration for empowering trajectory prediction with pre-trained large language models. IEEE Transactions on Intelligent Vehicles,

    2024

  2. [18]

    Deep causal inference for understanding the impact of meteorologi- cal variations on traffic

    [Li et al., 2024] Can Li, Wei Liu, and Hai Yang. Deep causal inference for understanding the impact of meteorologi- cal variations on traffic. Transportation Research Part C: Emerging Technologies, 165:104744,

    2024

  3. [19]

    [Liao et al., 2024b] Haicheng Liao, Yongkang Li, Zhenning Li, Chengyue Wang, Guofa Li, Chunlin Tian, Zilin Bian, Kaiqun Zhu, Zhiyong Cui, and Jia Hu

    AI for Good. [Liao et al., 2024b] Haicheng Liao, Yongkang Li, Zhenning Li, Chengyue Wang, Guofa Li, Chunlin Tian, Zilin Bian, Kaiqun Zhu, Zhiyong Cui, and Jia Hu. Less is more: Ef- ficient brain-inspired learning for autonomous driving tra- jectory prediction. In ECAI 2024, pa...

    2024

  4. [20]

    Mftraj: Map-free, behavior-driven trajectory prediction for autonomous driv- ing

    [Liao et al., 2024e] Haicheng Liao, Zhenning Li, Chengyue Wang, Huanming Shen, Bonan Wang, Dongping Liao, Guofa Li, and Chengzhong Xu. Mftraj: Map-free, behavior-driven trajectory prediction for autonomous driv- ing. arXiv preprint arXiv:2405.01266,

    arXiv

  5. [21]

    Human observation-inspired trajectory prediction for autonomous driving in mixed-autonomy traffic environments

    [Liao et al., 2024f] Haicheng Liao, Shangqian Liu, Yongkang Li, Zhenning Li, Chengyue Wang, Yun- jian Li, Shengbo Eben Li, and Chengzhong Xu. Human observation-inspired trajectory prediction for autonomous driving in mixed-autonomy traffic environments. In 2024 ICRA, pages 142...

    2024

  6. [22]

    Safecast: Risk-responsive motion forecasting for au- tonomous vehicles

    [Liao et al., 2025a] Haicheng Liao, Hanlin Kong, Bin Rao, Bonan Wang, Chengyue Wang, Guyang Yu, Yuming Huang, Ruru Tang, Chengzhong Xu, and Zhenning Li. Safecast: Risk-responsive motion forecasting for au- tonomous vehicles. arXiv preprint arXiv:2503.22541 ,

    arXiv

  7. [24]

    Attention based vehicle trajectory prediction

    [Messaoud et al., 2020] Kaouther Messaoud, Itheri Yahiaoui, Anne Verroust-Blondet, and Fawzi Nashashibi. Attention based vehicle trajectory prediction. IEEE Transactions on Intelligent Vehicles, 6(1):175–185,

    2020

  8. [25]

    Multi-agent trajectory prediction with heterogeneous edge-enhanced graph attention network

    [Mo et al., 2022] Xiaoyu Mo, Zhiyu Huang, Yang Xing, and Chen Lv. Multi-agent trajectory prediction with heterogeneous edge-enhanced graph attention network. IEEE Transactions on Intelligent Transportation Systems, 23(7):9554–9567,

    2022

  9. [26]

    Causality

    [Pearl, 2009] Judea Pearl. Causality. Cambridge university press,

    2009

  10. [28]

    Trajectron++: Dynamically-feasible trajectory forecasting with heteroge- neous data

    [Salzmann et al., 2020] Tim Salzmann, Boris Ivanovic, Punarjay Chakravarty, and Marco Pavone. Trajectron++: Dynamically-feasible trajectory forecasting with heteroge- neous data. In ECCV, pages 683–700. Springer,

    2020

  11. [29]

    Trajectory prediction for autonomous driving based on multiscale spatial-temporal graph

    [Tang et al., 2023] Luqi Tang, Fuwu Yan, Bin Zou, Wenbo Li, Chen Lv, and Kewei Wang. Trajectory prediction for autonomous driving based on multiscale spatial-temporal graph. IET Intelligent Transport Systems, 17(2):386–399,

    2023

  12. [30]

    Wsip: wave superposition inspired pooling for dynamic interactions-aware trajectory predic- tion

    [Wang et al., 2023] Renzhi Wang, Senzhang Wang, Hao Yan, and Xiang Wang. Wsip: wave superposition inspired pooling for dynamic interactions-aware trajectory predic- tion. In AAAI, volume 37, pages 4685–4692,

    2023

  13. [31]

