Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

DeMo++: Motion Decoupling for Autonomous Driving

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read This paper claims that decoupling trajectory prediction into holistic motion intentions and fine temporal state queries, implemented with a hybrid Attention-Mamba decoder, yields state-of-the-art forecasts and plans on Argoverse 2…

desk verdict Useful extension of the DeMo line, but the headline SOTA on Argoverse 2 and nuPlan rests on an incomparable reorganized protocol; the nuScenes and NAVSIM numbers are cleaner evidence. read the letter →

arxiv 2507.17342 v2 pith:FSOE7BYP submitted 2025-07-23 cs.CV

classification cs.CV
keywords autonomousdrivingmotionforecastingplanningend-to-endtrajectorypredictionstatespacemodelsMambaquerydecoupling
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 argues that the dominant one-query-one-trajectory design in autonomous driving, where each learned query directly proposes one future trajectory, cannot fully represent how a trajectory unfolds over time. DeMo++ therefore splits trajectory modeling into holistic motion modes that capture possible directions of travel and fine temporal state queries that track the agent's progress at each future instant. The paper reports that this decoupling, implemented with a hybrid Attention and Mamba network plus cross-scene intention interaction and state-anchor refinement, reaches state-of-the-art results on Argoverse 2 and nuScenes for motion forecasting, nuPlan for motion planning, and NAVSIM for end-to-end planning. A sympathetic reader would care because the same representation and training recipe transfers across prediction, planning, and sensor-based end-to-end driving, suggesting a single trajectory-decoding idea can replace task-specific designs.

What carries the argument

The central object is the pair of decoupled query sets. Mode queries are learned embeddings that localize directional intention, while state queries are timestamp-initialized embeddings that track future progress, with one state query per future time step. The named modules are the Mode Localization Module, the State Consistency Module, and the Hybrid Coupling Module, and together they let intentions and states interact with the scene context, with each other, and across time; Mamba, a linear-time sequence model with selective state spaces, is used where temporal consistency matters. Cross-scene intention interaction projects historical trajectories into the current coordinate frame and uses trajectory embeddings inside a lightweight transformer to update current queries, while state-anchor-based refinement re-attends to scene context around each predicted waypoint under a distance mask. This combination is what the paper credits for both diversity of intentions and spatiotemporal accuracy.

What would settle it

Run DeMo++ on the official Argoverse 2 test server under the standard five-second-history protocol and compare with the paper's reported b-minFDE6 value of 1.74; alternatively, train a strong one-query baseline from the paper's tables on the same three-sub-scene sliding-window protocol. If DeMo++'s margin shrinks or reverses in either comparison, the claimed state-of-the-art performance depends on the reorganization rather than on decoupled queries.

Watch

Extended reading notes

Core claim

The central claim is that future trajectories are better represented by two coupled sets of queries than by one set. Mode queries localize the plausible directional intentions of an agent, while state queries, one per future timestamp, carry the evolving position and context of the trajectory. A mode-localization module with cross- and self-attention, and a state-consistency module with scene cross-attention plus bidirectional Mamba, are trained with auxiliary losses so each query type becomes meaningful; the two are then added into hybrid spatiotemporal queries and refined by hybrid attention and Mamba layers. The paper further claims two extensions of this decoupling help in continuous driving: cross-scene intention interaction aligns current mode and state queries with those from overlapping earlier sub-scenes, and state-anchor-based refinement uses distance-masked cross-attention at each waypoint to improve proposal trajectories and their probabilities. On the paper's own evaluations, DeMo++ outperforms all compared methods on the four named benchmarks, with DeMo-E2E++ achieving the best PDM score among end-to-end planners on NAVSIM.

Load-bearing premise

The load-bearing assumption is that cutting each Argoverse 2 scene into three sub-scenes with 3 seconds of history and each nuPlan scene into two sub-scenes with 1.5 seconds of history preserves the official evaluation task, so the reported scores can be fairly compared with methods that used the full 5 seconds of history.

Editorial extensions

If this is right

  • If the decoupled-query representation is correct, one-query-one-trajectory methods leave spatiotemporal detail on the table, and adding explicit state queries should improve displacement errors without ensembling.
  • Cross-scene intention interaction implies that snapshot-based datasets can be reorganized into continuous driving sequences, and that forecasts should stay coherent across overlapping time windows in real deployment.
  • State-anchor-based refinement should reduce unreasonable proposals such as collisions, because each trajectory is locally re-attended against nearby map elements and agents before its final probability is assigned.
  • The same mode-state decoding pipeline transfers from forecasting to ego planning and to end-to-end planning from sensor data, so a single trajectory decoder can serve multiple levels of the driving stack.
  • The hybrid Attention-Mamba decoder is reported as smaller and faster than transformer-only alternatives, which would make the approach practical for onboard deployment.

Reading between the lines

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

  • An extension the paper does not test is a matched baseline that runs a prior one-query model on the same three-sub-scene Argoverse 2 protocol; that comparison would separate gains from decoupling from gains caused by training on additional sub-scenes.
  • The sub-scene reorganization effectively expands the training set and shortens history, so part of the reported margin may come from data augmentation rather than the architecture; evaluating DeMo++ on the official server with the standard five-second history would test this.
  • The cross-scene mechanism suggests a natural temporal-consistency metric, such as disagreement between overlapping predictions from consecutive sub-scenes, which the paper does not report and which could quantify the claimed continuity benefit.
  • For end-to-end driving, NAVSIM has no sequential sensor frames, so DeMo-E2E++ drops cross-scene interaction; reorganizing sensor data or adding a recurrent memory could transfer that benefit to planning as well.
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

3 major / 4 minor

Summary. DeMo++ proposes a unified motion forecasting and planning framework that decouples trajectory decoding into mode queries (directional intentions) and state queries (spatiotemporal states), processed by a hybrid Attention-Mamba architecture. The paper adds two components: cross-scene intention interaction, enabled by a sliding-window reorganization of snapshot datasets into sub-scenes, and state-anchor-based refinement of proposal trajectories. The framework is extended to end-to-end planning (DeMo-E2E++) using raw sensor inputs. The authors report state-of-the-art results on Argoverse 2, nuScenes, nuPlan, and NAVSIM, and provide extensive ablations on the Argoverse 2 validation split, efficiency comparisons, qualitative results, and a failure analysis.

