pith. machine review for the scientific record. sign in

arxiv: 2605.02762 · v1 · submitted 2026-05-04 · 💻 cs.CV

Recognition: 2 theorem links

· Lean Theorem

Unified Map Prior Encoder for Mapping and Planning

Zongzheng Zhang , Sizhe Zou , Guantian Zheng , Zhenxin Zhu , Yu Gao , Guoxuan Chi , Shuo Wang , Yuwen Heng
show 7 more authors
Zhigang Sun Yiru Wang Hao Sun Chao Ma Zhen Li Anqing Jiang Hao Zhao
Authors on Pith no claims yet

Pith reviewed 2026-05-08 18:11 UTC · model grok-4.3

classification 💻 cs.CV
keywords map prior fusiononline mappingend-to-end planningbird's eye viewvector mapsraster mapsautonomous driving
0
0 comments X

The pith

A unified encoder aligns and fuses any mix of vector and raster map priors into BEV features to raise mapping accuracy and lower planning errors.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Autonomous driving systems usually treat map priors as optional because formats differ and sensor pose can drift. The paper shows that a single encoder can take any subset of HD vector maps, SD vector maps, raster SD maps, or satellite images, align them to the current view, and merge them with bird's-eye-view scene features. Vector lines receive SE(2) corrections and confidence-weighted attention while raster images use conditioned backbones and zero-init residual links. The result is higher map precision and safer planned paths even when some priors are missing at test time. A reader should care because the method turns existing map data into a reliable signal rather than a source of noise.

Core claim

The paper claims that a vector encoder that pre-aligns polylines, encodes points with multi-frequency sinusoids, and emits tokens with per-polyline scores, followed by a raster encoder that applies FiLM-conditioned ResNet stages and SE(2) micro-alignment, then fuses the two streams in vector-first order via cross-attention with bias and normalized channel gating, produces BEV features that improve both mapping and planning over sensor-only baselines.

What carries the argument

The Unified Map Prior Encoder with a vector branch (SE(2) pre-alignment, sinusoidal point encoding, confidence scores) and a raster branch (FiLM-conditioned ResNet-18, zero-initialized residual fusion), merged by cross-attention and gating.

If this is right

  • Mapping models reach higher mAP when any combination of priors is supplied at test time.
  • End-to-end planners produce trajectories with lower average L2 error and lower collision rates.
  • A model trained on the full set of priors still outperforms single-prior baselines when only one prior arrives at inference.
  • The same fusion pattern works across different backbone mapping and planning networks.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Similar alignment-plus-gating blocks could be added to other multi-sensor fusion pipelines that suffer from inconsistent data availability.
  • The method suggests that explicit geometric correction before semantic fusion is more important than simply increasing model capacity.
  • Testing the same architecture on datasets with seasonal map changes would reveal whether the confidence scores adapt to outdated priors.
  • Extending the raster branch to accept live aerial imagery could further reduce reliance on pre-built HD maps.

Load-bearing premise

The supplied map priors can be brought into the current vehicle frame by simple rigid transformations and their usefulness can be summarized by scalar scores that do not mislead the later fusion steps.

What would settle it

A test set containing large unmodeled pose drift where adding the priors through the encoder lowers mapping mAP or raises collision rate compared with the no-prior baseline would show the alignment and gating steps are not sufficient.

Figures

Figures reproduced from arXiv: 2605.02762 by Anqing Jiang, Chao Ma, Guantian Zheng, Guoxuan Chi, Hao Sun, Hao Zhao, Shuo Wang, Sizhe Zou, Yiru Wang, Yu Gao, Yuwen Heng, Zhen Li, Zhenxin Zhu, Zhigang Sun, Zongzheng Zhang.

