pith. sign in

arxiv: 2512.20563 · v2 · submitted 2025-12-23 · 💻 cs.CV · cs.AI· cs.LG· cs.RO

LEAD: Minimizing Learner-Expert Asymmetry in End-to-End Driving

Long Nguyen , Micha Fauth , Bernhard Jaeger , Daniel Dauner , Maximilian Igl , Andreas Geiger , Kashyap Chitta This is my paper

Pith reviewed 2026-05-16 19:59 UTC · model grok-4.3

classification 💻 cs.CV cs.AIcs.LGcs.RO
keywords imitation learningend-to-end drivinglearner-expert asymmetryCARLA simulatorautonomous vehiclesclosed-loop evaluationTransFusersim-to-real transfer
0
0 comments X

The pith

Narrowing the gaps in visibility, uncertainty, and route information between expert demonstrations and sensor-based student policies allows imitation learning to reach new state-of-the-art closed-loop performance in CARLA driving simulators

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

The paper examines why imitation learning policies trained on simulator data fail to perform robustly in closed-loop driving. It identifies key asymmetries: experts have perfect visibility ignoring occlusions and know other vehicles' actions, while students use limited sensors and receive only a single target point for navigation. The authors propose and test practical interventions to reduce these differences. After implementing these changes, their updated TransFuser v6 model sets new records on major CARLA benchmarks. This suggests that aligning expert and student information is crucial for effective sim-to-real transfer in autonomous driving.

Core claim

The central claim is that misalignment between privileged expert demonstrations and sensor-based student observations limits imitation learning in simulation, and that targeted modifications to narrow gaps in visibility, uncertainty, and navigational intent enable a student policy to achieve new state-of-the-art results on CARLA closed-loop benchmarks, with 95 DS on Bench2Drive and more than doubling prior performances on Longest6 v2 and Town13.

What carries the argument

The interventions to minimize learner-expert asymmetry, which adjust expert observations to match student limitations and enhance student inputs for better intent specification

Load-bearing premise

The observed performance improvements stem mainly from the proposed reductions in learner-expert asymmetry rather than other unmentioned changes to the model or training process

What would settle it

Running an ablation study that applies the asymmetry reductions one at a time while holding all other factors constant and measuring the incremental gains on the same benchmarks

Figures

Figures reproduced from arXiv: 2512.20563 by Andreas Geiger, Bernhard Jaeger, Daniel Dauner, Kashyap Chitta, Long Nguyen, Maximilian Igl, Micha Fauth.

Figure 1
Figure 1. Figure 1: Performing a task well and teaching it well are not the same. An expert driver (blue bounding box) is most useful when its behavior can be transferred to a student policy (green bounding box) effectively. Current expert drivers for CARLA do not fulfill this requirement. We focus on three common asymmetries that hinder effective transfer. Visibility asymmetry: the expert reacts to occluded actors, leading t… view at source ↗
Figure 2
Figure 2. Figure 2: summarizes how state and intent alignment con￾tribute to infraction counts. While infractions in general decrease with each improvement, the weakened target point bias, achieved through intent alignment, leads to an increase in route deviation, since the model no longer aggressively snaps back toward the target points after getting off route. Late Goal Conditioning as a Bottleneck: Although the GRU was ori… view at source ↗
read the original abstract

Simulators can generate virtually unlimited driving data, yet imitation learning policies in simulation still struggle to achieve robust closed-loop performance. Motivated by this gap, we empirically study how misalignment between privileged expert demonstrations and sensor-based student observations can limit the effectiveness of imitation learning. More precisely, experts have significantly higher visibility (e.g., ignoring occlusions) and far lower uncertainty (e.g., knowing other vehicles' actions), making them difficult to imitate reliably. Furthermore, navigational intent (i.e., the route to follow) is under-specified in student models at test time via only a single target point. We demonstrate that these asymmetries can measurably limit driving performance in CARLA and offer practical interventions to address them. After careful modifications to narrow the gaps between expert and student, our TransFuser v6 (TFv6) student policy achieves a new state of the art on all major publicly available CARLA closed-loop benchmarks, reaching 95 DS on Bench2Drive and more than doubling prior performances on Longest6~v2 and Town13. Additionally, by integrating perception supervision from our dataset into a shared sim-to-real pipeline, we show consistent gains on the NAVSIM and Waymo Vision-Based End-to-End driving benchmarks. Our code, data, and models are publicly available at https://github.com/autonomousvision/lead.