Significance. If the reported results were obtained under the official benchmark protocols, the decoupled query representation would be a meaningful alternative to the one-query-one-trajectory paradigm, and the hybrid Attention-Mamba combination would be a useful architectural contribution. The manuscript is transparent about its relationship to the prior DeMo and RealMotion works, and the ablation study is thorough, covering component contributions, sequence-modeling choices, state-query counts, and layer depths. However, the blanket state-of-the-art claim rests substantially on the Argoverse 2 and nuPlan numbers, which are obtained under a reorganized sub-scene protocol with shorter history and no matched baseline. As a result, the significance of those two benchmark claims is currently unestablished, and the abstract's sweeping SOTA statement is not supported for them.

major comments (3)
  1. [Section V-A-c and Tables I, III] The implementation details state that for DeMo++ the Argoverse 2 dataset is reorganized into three sub-scenes with 3 s of history and nuPlan into two sub-scenes with 1.5 s of history, while the future horizon is kept at 6 s and 8 s respectively. No baseline is reported under this same reorganized protocol, and the official Argoverse 2 benchmark provides 5 s of history. The reported gains over prior methods in Tables I and III therefore conflate the architectural contributions with the input history length, the number of training samples, and the evaluation distribution, so the abstract's state-of-the-art claim is not supported for these two benchmarks.
  2. [Section III-D-b and III-D-c] The cross-scene intention interaction uses historical mode and state queries from earlier sub-scenes produced by the sliding-window reorganization. Because the official Argoverse 2 and nuPlan evaluations are snapshot-based and do not provide such sequential cross-scene context, the test-split numbers in Tables I and III cannot be reproduced by a standard official submission, and the method's benefit on the official task remains undemonstrated.
  3. [Section V-A-c and Table I] The paper does not state whether the 'DeMo' results in Tables I and III are obtained under the original standard protocol or under the same reorganized sub-scene protocol as DeMo++. Since DeMo is the primary baseline for isolating the contribution of the new components (cross-scene intention interaction and state-anchor-based refinement), this ambiguity is critical; if DeMo uses 5 s history and DeMo++ uses 3 s, the improvements attributed to the new components are not controlled, and if DeMo also uses the reorganized protocol, the comparison with prior work is affected by the same issue.
minor comments (4)
  1. [Abstract and Section IV] The benchmark name appears as 'NA VSIM' (with a space) in the abstract and in Section IV, while 'NAVSIM' is used elsewhere; please unify the spelling.
  2. [Table V] The header of Table V appears corrupted ('State Decouple Agg. Aux. CII RefineQuery Query Module Loss'); please fix the column formatting so that each component name is aligned with its column.
  3. [Section V-D and Table XII] Table XII shows that the state-query output has a lower minFDE1 (3.84) than the final hybrid output (3.93); the text does not discuss this apparent trade-off, and a few sentences of explanation would help the reader interpret the decoupling analysis.
  4. [Section VI, Limitations] The Limitations section only discusses model efficiency; it should also acknowledge that the reorganized sub-scene protocol is a deviation from the official benchmarks and that matched baselines are needed for a fair comparison with prior work.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the mode/state decoupling, auxiliary losses, and refinement modules are architectural training choices validated on benchmark splits; the main caveat is an evaluation-protocol comparability issue, not a circular derivation.

full rationale

DeMo++ does not derive its final predictions from fitted parameters renamed as predictions. The decoder equations (2)-(4) specify how state and mode queries interact with scene context and with each other, and the auxiliary losses L_ts (Eq. 8) and L_m (Eq. 9) supervise intermediate representations against ground truth, which is a standard training regularizer rather than circular reasoning. The ablation studies (Tables V-XI) compare model variants on a validation split, and the final test-split numbers are external benchmark measurements with independently reported baselines. The self-citations to DeMo [22] and RealMotion [23] are prior published models used as predecessors and baselines; no uniqueness or theoretical claim is imported from those papers to force the current architecture. The one substantive concern is not circular: Section III-D-a reorganizes Argoverse 2 into three sub-scenes with 3 seconds of history and nuPlan into two sub-scenes with 1.5 seconds of history, while Tables I and III compare against methods evaluated on the standard 5-second and 2-second protocols. That is a comparability and external-validity risk that the Limitations section does not disclose, but no equation reduces to its own input and no fitted parameter is relabeled as a prediction. Hence the circularity score is 0.

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

No new physical entities are introduced. The free parameters listed are architectural and data-processing choices that the central SOTA claim depends on, especially the sub-scene history length and the sub-scene count. The most load-bearing axiom is that the reorganized sub-scene protocol remains equivalent to the official benchmark, which is assumed but not demonstrated.