Figure 1
Figure 1. Figure 1: Unified Map Prior Encoder (UMPE). UMPE ingests an arbitrary subset of four map priors—vector (HD/SD vectorized maps) and raster (rasterized SD map, satellite imagery), and processes them via a vector encoder and a raster encoder. The resulting priors are fused with BEV features, supporting both online HD mapping and end-to-end planning tasks. and are merged into a single BEV representation shared by mappin… view at source ↗
Figure 2
Figure 2. Figure 2: Unified Map Prior Encoder (UMPE) architecture. (a) Vector Encoder: HD/SD polylines are SE(2) pre-aligned and encoded; BEV queries attend to each source with confidence-biased dual cross-attention. Presence-normalized, channel-wise gating mixes sources to produce fused vector tokens Y¯ . (b) Raster Encoder: rasterized SD map and satellite imagery pass through a shared FiLM-conditioned ResNet, then undergo S… view at source ↗
Figure 3
Figure 3. Figure 3: Online mapping visualization on nuScenes. Adding UMPE to both MapTRv2 [9] and MapQR [10] produces more accurate maps, especially in the green-highlighted regions: baselines show broken pedestrian crossings, kinked boundaries and missing dividers; UMPE straightens, restores them. Following standard protocols for vectorized mapping, we report mAP computed from average precision over Chamfer distance threshol… view at source ↗
Figure 4
Figure 4. Figure 4: End-to-end Planning visualization on nuScenes. The ego vehicle is turning left. VAD without priors drifts toward the oncoming lane; adding the vector encoder or raster encoder improves lane adherence but leaves lateral error, while VAD+UMPE produces a trajectory tightly overleaps the GT. TABLE III MAPPING RESULTS ON ARGOVERSE 2 [12] VALIDATION DATASET. Method APped APdiv APbou mAP VectorMapNet [15] 35.6 34… view at source ↗
read the original abstract

Online mapping and end-to-end (E2E) planning in autonomous driving remain largely sensor-centric, leaving rich map priors, including HD/SD vector maps, rasterized SD maps, and satellite imagery, underused because of heterogeneity, pose drift, and inconsistent availability at test time. We present UMPE, a Unified Map Prior Encoder that can ingest any subset of four priors and fuse them with BEV features for both mapping and planning. UMPE has two branches. The vector encoder pre-aligns HD/SD polylines with a frame-wise SE(2) correction, encodes points via multi-frequency sinusoidal features, and produces polyline tokens with confidence scores. BEV queries then apply cross-attention with confidence bias, followed by normalized channel-wise gating to avoid length imbalance and softly down-weight uncertain sources. The raster encoder shares a ResNet-18 backbone conditioned by FiLM with scaling and shift at every stage, performs SE(2) micro-alignment, and injects priors through zero-initialized residual fusion, so the network starts from a do-no-harm baseline and learns to add only useful prior evidence. A vector-then-raster fusion order reflects the inductive bias of geometry first, appearance second. On nuScenes mapping, UMPE lifts MapTRv2 from 61.5 to 67.4 mAP (+5.9) and MapQR from 66.4 to 71.7 mAP (+5.3). On Argoverse2, UMPE adds +4.1 mAP over strong baselines. UMPE is compositional: when trained with all priors, it outperforms single-prior models even when only one prior is available at test time, demonstrating powerset robustness. For E2E planning with the VAD backbone on nuScenes, UMPE reduces trajectory error from 0.72 to 0.42 m L2 on average (-0.30 m) and collision rate from 0.22% to 0.12% (-0.10%), surpassing recent prior-injection methods. These results show that a unified, alignment-aware treatment of heterogeneous map priors yields better mapping and better planning.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript presents UMPE, a Unified Map Prior Encoder that ingests heterogeneous map priors (HD/SD vector maps, rasterized maps, satellite imagery) and fuses them with BEV features for online mapping and end-to-end planning. The vector encoder applies frame-wise SE(2) micro-alignment to polylines, uses multi-frequency sinusoidal point encoding, and produces tokens with confidence scores that bias cross-attention and enable normalized gating. The raster encoder employs a FiLM-conditioned ResNet-18 with SE(2) alignment and zero-initialized residual fusion. Vector-then-raster fusion order is used. On nuScenes, UMPE improves MapTRv2 mAP from 61.5 to 67.4 and MapQR from 66.4 to 71.7; with VAD planner it reduces average L2 trajectory error from 0.72 m to 0.42 m and collision rate from 0.22% to 0.12%. Comparable gains (+4.1 mAP) are reported on Argoverse2, with compositional robustness when subsets of priors are available at test time.