    Congestion-aware multi-agent trajectory prediction for collision avoidance

    [Xie et al., 2021] Xu Xie, Chi Zhang, Yixin Zhu, Ying Nian Wu, and Song-Chun Zhu. Congestion-aware multi-agent trajectory prediction for collision avoidance. In 2021 ICRA, pages 13693–13700. IEEE,

    2021

  14. [32]

    Adapting to length shift: Flexilength network for trajectory prediction

    [Xu and Fu, 2024] Yi Xu and Yun Fu. Adapting to length shift: Flexilength network for trajectory prediction. In CVPR, pages 15226–15237,

    2024

  15. [33]

    A multi-task learning network with a collision-aware graph transformer for traffic-agents trajectory prediction

    [Yang et al., 2024] Biao Yang, Fucheng Fan, Rongrong Ni, Hai Wang, Ammar Jafaripournimchahi, and Hongyu Hu. A multi-task learning network with a collision-aware graph transformer for traffic-agents trajectory prediction. IEEE Transactions on Intelligent Transportation Systems,...

    2024

  16. [2009]

    Emsin: Enhanced multi-stream in- teraction network for vehicle trajectory prediction

    [Ren et al., 2024] Yilong Ren, Zhengxing Lan, Lingshan Liu, and Haiyang Yu. Emsin: Enhanced multi-stream in- teraction network for vehicle trajectory prediction. IEEE Transactions on Fuzzy Systems,

    2024

  17. [2014]

    nuscenes: A multimodal dataset for autonomous driving

    [Caesar et al., 2020] Holger Caesar, Varun Bankiti, Alex H Lang, Sourabh V ora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Giancarlo Baldan, and Oscar Beijbom. nuscenes: A multimodal dataset for autonomous driving. In CVPR, pages 11621–11631,

    2020

  18. [2018]

    Multimodal trajectory prediction conditioned on lane-graph traversals

    [Deo et al., 2022] Nachiket Deo, Eric Wolff, and Oscar Bei- jbom. Multimodal trajectory prediction conditioned on lane-graph traversals. In CoRL, pages 203–212. PMLR,

    2022

  19. [2019]

    Human trajectory prediction via counter- factual analysis

    [Chen et al., 2021] Guangyi Chen, Junlong Li, Jiwen Lu, and Jie Zhou. Human trajectory prediction via counter- factual analysis. In ICCV, pages 9824–9833,

    2021

  20. [2020]

    Multipath: Multiple probabilistic anchor trajectory hypotheses for behavior prediction,

    [Chai et al., 2019] Yuning Chai, Benjamin Sapp, Mayank Bansal, and Dragomir Anguelov. Multipath: Multiple probabilistic anchor trajectory hypotheses for behavior prediction,

    2019

  21. [2021]

    Causal intervention for human trajectory prediction with cross attention mechanism

    [Ge et al., 2023] Chunjiang Ge, Shiji Song, and Gao Huang. Causal intervention for human trajectory prediction with cross attention mechanism. In AAAI, volume 37, pages 658–666,

    2023

  22. [2022]

    Q-eanet: Implicit social modeling for trajectory prediction via experience-anchored queries

    [Chen et al., 2024] Jiuyu Chen, Zhongli Wang, Jian Wang, and Baigen Cai. Q-eanet: Implicit social modeling for trajectory prediction via experience-anchored queries. IET Intelligent Transport Systems, 18(6):1004–1015,

    2024

  23. [2023]

    Kolmogorov–smirnov test: Overview

    [Berger and Zhou, 2014] Vance W Berger and YanYan Zhou. Kolmogorov–smirnov test: Overview. Wiley stat- sref: Statistics reference online,

    2014

  24. [2024]

    Dacr- amtp: Adaptive multi-modal vehicle trajectory prediction for dynamic drivable areas based on collision risk

    [Cong et al., 2023] Peichao Cong, Yixuan Xiao, Xianquan Wan, Murong Deng, Jiaxing Li, and Xin Zhang. Dacr- amtp: Adaptive multi-modal vehicle trajectory prediction for dynamic drivable areas based on collision risk. IEEE Transactions on Intelligent Vehicles,

    2023

  25. [2025]

    Laformer: Trajectory pre- diction for autonomous driving with lane-aware scene con- straints

    [Liu et al., 2024] Mengmeng Liu, Hao Cheng, Lin Chen, Hellward Broszio, Jiangtao Li, Runjiang Zhao, Monika Sester, and Michael Ying Yang. Laformer: Trajectory pre- diction for autonomous driving with lane-aware scene con- straints. In CVPR, pages 2039–2049,

    2024

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.