free parameters (5)
  • Sub-scene count for data reorganization = 3 (Argoverse 2), 2 (nuPlan)
    Sliding window divides each scene into this many sub-scenes, tripling training samples for Argoverse 2. Not ablated.
  • History length per sub-scene = 3s (Argoverse 2), 1.5s (nuPlan)
    Shorter than the standard 5s history; changes the input to the final prediction and is load-bearing for the protocol comparison.
  • Number of state queries = 60
    Ablated over 10/20/30/60; 60 chosen, matching the 6s future at 10Hz.
  • State anchor refinement radius = 50m
    Distance mask threshold for refinement cross-attention; no ablation reported.
  • Number of planning modes in NAVSIM = 20
    Set for end-to-end planning; not ablated.
assumptions (5)
  • standard math Attention and Mamba operations as specified in Eqs. (1)-(4) provide sufficient sequence and scene modeling.
    Invoked throughout Section III.
  • domain assumption Vectorized scene representation with PointNet polyline encoder and agent-centric coordinates is an appropriate input encoding.
    Section III-A, following references [7], [13], [14].
  • ad hoc to paper Mode and state queries can be summed to form hybrid queries without loss of information.
    Section III-C-c states 'we simply add Qm and Qs together'; this is a design choice not derived from first principles.
  • ad hoc to paper Reorganized sub-scenes with shorter history preserve the official evaluation task for Argoverse 2 and nuPlan.
    Section III-D-a introduces the sliding window reorganization, but the paper does not prove equivalence to the standard benchmark.
  • domain assumption Bidirectional Mamba better models temporal consistency of state queries than unidirectional Mamba, attention, GRU, or Conv1d.
    Section V-C-b, Table VI, chosen based on empirical ablation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DeMo++: Motion Decoupling for Autonomous Driving." pith.science (2026). https://pith.science/paper/FSOE7BYP

@misc{pith2026250717342,
  author       = {Pith},
  title        = {Pith review of: DeMo++: Motion Decoupling for Autonomous Driving},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FSOE7BYP}},
  note         = {Machine review of arXiv:2507.17342}
}
read the original abstract

Motion forecasting and planning are tasked with estimating the trajectories of traffic agents and the ego vehicle, respectively, to ensure the safety and efficiency of autonomous driving systems in dynamically changing environments. State-of-the-art methods typically adopt a one-query-one-trajectory paradigm, where each query corresponds to a unique trajectory for predicting multi-mode trajectories. While this paradigm can produce diverse motion intentions, it often falls short in modeling the intricate spatiotemporal evolution of trajectories, which can lead to collisions or suboptimal outcomes. To overcome this limitation, we propose DeMo++, a framework that decouples motion estimation into two distinct components: holistic motion intentions to capture the diverse potential directions of movement, and fine spatiotemporal states to track the agent's dynamic progress within the scene and enable a self-refinement capability. Further, we introduce a cross-scene trajectory interaction mechanism to explore the relationships between motions in adjacent scenes. This allows DeMo++ to comprehensively model both the diversity of motion intentions and the spatiotemporal evolution of each trajectory. To effectively implement this framework, we developed a hybrid model combining Attention and Mamba. This architecture leverages the strengths of both mechanisms for efficient scene information aggregation and precise trajectory state sequence modeling. Extensive experiments demonstrate that DeMo++ achieves state-of-the-art performance across various benchmarks, including motion forecasting (Argoverse 2 and nuScenes), motion planning (nuPlan), and end-to-end planning (NAVSIM).

Figures

Figures reproduced from arXiv: 2507.17342 by the authors.

Figure 1
Figure 1. Conceptual illustration of future trajectory representation. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of our DeMo++ framework: The HD maps and agents are first processed by the encoder to obtain the scene context. The decoding pipeline includes: (a) the Mode Localization Module, which processes mode queries by interacting with the scene context from the encoder and among themselves; (b) the State Consistency Module, which processes state queries; and (c) the Hybrid Coupling Module, which combines these quer… view at source ↗
Figure 3
Figure 3. (a) Cross-scene intention interaction: the mode queries interact with historical mode queries using trajectory embeddings (TE); similarly, the state queries interact in the same manner. (b) State anchor-based refinement: the state queries within each trajectory interact with the scene context to refine both the predicted trajectories and their associated probabilities. over time. Motivated by this, we introduce inte… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Illustration of our data reorganization strategy: starting from (a) an independent scene, we (b) reorganize the trajectories into segments and aggregate surrounding elements, resulting in (c) continuous sub-scenes. E. State anchor-based refinement To fully leverage the…
Figure 5
Figure 5. Figure 5: Overview of our DeMo-E2E++ framework. Multi-view images and LiDAR data are first processed by a multi-modal sensor encoder to extract BEV features and agent features, which together constitute the scene context. For trajectory decoding, two types of queries—mode querie…
Figure 6
Figure 6. Figure 6: Qualitative results for the motion forecasting task [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Qualitative results for end-to-end planning on the NAVSIM dataset. The visualization includes three front-facing camera views: front-left, front, and front-right. The trajectory planned by DeMo-E2E++ is shown in orange, while the ground-truth trajectory is shown in gre…

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Mind the Noise: Sensitivity of Transformer-based Interaction-Aware Trajectory Prediction Models to Noisy Data

    cs.AI 2026-06 unverdicted novelty 4.0 of 10

    Noise in input state data degrades Transformer trajectory prediction accuracy by factors of 1.3x to 3.9x under realistic conditions.

Reference graph

Works this paper leans on

107 extracted references · 63 canonical work pages · cited by 1 Pith paper

  1. [1]

    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, 2022

  2. [2]

    Scalability in perception for autonomous driving: Waymo open dataset,

    P. Sun, H. Kretzschmar, X. Dotiwalla, A. Chouard, V . Patnaik, P. Tsui, J. Guo, Y . Zhou, Y . Chai, B. Caine et al. , “Scalability in perception for autonomous driving: Waymo open dataset,” in IEEE Conference on Computer Vision and Pattern Recognition , 2020

  3. [3]

    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,” in Advances in Neural Information Processing Systems , 2021

  4. [4]

    Carla: An open urban driving simulator,

    A. Dosovitskiy, G. Ros, F. Codevilla, A. Lopez, and V . Koltun, “Carla: An open urban driving simulator,” in Conference on Robot Learning , 2017

  5. [5]