Significance. If the empirical gains prove robust, the work is significant for providing a practical, alignment-aware mechanism to exploit underused map priors in autonomous driving despite pose drift and inconsistent availability. The do-no-harm residual design, powerset robustness, and joint mapping-planning improvements represent useful engineering contributions that could be adopted in BEV pipelines.

major comments (2)
  1. [Vector encoder description (abstract and method)] The reported gains (e.g., +5.9 mAP on MapTRv2, -0.30 m L2 with VAD) rest on the vector encoder's SE(2) micro-alignment and per-polyline confidence scores being sufficiently accurate to modulate cross-attention and gating without amplifying pose errors. No independent diagnostics (alignment error histograms, calibration plots, or drift-robustness tests) are provided to verify this load-bearing assumption under realistic conditions.
  2. [Experiments and results] Ablation studies isolating the SE(2) correction, confidence gating, raster residual fusion, and vector-first order are absent, as is error analysis on cases where priors are noisy or misaligned. This makes it difficult to attribute the central performance deltas specifically to the proposed components rather than other implementation details.
minor comments (2)
  1. [Fusion module] The normalized channel-wise gating operation would benefit from an explicit equation to clarify how length imbalance is avoided and uncertain sources are down-weighted.
  2. [Figures and tables] Figure captions and axis labels in the mapping and planning result tables could be expanded for standalone readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive review and positive assessment of UMPE's practical contributions. We address each major comment below and will revise the manuscript accordingly to provide the requested diagnostics and ablations.

read point-by-point responses

  1. Referee: [Vector encoder description (abstract and method)] The reported gains (e.g., +5.9 mAP on MapTRv2, -0.30 m L2 with VAD) rest on the vector encoder's SE(2) micro-alignment and per-polyline confidence scores being sufficiently accurate to modulate cross-attention and gating without amplifying pose errors. No independent diagnostics (alignment error histograms, calibration plots, or drift-robustness tests) are provided to verify this load-bearing assumption under realistic conditions.

    Authors: We acknowledge that the manuscript does not include independent diagnostics such as alignment error histograms, confidence calibration plots, or explicit drift-robustness tests for the SE(2) micro-alignment and per-polyline confidence scores. The reported gains and the powerset robustness results (outperforming single-prior models even with partial priors at test time) provide indirect evidence that the mechanisms do not amplify pose errors under the evaluated conditions. To directly address the concern, we will add these diagnostics in the revision, including histograms of SE(2) correction magnitudes on nuScenes and Argoverse2, calibration curves for confidence scores, and tests with simulated pose drift. revision: yes

  2. Referee: [Experiments and results] Ablation studies isolating the SE(2) correction, confidence gating, raster residual fusion, and vector-first order are absent, as is error analysis on cases where priors are noisy or misaligned. This makes it difficult to attribute the central performance deltas specifically to the proposed components rather than other implementation details.

    Authors: We agree that the current experiments do not isolate the individual contributions of SE(2) correction, confidence gating, raster residual fusion, and vector-first fusion order, nor do they include dedicated error analysis on noisy or misaligned priors. While the manuscript reports overall gains, subset robustness, and comparisons to prior-injection baselines, these do not fully disentangle the components. We will add targeted ablations for each element and case studies analyzing performance on subsets with injected noise or misalignment in the revised experiments section. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical architecture with independent experimental validation

full rationale

The paper presents UMPE as an engineering architecture (vector encoder with SE(2) pre-alignment and confidence-gated cross-attention; raster encoder with FiLM-conditioned ResNet and zero-init residual fusion; vector-then-raster order) whose performance claims rest entirely on held-out nuScenes/Argoverse2 metrics. No equation or result is shown to reduce by construction to a fitted parameter, self-citation, or renamed input; the design choices are presented as inductive biases rather than derived predictions. The central results (mAP deltas, L2/collision reductions) are falsifiable experimental outcomes, not tautological re-statements of the method.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The method rests on standard computer-vision backbones and attention mechanisms; no new physical axioms or invented entities are introduced. The only free parameters are the learned weights of the encoders and the gating layers, which are fitted on the training split.

free parameters (2)
  • SE(2) correction parameters
    Learned per-frame pose correction for vector polylines; fitted during training.
  • FiLM scaling and shift parameters
    Conditioning parameters for the raster ResNet; fitted end-to-end.
axioms (2)
  • domain assumption BEV features can be meaningfully augmented by cross-attention to aligned map tokens
    Invoked in the vector branch description.
  • domain assumption Zero-initialized residual fusion starts from a do-no-harm baseline
    Stated in the raster encoder paragraph.

pith-pipeline@v0.9.0 · 5741 in / 1637 out tokens · 41932 ms · 2026-05-08T18:11:10.792751+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

52 extracted references · 15 canonical work pages · 3 internal anchors

  1. [1]

    Augmenting lane perception and topol- ogy understanding with standard definition navigation maps,

    K. Z. Luo, X. Weng, Y . Wang, S. Wu, J. Li, K. Q. Weinberger, Y . Wang, and M. Pavone, “Augmenting lane perception and topol- ogy understanding with standard definition navigation maps,”ICRA, pp. 4029–4035, 2023

    2023

  2. [2]

    Enhancing online road network perception and reasoning with standard definition maps,

    H. Zhang, D. Paz, Y . Guo, A. Das, X. Huang, K. Haug, H. I. Christensen, and L. Ren, “Enhancing online road network perception and reasoning with standard definition maps,” inIROS, 2024

    2024

  3. [3]

    Sdtagnet: Leveraging text-annotated navigation maps for online hd map construction,

    F. Immel, J.-H. Pauls, R. Fehler, F. Bieder, J. Merkert, and C. Stiller, “Sdtagnet: Leveraging text-annotated navigation maps for online hd map construction,”arXiv preprint arXiv:2506.08997, 2025

    work page arXiv 2025

  4. [4]

    Comple- menting onboard sensors with satellite maps: a new perspective for hd map construction,

    W. Gao, J. Fu, Y . Shen, H. Jing, S. Chen, and N. Zheng, “Comple- menting onboard sensors with satellite maps: a new perspective for hd map construction,” inICRA, pp. 11103–11109, IEEE, 2024

    2024

  5. [5]

    Smart: Advancing scalable map priors for driving topology reasoning,

    J. Ye, D. Paz, H. Zhang, Y . Guo, X. Huang, H. I. Christensen, Y . Wang, and L. Ren, “Smart: Advancing scalable map priors for driving topology reasoning,”arXiv preprint arXiv:2502.04329, 2025

    work page arXiv 2025

  6. [6]

    Sept: Standard-definition map enhanced scene perception and topology reasoning for autonomous driving,

    M. Pei, J. Shan, P. Li, J. Shi, J. Huo, Y . Gao, and S. Shen, “Sept: Standard-definition map enhanced scene perception and topology reasoning for autonomous driving,”RAL, 2025

    2025

  7. [7]

    Leveraging sd map to augment hd map-based trajectory prediction,

    Z. Dong, R. Ding, W. Li, P. Zhang, G. Tang, and J. Guo, “Leveraging sd map to augment hd map-based trajectory prediction,” inCVPR, pp. 17219–17228, 2025

    2025

  8. [8]

    arXiv preprint arXiv:2409.05352 , year=

    S. Zeng, X. Chang, X. Liu, Z. Pan, and X. Wei, “Driving with prior maps: Unified vector prior encoding for autonomous vehicle mapping,” arXiv preprint arXiv:2409.05352, 2024

    work page arXiv 2024

  9. [9]

    Maptrv2: An end-to-end framework for online vectorized hd map construction,

    B. Liao, S. Chen, Y . Zhang, B. Jiang, Q. Zhang, W. Liu, C. Huang, and X. Wang, “Maptrv2: An end-to-end framework for online vectorized hd map construction,”IJCV, vol. 133, no. 3, pp. 1352–1374, 2025

    2025

  10. [10]

    Leveraging enhanced queries of point sets for vectorized map construction,

    Z. Liu, X. Zhang, G. Liu, J. Zhao, and N. Xu, “Leveraging enhanced queries of point sets for vectorized map construction,” inEuropean Conference on Computer Vision, pp. 461–477, Springer, 2024