    Navsim: Data-driven non-reactive autonomous vehicle simulation and benchmarking,

    D. Dauner, M. Hallgarten, T. Li, X. Weng, Z. Huang, Z. Yang, H. Li, I. Gilitschenski, B. Ivanovic, M. Pavone et al. , “Navsim: Data-driven non-reactive autonomous vehicle simulation and benchmarking,” in Advances in Neural Information Processing Systems , 2024

  6. [6]

    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

  7. [7]

    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 Conference on Computer Vision and Pattern Recognition, 2020

  8. [8]

    Learning lane graph representations for motion forecasting,

    M. Liang, B. Yang, R. Hu, Y . Chen, R. Liao, S. Feng, and R. Urta- sun, “Learning lane graph representations for motion forecasting,” in European Conference on Computer Vision , 2020

Show all 107 references
  1. [9]

    Scene transformer: A unified architecture for predicting future trajectories of multiple agents,

    J. Ngiam, V . Vasudevan, B. Caine, Z. Zhang, H.-T. L. Chiang, J. Ling, R. Roelofs, A. Bewley, C. Liu, A. Venugopal, D. J. Weiss, B. Sapp, Z. Chen, and J. Shlens, “Scene transformer: A unified architecture for predicting future trajectories of multiple agents,” in International...

  2. [10]

    Rethinking imitation-based planners for autonomous driving,

    J. Cheng, Y . Chen, X. Mei, B. Yang, B. Li, and M. Liu, “Rethinking imitation-based planners for autonomous driving,” in IEEE Interna- tional Conference on Robotics and Automation , 2024

  3. [11]

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

    J. Gu, C. Sun, and H. Zhao, “Densetnt: End-to-end trajectory prediction from dense goal sets,” in IEEE International Conference on Computer Vision, 2021

  4. [12]

    Eda: Evolving and distinct anchors for multimodal motion prediction,

    L. Lin, X. Lin, T. Lin, L. Huang, R. Xiong, and Y . Wang, “Eda: Evolving and distinct anchors for multimodal motion prediction,” in AAAI Conference on Artificial Intelligence , 2024

  5. [13]

    Motion transformer with global intention localization and local movement refinement,

    S. Shi, L. Jiang, D. Dai, and B. Schiele, “Motion transformer with global intention localization and local movement refinement,” in Ad- vances in Neural Information Processing Systems , 2022

  6. [14]

    Query-centric trajectory prediction,

    Z. Zhou, J. Wang, Y .-H. Li, and Y .-K. Huang, “Query-centric trajectory prediction,” in IEEE Conference on Computer Vision and Pattern Recognition, 2023

  7. [15]

    Diffusiondrive: Truncated diffusion model for end-to-end autonomous driving,

    B. Liao, S. Chen, H. Yin, B. Jiang, C. Wang, S. Yan, X. Zhang, X. Li, Y . Zhang, Q. Zhang, and X. Wang, “Diffusiondrive: Truncated diffusion model for end-to-end autonomous driving,” in IEEE Conference on Computer Vision and Pattern Recognition , 2025

  8. [16]

    Goalflow: Goal-driven flow matching for multimodal trajectories generation in end-to-end autonomous driving,

    Z. Xing, X. Zhang, Y . Hu, B. Jiang, T. He, Q. Zhang, X. Long, and W. Yin, “Goalflow: Goal-driven flow matching for multimodal trajectories generation in end-to-end autonomous driving,” in IEEE Conference on Computer Vision and Pattern Recognition , 2025

  9. [17]

    Diffusion-based planning for au- tonomous driving with flexible guidance,

    Y . Zheng, R. Liang, K. ZHENG, J. Zheng, L. Mao, J. Li, W. Gu, R. Ai, S. E. Li, X. Zhan, and J. Liu, “Diffusion-based planning for au- tonomous driving with flexible guidance,” in International Conference on Learning Representations , 2025

  10. [18]

    Reasoning multi-agent behavioral topology for interactive autonomous driving,

    H. Liu, L. Chen, Y . Qiao, C. Lv, and H. Li, “Reasoning multi-agent behavioral topology for interactive autonomous driving,” in Advances in Neural Information Processing Systems , 2024

  11. [19]

    End-to-end object detection with transformers,

    N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, and S. Zagoruyko, “End-to-end object detection with transformers,” in European Conference on Computer Vision , 2020

  12. [20]

    Dab-detr: Dynamic anchor boxes are better queries for detr,

    S. Liu, F. Li, H. Zhang, X. Yang, X. Qi, H. Su, J. Zhu, and L. Zhang, “Dab-detr: Dynamic anchor boxes are better queries for detr,” in International Conference on Learning Representations , 2022

  13. [21]

    Mamba: Linear-time sequence modeling with selective state spaces,

    A. Gu and T. Dao, “Mamba: Linear-time sequence modeling with selective state spaces,” arXiv preprint arXiv:2312.00752 , 2023

  14. [22]

    Demo: Decoupling motion fore- casting into directional intentions and dynamic states,

    B. Zhang, N. Song, and L. Zhang, “Demo: Decoupling motion fore- casting into directional intentions and dynamic states,” in Advances in Neural Information Processing Systems , 2024

  15. [23]

    Motion forecasting in continuous driving,

    N. Song, B. Zhang, X. Zhu, and L. Zhang, “Motion forecasting in continuous driving,” in Advances in Neural Information Processing Systems, 2024

  16. [24]

    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,” in Conference on Robot Learning , 2020

  17. [25]

    Home: Heatmap output for future motion estimation,

    T. Gilles, S. Sabatini, D. Tsishkou, B. Stanciulescu, and F. Moutarde, “Home: Heatmap output for future motion estimation,” in IEEE Inter- national Intelligent Transportation Systems Conference , 2021

  18. [26]

    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,” in IEEE Conference on Computer Vision and Pattern Recognition, 2020

  19. [27]

    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,” in IEEE International Conference on Robotics ...