    2024

  11. [11]

    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,” inCVPR, 2020

    2020

  12. [12]

    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,”arXiv preprint arXiv:2301.00493, 2023

    work page internal anchor Pith review arXiv 2023

  13. [13]

    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,” inCVPR, pp. 8340–8350, 2023

    2023

  14. [14]

    Hdmapnet: An online hd map construction and evaluation framework,

    Q. Li, Y . Wang, Y . Wang, and H. Zhao, “Hdmapnet: An online hd map construction and evaluation framework,” in2022 International Conference on Robotics and Automation (ICRA), IEEE, 2022

    2022

  15. [15]

    Vectormapnet: End-to-end vectorized hd map learning,

    Y . Liu, T. Yuan, Y . Wang, Y . Wang, and H. Zhao, “Vectormapnet: End-to-end vectorized hd map learning,” inInternational Conference on Machine Learning, pp. 22352–22369, PMLR, 2023

    2023

  16. [16]

    Maptr: Structured modeling and learning for online vectorized hd map construction,

    B. Liao, S. Chen, X. Wang, T. Cheng, Q. Zhang, W. Liu, and C. Huang, “Maptr: Structured modeling and learning for online vectorized hd map construction,”arXiv preprint arXiv:2208.14437, 2022

    work page arXiv 2022

  17. [17]

    Streammapnet: Streaming mapping network for vectorized online hd map construc- tion,

    T. Yuan, Y . Liu, Y . Wang, Y . Wang, and H. Zhao, “Streammapnet: Streaming mapping network for vectorized online hd map construc- tion,” inCVPR, pp. 7356–7365, 2024

    2024

  18. [18]

    Maptracker: Track- ing with strided memory fusion for consistent vector hd mapping,

    J. Chen, Y . Wu, J. Tan, H. Ma, and Y . Furukawa, “Maptracker: Track- ing with strided memory fusion for consistent vector hd mapping,” in ECCV, pp. 90–107, Springer, 2024

    2024

  19. [19]

    Himap: Hybrid representation learning for end-to-end vectorized hd map construction,

    Y . Zhou, H. Zhang, J. Yu, Y . Yang, S. Jung, S.-I. Park, and B. Yoo, “Himap: Hybrid representation learning for end-to-end vectorized hd map construction,” inCVPR, pp. 15396–15406, 2024

    2024

  20. [20]

    Chameleon: Fast-slow neuro-symbolic lane topology extraction,

    Z. Zhang, X. Li, S. Zou, G. Chi, S. Li, X. Qiu, G. Wang, G. Zheng, L. Wang, H. Zhao,et al., “Chameleon: Fast-slow neuro-symbolic lane topology extraction,” in2025 IEEE International Conference on Robotics and Automation (ICRA), pp. 3752–3758, IEEE, 2025

    2025

  21. [21]

    Reusing attention for one-stage lane topology understanding,

    Y . Li, Z. Zhang, X. Qiu, X. Li, Z. Liu, L. Wang, R. Li, Z. Zhu, H.-a. Gao, X. Lin,et al., “Reusing attention for one-stage lane topology understanding,” in2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), IEEE, 2025

    2025

  22. [22]

    Openlane-v2: A topology reasoning bench- mark for unified 3d hd mapping,

    H. Wang, T. Li, Y . Li, L. Chen, C. Sima, Z. Liu, B. Wang, P. Jia, Y . Wang, S. Jiang,et al., “Openlane-v2: A topology reasoning bench- mark for unified 3d hd mapping,”NeurIPS, pp. 18873–18884, 2023

    2023

  23. [23]

    Graph-based Topology Reasoning for Driving Scenes,

    T. Li, L. Chen, H. Wang, Y . Li, J. Yang, X. Geng, S. Jiang, Y . Wang, H. Xu, C. Xu,et al., “Graph-based topology reasoning for driving scenes,”arXiv preprint arXiv:2304.05277, 2023

    work page arXiv 2023

  24. [24]

    TopoMLP: A Simple yet Strong Pipeline for Driving Topology Reasoning,

    D. Wu, J. Chang, F. Jia, Y . Liu, T. Wang, and J. Shen, “Topomlp: A simple yet strong pipeline for driving topology reasoning,”arXiv preprint arXiv:2310.06753, 2023

    work page arXiv 2023

  25. [25]