  20. [28]

    Tnt: Target-driven trajectory prediction,

    H. Zhao, J. Gao, T. Lan, C. Sun, B. Sapp, B. Varadarajan, Y . Shen, Y . Shen, Y . Chai, C. Schmid et al. , “Tnt: Target-driven trajectory prediction,” in Conference on Robot Learning , 2021

  21. [29]

    Hierarchical vector transformer for multi-agent motion prediction,

    Z. Zhou, L. Ye, J. Wang, K. Wu, and K. H. Lu, “Hierarchical vector transformer for multi-agent motion prediction,” in IEEE Conference on Computer Vision and Pattern Recognition , 2022

  22. [30]

    Multimodal trajectory prediction conditioned on lane-graph traversals,

    N. Deo, E. Wolff, and O. Beijbom, “Multimodal trajectory prediction conditioned on lane-graph traversals,” in Conference on Robot Learn- ing, 2022

  23. [31]

    Gohome: Graph-oriented heatmap output for future motion estima- tion,

    T. Gilles, S. Sabatini, D. Tsishkou, B. Stanciulescu, and F. Moutarde, “Gohome: Graph-oriented heatmap output for future motion estima- tion,” in IEEE International Conference on Robotics and Automation , 2022

  24. [32]

    Multi-agent trajectory prediction by combining egocentric and allocentric views,

    X. Jia, L. Sun, H. Zhao, M. Tomizuka, and W. Zhan, “Multi-agent trajectory prediction by combining egocentric and allocentric views,” in Conference on Robot Learning , 2022

  25. [33]

    Hdgt: Heterogeneous driving graph transformer for multi-agent trajectory prediction via scene encoding,

    X. Jia, P. Wu, L. Chen, Y . Liu, H. Li, and J. Yan, “Hdgt: Heterogeneous driving graph transformer for multi-agent trajectory prediction via scene encoding,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023

  26. [34]

    Fjmp: Factor- ized joint multi-agent motion prediction over learned directed acyclic interaction graphs,

    L. Rowe, M. Ethier, E.-H. Dykhne, and K. Czarnecki, “Fjmp: Factor- ized joint multi-agent motion prediction over learned directed acyclic interaction graphs,” in IEEE Conference on Computer Vision and Pattern Recognition, 2023

  27. [35]

    Lanercnn: Distributed representations for graph-centric motion forecasting,

    W. Zeng, M. Liang, R. Liao, and R. Urtasun, “Lanercnn: Distributed representations for graph-centric motion forecasting,” in International Conference on Intelligent Robots and Systems , 2021

  28. [36]

    Trajectory prediction with graph-based dual-scale context fusion,

    L. Zhang, P. Li, J. Chen, and S. Shen, “Trajectory prediction with graph-based dual-scale context fusion,” in International Conference on Intelligent Robots and Systems , 2022

  29. [37]

    Multimodal motion prediction with stacked transformers,

    Y . Liu, J. Zhang, L. Fang, Q. Jiang, and B. Zhou, “Multimodal motion prediction with stacked transformers,” in IEEE Conference on Computer Vision and Pattern Recognition , 2021

  30. [38]

    Wayformer: Motion forecasting via simple & efficient attention networks,

    N. Nayakanti, R. Al-Rfou, A. Zhou, K. Goel, K. S. Refaat, and B. Sapp, “Wayformer: Motion forecasting via simple & efficient attention networks,” in IEEE International Conference on Robotics and Automation, 2023

  31. [39]

    Real-time motion prediction via heterogeneous polyline transformer with relative pose encoding,

    Z. Zhang, A. Liniger, C. Sakaridis, F. Yu, and L. V . Gool, “Real-time motion prediction via heterogeneous polyline transformer with relative pose encoding,” Advances in Neural Information Processing Systems , 2023

  32. [40]

    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,” Advances in Neural Information Processing Systems , 2017

  33. [41]

    Traj-mae: Masked autoencoders for trajectory prediction,

    H. Chen, J. Wang, K. Shao, F. Liu, J. Hao, C. Guan, G. Chen, and P.- A. Heng, “Traj-mae: Masked autoencoders for trajectory prediction,” in IEEE International Conference on Computer Vision , 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,” in IEEE International Conference on Computer Vision , 2023. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 14

  35. [43]

    Sept: Towards efficient scene representation learning for motion prediction,

    Z. Lan, Y . Jiang, Y . Mu, C. Chen, and S. E. Li, “Sept: Towards efficient scene representation learning for motion prediction,” in International Conference on Learning Representations , 2024

  36. [44]

    T4p: Test- time training of trajectory prediction via masked autoencoder and actor- specific token memory,

    D. Park, J. Jeong, S.-H. Yoon, J. Jeong, and K.-J. Yoon, “T4p: Test- time training of trajectory prediction via masked autoencoder and actor- specific token memory,” in IEEE Conference on Computer Vision and Pattern Recognition, 2024

  37. [45]

    Hpnet: Dynamic trajectory forecasting with historical prediction attention,

    X. Tang, M. Kan, S. Shan, Z. Ji, J. Bai, and X. Chen, “Hpnet: Dynamic trajectory forecasting with historical prediction attention,” in IEEE Conference on Computer Vision and Pattern Recognition , 2024

  38. [46]

    Trajeglish: Learning the lan- guage of driving scenarios,

    J. Philion, X. B. Peng, and S. Fidler, “Trajeglish: Learning the lan- guage of driving scenarios,” in International Conference on Learning Representations, 2024

  39. [47]

    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,” in IEEE International Conference on Computer Vision, 2023

  40. [48]

    R-pred: Two-stage motion prediction via tube-query attention-based trajectory refinement,