    LaneSegNet: Map Learning with Lane Segment Perception for Autonomous Driving,

    T. Li, P. Jia, B. Wang, L. Chen, K. Jiang, J. Yan, and H. Li, “Lane- segnet: Map learning with lane segment perception for autonomous driving,”arXiv preprint arXiv:2312.16108, 2023

    work page arXiv 2023

  26. [26]

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

    S. Shi, L. Jiang, D. Dai, and B. Schiele, “Mtr++: Multi-agent mo- tion prediction with symmetric scene modeling and guided intention querying,”TPAMI, vol. 46, no. 5, pp. 3955–3971, 2024

    2024

  27. [27]

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

    J. Gu, C. Sun, and H. Zhao, “Densetnt: End-to-end trajectory predic- tion from dense goal sets,” inCVPR, pp. 15303–15312, 2021

    2021

  28. [28]

    Hivt: Hierarchical vector transformer for multi-agent motion prediction,

    Z. Zhou, L. Ye, J. Wang, K. Wu, and K. Lu, “Hivt: Hierarchical vector transformer for multi-agent motion prediction,” inCVPR, 2022

    2022

  29. [29]

    Query-centric trajectory prediction,

    Z. Zhou, J. Wang, Y .-H. Li, and Y .-K. Huang, “Query-centric trajectory prediction,” inCVPR, pp. 17863–17873, 2023

    2023

  30. [30]

    Accelerating online mapping and behavior prediction via direct bev feature attention,

    X. Gu, G. Song, I. Gilitschenski, M. Pavone, and B. Ivanovic, “Accelerating online mapping and behavior prediction via direct bev feature attention,” inECCV, pp. 412–428, Springer, 2024

    2024

  31. [31]

    Pro- ducing and leveraging online map uncertainty in trajectory prediction,

    X. Gu, G. Song, I. Gilitschenski, M. Pavone, and B. Ivanovic, “Pro- ducing and leveraging online map uncertainty in trajectory prediction,” inCVPR, pp. 14521–14530, 2024

    2024

  32. [32]

    Presight: Enhancing autonomous vehicle perception with city-scale nerf priors,

    T. Yuan, Y . Mao, J. Yang, Y . Liu, Y . Wang, and H. Zhao, “Presight: Enhancing autonomous vehicle perception with city-scale nerf priors,” inECCV, pp. 323–339, Springer, 2024

    2024

  33. [33]

    Neural map prior for autonomous driving,

    X. Xiong, Y . Liu, T. Yuan, Y . Wang, Y . Wang, and H. Zhao, “Neural map prior for autonomous driving,” inCVPR, pp. 17535–17544, 2023

    2023

  34. [34]

    P-mapnet: Far-seeing map generator enhanced by both sdmap and hdmap priors,

    Z. Jiang, Z. Zhu, P. Li, H.-a. Gao, T. Yuan, Y . Shi, H. Zhao, and H. Zhao, “P-mapnet: Far-seeing map generator enhanced by both sdmap and hdmap priors,”RAL, 2024

    2024

  35. [35]

    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 CVPR, pp. 17853–17862, 2023

    2023

  36. [36]

    Is ego status all you need for open-loop end-to-end autonomous driving?,

    Z. Li, Z. Yu, S. Lan, J. Li, J. Kautz, T. Lu, and J. M. Alvarez, “Is ego status all you need for open-loop end-to-end autonomous driving?,” inCVPR, pp. 14864–14873, 2024

    2024

  37. [37]

    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:2406.06978, 2024

    work page internal anchor Pith review arXiv 2024

  38. [38]

    Delving into mapping uncertainty for mapless trajectory prediction,

    Z. Zhang, X. Qiu, B. Zhang, G. Zheng, X. Gu, G. Chi, H.-a. Gao, L. Wang, Z. Liu, X. Li,et al., “Delving into mapping uncertainty for mapless trajectory prediction,” in2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), IEEE, 2025

    2025

  39. [39]