    S. Choi, J. Kim, J. Yun, and J. W. Choi, “R-pred: Two-stage motion prediction via tube-query attention-based trajectory refinement,” in IEEE International Conference on Computer Vision , 2023

  41. [49]

    Smartrefine: A scenario-adaptive refinement framework for efficient motion prediction,

    Y . Zhou, H. Shao, L. Wang, S. L. Waslander, H. Li, and Y . Liu, “Smartrefine: A scenario-adaptive refinement framework for efficient motion prediction,” in IEEE Conference on Computer Vision and Pattern Recognition, 2024

  42. [50]

    THOMAS: Trajectory heatmap output with learned multi-agent sam- pling,

    T. Gilles, S. Sabatini, D. Tsishkou, B. Stanciulescu, and F. Moutarde, “THOMAS: Trajectory heatmap output with learned multi-agent sam- pling,” in International Conference on Learning Representations, 2022

  43. [51]

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

    S. Shi, L. Jiang, D. Dai, and B. Schiele, “Mtr++: Multi-agent motion prediction with symmetric scene modeling and guided intention query- ing,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024

  44. [52]

    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

  45. [53]

    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. Ramananet al., “Argoverse: 3d tracking and forecasting with rich maps,” in IEEE Conference on Computer Vision and Pattern Recognition , 2019

  46. [54]

    nuscenes: A multimodal dataset for autonomous driving,

    H. Caesar, V . Bankiti, A. H. Lang, S. V ora, V . E. Liong, Q. Xu, A. Krishnan, Y . Pan, G. Baldan, and O. Beijbom, “nuscenes: A multimodal dataset for autonomous driving,” in IEEE Conference on Computer Vision and Pattern Recognition , 2020

  47. [55]

    Congested traffic states in empirical observations and microscopic simulations,

    M. Treiber, A. Hennecke, and D. Helbing, “Congested traffic states in empirical observations and microscopic simulations,” Physical review E, 2000

  48. [56]

    Parting with misconceptions about learning-based vehicle motion planning,

    D. Dauner, M. Hallgarten, A. Geiger, and K. Chitta, “Parting with misconceptions about learning-based vehicle motion planning,” in CoRL, 2023

  49. [57]

    Pluto: Pushing the limit of imita- tion learning-based planning for autonomous driving,

    J. Cheng, Y . Chen, and Q. Chen, “Pluto: Pushing the limit of imita- tion learning-based planning for autonomous driving,” arXiv preprint arXiv:2404.14327, 2024

  50. [58]

    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,” in IEEE Conference on Computer Vision and Pattern Recognition , 2023

  51. [59]

    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,” in IEEE International Conference on Computer Vision, 2023

  52. [60]

    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,” in IEEE International Conference on Robotics and Automation , 2025

  53. [61]

    Bridging past and future: End-to-end autonomous driving with historical prediction and plan- ning,

    B. Zhang, N. Song, X. Jin, and L. Zhang, “Bridging past and future: End-to-end autonomous driving with historical prediction and plan- ning,” in IEEE Conference on Computer Vision and Pattern Recogni- tion, 2025

  54. [62]

    Bevformer: Learning bird’s-eye-view representation from multi-camera images via spatiotemporal transformers,

    Z. Li, W. Wang, H. Li, E. Xie, C. Sima, T. Lu, Y . Qiao, and J. Dai, “Bevformer: Learning bird’s-eye-view representation from multi-camera images via spatiotemporal transformers,” in European Conference on Computer Vision , 2022

  55. [63]

    Exploring object- centric temporal modeling for efficient multi-view 3d object detection,

    S. Wang, Y . Liu, T. Wang, Y . Li, and X. Zhang, “Exploring object- centric temporal modeling for efficient multi-view 3d object detection,” in IEEE International Conference on Computer Vision , 2023

  56. [64]

    Trans- fuser: Imitation with transformer-based sensor fusion for autonomous driving,

    K. Chitta, A. Prakash, B. Jaeger, Z. Yu, K. Renz, and A. Geiger, “Trans- fuser: Imitation with transformer-based sensor fusion for autonomous driving,” IEEE Transactions on Pattern Analysis and Machine Intelli- gence, 2022

  57. [65]

    St-p3: End-to-end vision-based autonomous driving via spatial-temporal feature learning,

    S. Hu, L. Chen, P. Wu, H. Li, J. Yan, and D. Tao, “St-p3: End-to-end vision-based autonomous driving via spatial-temporal feature learning,” in European Conference on Computer Vision , 2022

  58. [66]

    Driveadapter: Breaking the coupling barrier of perception and planning in end-to-end autonomous driving,

    X. Jia, Y . Gao, L. Chen, J. Yan, P. L. Liu, and H. Li, “Driveadapter: Breaking the coupling barrier of perception and planning in end-to-end autonomous driving,” in IEEE International Conference on Computer Vision, 2023

  59. [67]

    Think twice before driving: Towards scalable decoders for end-to-end autonomous driving,

    X. Jia, P. Wu, L. Chen, J. Xie, C. He, J. Yan, and H. Li, “Think twice before driving: Towards scalable decoders for end-to-end autonomous driving,” in IEEE Conference on Computer Vision and Pattern Recog- nition, 2023

  60. [68]

    Vadv2: End-to-end vectorized autonomous driving via probabilistic planning,

    S. Chen, B. Jiang, H. Gao, B. Liao, Q. Xu, Q. Zhang, C. Huang, W. Liu, and X. Wang, “Vadv2: End-to-end vectorized autonomous driving via probabilistic planning,” arXiv preprint arXiv:2402.13243 , 2024

  61. [69]

    Enhancing end-to-end autonomous driving with latent world model,

    Y . Li, L. Fan, J. He, Y . Wang, Y . Chen, Z. Zhang, and T. Tan, “Enhancing end-to-end autonomous driving with latent world model,” in International Conference on Learning Representations , 2025

  62. [70]

    Navigation-guided sparse scene representation for end-to-end autonomous driving,

    P. Li and D. Cui, “Navigation-guided sparse scene representation for end-to-end autonomous driving,” in International Conference on Learning Representations, 2025

  63. [71]

    Bench2drive: Towards multi-ability benchmarking of closed-loop end-to-end autonomous driving,

    X. Jia, Z. Yang, Q. Li, Z. Zhang, and J. Yan, “Bench2drive: Towards multi-ability benchmarking of closed-loop end-to-end autonomous driving,” in Advances in Neural Information Processing Systems, 2024

  64. [72]

    Drivetransformer: Unified trans- former for scalable end-to-end autonomous driving,

    X. Jia, J. You, Z. Zhang, and J. Yan, “Drivetransformer: Unified trans- former for scalable end-to-end autonomous driving,” in International Conference on Learning Representations , 2025

  65. [73]

    Hungry hungry hippos: Towards language modeling with state space models,

    D. Y . Fu, T. Dao, K. K. Saab, A. W. Thomas, A. Rudra, and C. Re, “Hungry hungry hippos: Towards language modeling with state space models,” in International Conference on Learning Representations , 2023

  66. [74]

    Efficiently modeling long sequences with structured state spaces,

    A. Gu, K. Goel, and C. Re, “Efficiently modeling long sequences with structured state spaces,” in International Conference on Learning Representations, 2022

  67. [75]

    Simplified state space layers for sequence modeling,

    J. T. Smith, A. Warrington, and S. Linderman, “Simplified state space layers for sequence modeling,” in International Conference on Learning Representations, 2023

  68. [76]

    Densemamba: State space models with dense hidden connection for efficient large language models,

    W. He, K. Han, Y . Tang, C. Wang, Y . Yang, T. Guo, and Y . Wang, “Densemamba: State space models with dense hidden connection for efficient large language models,” arXiv preprint arXiv:2403.00818 , 2024

  69. [77]

    Jamba: A hybrid transformer-mamba language model,

    O. Lieber, B. Lenz, H. Bata, G. Cohen, J. Osin, I. Dalmedi- gos, E. Safahi, S. Meirom, Y . Belinkov, S. Shalev-Shwartz et al. , “Jamba: A hybrid transformer-mamba language model,” arXiv preprint arXiv:2403.19887, 2024

  70. [78]

    Zigma: A dit-style zigzag mamba diffusion model,

    V . T. Hu, S. A. Baumann, M. Gui, O. Grebenkova, P. Ma, J. Fischer, and B. Ommer, “Zigma: A dit-style zigzag mamba diffusion model,” in European Conference on Computer Vision , 2024

  71. [79]

    Videomamba: State space model for efficient video understanding,

    K. Li, X. Li, Y . Wang, Y . He, Y . Wang, L. Wang, and Y . Qiao, “Videomamba: State space model for efficient video understanding,” in European Conference on Computer Vision , 2024

  72. [80]

    Motion mamba: Efficient and long sequence motion generation with hierarchical and bidirectional selective ssm,

    Z. Zhang, A. Liu, I. Reid, R. Hartley, B. Zhuang, and H. Tang, “Motion mamba: Efficient and long sequence motion generation with hierarchical and bidirectional selective ssm,” in European Conference on Computer Vision , 2024

  73. [81]

    Vision mamba: Efficient visual representation learning with bidirectional state space model,

    L. Zhu, B. Liao, Q. Zhang, X. Wang, W. Liu, and X. Wang, “Vision mamba: Efficient visual representation learning with bidirectional state space model,” in International Conference on Machine Learning, 2024

  74. [82]

    End-to-end driving with online trajectory evaluation via bev world model,

    Y . Li, Y . Wang, Y . Liu, J. He, L. Fan, and Z. Zhang, “End-to-end driving with online trajectory evaluation via bev world model,” in IEEE International Conference on Computer Vision , 2025

  75. [83]

    Hydra-mdp: End-to-end multimodal planning with multi- target hydra-distillation,

    Z. Li, K. Li, S. Wang, S. Lan, Z. Yu, Y . Ji, Z. Li, Z. Zhu, J. Kautz, Z. Wu et al., “Hydra-mdp: End-to-end multimodal planning with multi- target hydra-distillation,” arXiv preprint arXiv:2406.06978 , 2024

  76. [84]

    Deformable detr: Deformable transformers for end-to-end object detection,

    X. Zhu, W. Su, L. Lu, B. Li, X. Wang, and J. Dai, “Deformable detr: Deformable transformers for end-to-end object detection,” in International Conference on Learning Representations , 2021

  77. [85]

    Leveraging future relationship reasoning for vehicle trajectory prediction,

    D. Park, H. Ryu, Y . Yang, J. Cho, J. Kim, and K.-J. Yoon, “Leveraging future relationship reasoning for vehicle trajectory prediction,” in International Conference on Learning Representations , 2023

  78. [86]

    Simpl: A simple and efficient multi-agent motion prediction baseline for autonomous driving,

    L. Zhang, P. Li, S. Liu, and S. Shen, “Simpl: A simple and efficient multi-agent motion prediction baseline for autonomous driving,” IEEE Robotics and Automation Letters , 2024

  79. [87]

    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,” in IEEE International Conference on Robotics and Automation , 2023. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 15

  80. [88]

    Ganet: Goal area network for motion forecast- ing,

    M. Wang, X. Zhu, C. Yu, W. Li, Y . Ma, R. Jin, X. Ren, D. Ren, M. Wang, and W. Yang, “Ganet: Goal area network for motion forecast- ing,” in IEEE International Conference on Robotics and Automation , 2023

  81. [89]

    Prophnet: Efficient agent- centric motion forecasting with anchor-informed proposals,