    Uniuncer: Unified dynamic static uncertainty for end to end driving,

    Y . Gao, J. Wang, Z. Zhang, A. Jiang, Y . Wang, Y . Heng, S. Wang, H. Sun, Z. Hu, and H. Zhao, “Uniuncer: Unified dynamic static uncertainty for end to end driving,”arXiv preprint arXiv:2603.07686, 2026

    work page internal anchor Pith review arXiv 2026

  40. [40]

    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,et al., “Diffusiondrive: Truncated diffusion model for end-to-end autonomous driving,” inCVPR, 2025

    2025

  41. [41]

    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,” inCVPR, pp. 1602–1611, 2025

    2025

  42. [42]

    Genad: Generative end-to-end autonomous driving,

    W. Zheng, R. Song, X. Guo, C. Zhang, and L. Chen, “Genad: Generative end-to-end autonomous driving,” inECCV, Springer, 2024

    2024

  43. [43]

    Sparsedrive: End-to-end autonomous driving via sparse scene representation.arXiv preprint arXiv:2405.19620, 2024

    W. Sun, X. Lin, Y . Shi, C. Zhang, H. Wu, and S. Zheng, “Sparsedrive: End-to-end autonomous driving via sparse scene representation,”arXiv preprint arXiv:2405.19620, 2024

    work page arXiv 2024

  44. [44]

    Epona: Autoregressive diffusion world model for autonomous driving

    K. Zhang, Z. Tang, X. Hu, X. Pan, X. Guo, Y . Liu, J. Huang, L. Yuan, Q. Zhang, X.-X. Long,et al., “Epona: Autoregressive diffusion world model for autonomous driving,”arXiv:2506.24113, 2025

    work page arXiv 2025

  45. [45]

    Don’t shake the wheel: Momentum-aware planning in end-to-end autonomous driving,

    Z. Song, C. Jia, L. Liu, H. Pan, Y . Zhang, J. Wang, X. Zhang, S. Xu, L. Yang, and Y . Luo, “Don’t shake the wheel: Momentum-aware planning in end-to-end autonomous driving,” inCVPR, 2025

    2025

  46. [46]

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

    B. Zhang, N. Song, X. Jin, and L. Zhang, “Bridging past and future: End-to-end autonomous driving with historical prediction and planning,” inCVPR, pp. 6854–6863, 2025

    2025

  47. [47]

    Orion: A holistic end-to-end autonomous driving framework by vision-language instructed action generation.arXiv preprint arXiv:2503.19755,

    H. Fu, D. Zhang, Z. Zhao, J. Cui, D. Liang, C. Zhang, D. Zhang, H. Xie, B. Wang, and X. Bai, “Orion: A holistic end-to-end au- tonomous driving framework by vision-language instructed action generation,”arXiv preprint arXiv:2503.19755, 2025

    work page arXiv 2025

  48. [48]

    Gaussianfusion: Gaussian-based multi-sensor fusion for end-to-end autonomous driv- ing,

    S. Liu, Q. Liang, Z. Li, B. Li, and K. Huang, “Gaussianfusion: Gaussian-based multi-sensor fusion for end-to-end autonomous driv- ing,”arXiv preprint arXiv:2506.00034, 2025

    work page arXiv 2025

  49. [49]

    Openstreetmap: User-generated street maps,

    M. Haklay and P. Weber, “Openstreetmap: User-generated street maps,”IEEE Pervasive computing, vol. 7, no. 4, pp. 12–18, 2008

    2008

  50. [50]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inCVPR, pp. 770–778, 2016

    2016

  51. [51]

    Mind the map! accounting for existing map information when estimating online hdmaps from sensor,

    R. Sun, L. Yang, D. Lingrand, and F. Precioso, “Mind the map! accounting for existing map information when estimating online hdmaps from sensor,”arXiv preprint arXiv:2311.10517, 2023

    work page arXiv 2023

  52. [52]

    Mgmap: Mask-guided learning for online vectorized hd map construction,

    X. Liu, S. Wang, W. Li, R. Yang, J. Chen, and J. Zhu, “Mgmap: Mask-guided learning for online vectorized hd map construction,” in CVPR, pp. 14812–14821, 2024

    2024