    X. Wang, T. Su, F. Da, and X. Yang, “Prophnet: Efficient agent- centric motion forecasting with anchor-informed proposals,” in IEEE Conference on Computer Vision and Pattern Recognition , 2023

  82. [90]

    Cadet: a causal dis- entanglement approach for robust trajectory prediction in autonomous driving,

    M. Pourkeshavarz, J. Zhang, and A. Rasouli, “Cadet: a causal dis- entanglement approach for robust trajectory prediction in autonomous driving,” in IEEE Conference on Computer Vision and Pattern Recog- nition, 2024

  83. [91]

    Trajec- tron++: Dynamically-feasible trajectory forecasting with heterogeneous data,

    T. Salzmann, B. Ivanovic, P. Chakravarty, and M. Pavone, “Trajec- tron++: Dynamically-feasible trajectory forecasting with heterogeneous data,” in European Conference on Computer Vision , 2020

  84. [92]

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

    B. Kim, S. H. Park, S. Lee, E. Khoshimjonov, D. Kum, J. Kim, J. S. Kim, and J. W. Choi, “Lapred: Lane-aware prediction of multi- modal future trajectories of dynamic agents,” in IEEE Conference on Computer Vision and Pattern Recognition , 2021

  85. [93]

    Trajectory forecasts in unknown environments conditioned on grid-based plans,

    N. Deo and M. M. Trivedi, “Trajectory forecasts in unknown environments conditioned on grid-based plans,” arXiv preprint arXiv:2001.00735, 2020

  86. [94]

    Context-aware scene prediction network (caspnet),

    M. Sch ¨afer, K. Zhao, M. B ¨uhren, and A. Kummert, “Context-aware scene prediction network (caspnet),” in IEEE International Intelligent Transportation Systems Conference, 2022

  87. [95]

    Latent variable sequential set transformers for joint multi-agent motion prediction,

    R. Girgis, F. Golemo, F. Codevilla, M. Weiss, J. A. D’Souza, S. E. Ka- hou, F. Heide, and C. Pal, “Latent variable sequential set transformers for joint multi-agent motion prediction,” in International Conference on Learning Representations , 2022

  88. [96]

    Laformer: Trajectory prediction for autonomous driving with lane-aware scene constraints,

    M. Liu, H. Cheng, L. Chen, H. Broszio, J. Li, R. Zhao, M. Sester, and M. Y . Yang, “Laformer: Trajectory prediction for autonomous driving with lane-aware scene constraints,” in IEEE Conference on Computer Vision and Pattern Recognition , 2024

  89. [97]

    Gameformer: Game-theoretic modeling and learning of transformer-based interactive prediction and planning for autonomous driving,

    Z. Huang, H. Liu, and C. Lv, “Gameformer: Game-theoretic modeling and learning of transformer-based interactive prediction and planning for autonomous driving,” in IEEE International Conference on Com- puter Vision, 2023

  90. [98]

    Urban driver: Learning to drive from real-world demonstrations using policy gradients,

    O. Scheel, L. Bergamini, M. Wolczyk, B. Osi ´nski, and P. Ondruska, “Urban driver: Learning to drive from real-world demonstrations using policy gradients,” in Conference on Robot Learning , 2022

  91. [99]

    Plant: Explainable planning transformers via object-level representations,

    K. Renz, K. Chitta, O.-B. Mercea, A. S. Koepke, Z. Akata, and A. Geiger, “Plant: Explainable planning transformers via object-level representations,” in Conference on Robot Learning , 2022

  92. [100]

    From prediction to planning with goal conditioned lane graph traversals,

    M. Hallgarten, M. Stoll, and A. Zell, “From prediction to planning with goal conditioned lane graph traversals,” in IEEE International Conference on Intelligent Transportation Systems , 2023

  93. [101]

    Decoupled weight decay regularization,

    I. Loshchilov and F. Hutter, “Decoupled weight decay regularization,” in International Conference on Learning Representations , 2019

  94. [102]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in IEEE Conference on Computer Vision and Pattern Recognition, 2016

  95. [103]

    Para-drive: Parallelized architecture for real-time autonomous driving,

    X. Weng, B. Ivanovic, Y . Wang, Y . Wang, and M. Pavone, “Para-drive: Parallelized architecture for real-time autonomous driving,” in IEEE Conference on Computer Vision and Pattern Recognition , 2024

  96. [104]

    Hydra-mdp++: Advancing end-to-end driving via expert- guided hydra-distillation,

    K. Li, Z. Li, S. Lan, Y . Xie, Z. Zhang, J. Liu, Z. Wu, Z. Yu, and J. M. Alvarez, “Hydra-mdp++: Advancing end-to-end driving via expert- guided hydra-distillation,” arXiv preprint arXiv:2503.12820 , 2025

  97. [105]

    Drama: An efficient end-to-end motion planner for autonomous driving with mamba,

    C. Yuan, Z. Zhang, J. Sun, S. Sun, Z. Huang, C. D. W. Lee, D. Li, Y . Han, A. Wong, K. P. Tee et al. , “Drama: An efficient end-to-end motion planner for autonomous driving with mamba,” arXiv preprint arXiv:2408.03601, 2024

  98. [106]

    Hydra-next: Robust closed-loop driving with open-loop training,

    Z. Li, S. Wang, S. Lan, Z. Yu, Z. Wu, and J. M. Alvarez, “Hydra-next: Robust closed-loop driving with open-loop training,” arXiv preprint arXiv:2503.12030, 2025

  99. [107]

    Empirical evaluation of gated recurrent neural networks on sequence modeling,

    J. Chung, C. Gulcehre, K. Cho, and Y . Bengio, “Empirical evaluation of gated recurrent neural networks on sequence modeling,” arXiv preprint arXiv:1412.3555, 2014. Bozhou Zhang received his BEng degree from Beihang University. He is now a PhD student in the School of Data Sci...

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.