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 empirically studies expert-learner asymmetries in imitation learning for end-to-end driving, focusing on visibility (occlusion handling), uncertainty (privileged knowledge of other agents), and route specification (single target point vs. richer intent). The authors apply targeted interventions to TransFuser, yielding TFv6, which reports new state-of-the-art closed-loop results on CARLA benchmarks (95 DS on Bench2Drive; more than doubling prior scores on Longest6 v2 and Town13) plus gains on NAVSIM and Waymo via sim-to-real transfer. Code, data, and models are released publicly.

Significance. If the performance deltas are attributable to the asymmetry reductions, the work offers concrete, practical guidance for closing the expert-student gap in simulation-based driving policies and establishes stronger baselines for CARLA evaluation. Public release of code and models supports reproducibility and extension.

major comments (2)
  1. [§5] §5 (Experiments): The manuscript does not present controlled ablations that hold model capacity, dataset size, optimizer, and augmentation fixed while toggling only the asymmetry components (e.g., single vs. richer route input or occlusion-aware vs. privileged labels). Without an otherwise identical v5 baseline, concurrent unstated changes remain a plausible alternative explanation for the closed-loop gains.
  2. [Table 2] Table 2 and Table 3: Driving scores on Longest6 v2 and Town13 lack reported standard deviations or results across multiple random seeds. Given the stochasticity of closed-loop CARLA evaluation, this makes it difficult to assess whether the reported >2× improvements are statistically robust.
minor comments (2)
  1. The abstract and introduction refer to 'careful modifications' without a concise upfront enumeration of the exact changes; adding a short bullet list would improve readability.
  2. [Figure 3] Figure 3 (route input visualization): The distinction between single-point and richer route representations could be clarified with an explicit side-by-side comparison in the caption.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive recommendation of minor revision and the constructive comments on experimental controls and statistical reporting. We address each major comment below and will update the manuscript accordingly.

read point-by-point responses

  1. Referee: [§5] The manuscript does not present controlled ablations that hold model capacity, dataset size, optimizer, and augmentation fixed while toggling only the asymmetry components (e.g., single vs. richer route input or occlusion-aware vs. privileged labels). Without an otherwise identical v5 baseline, concurrent unstated changes remain a plausible alternative explanation for the closed-loop gains.

    Authors: We acknowledge the value of strictly controlled ablations to isolate the effect of each asymmetry reduction. The current manuscript presents incremental results from TransFuser v5 to v6 with targeted changes for visibility, uncertainty, and route specification. To strengthen attribution, we will add a new controlled ablation table in the revision that starts from an identical v5 configuration (fixed capacity, dataset, optimizer, and augmentations) and toggles only the asymmetry interventions one at a time. This will directly address the concern about alternative explanations. revision: yes

  2. Referee: [Table 2] Table 2 and Table 3: Driving scores on Longest6 v2 and Town13 lack reported standard deviations or results across multiple random seeds. Given the stochasticity of closed-loop CARLA evaluation, this makes it difficult to assess whether the reported >2× improvements are statistically robust.

    Authors: We agree that standard deviations and multi-seed results are important for assessing robustness in stochastic closed-loop settings. In the revised manuscript, we will rerun the evaluations on Longest6 v2 and Town13 across three random seeds and report mean driving scores with standard deviations. This will provide clearer evidence of the statistical reliability of the reported gains. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical interventions validated on external benchmarks

full rationale

The manuscript presents an empirical study of expert-student asymmetries in imitation learning for CARLA driving, followed by practical modifications to TransFuser yielding TFv6 and new benchmark results (95 DS on Bench2Drive, doubled scores on Longest6 v2 and Town13). No equations, derivations, or predictions are defined that reduce to inputs by construction. Claims rest on described interventions and comparisons against public benchmarks rather than self-citations, fitted parameters renamed as predictions, or ansatzes smuggled via prior work. The derivation chain is self-contained against external evaluation.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim depends on the empirical validity of the three identified asymmetries and the assumption that the listed modifications directly close the performance gap without introducing new confounding factors.

axioms (1)
  • domain assumption Imitation learning performance is limited by observation mismatch between expert and student rather than by other factors such as model capacity or optimization.
    Invoked in the motivation and intervention design sections.

pith-pipeline@v0.9.0 · 5563 in / 1258 out tokens · 17275 ms · 2026-05-16T19:59:15.358079+00:00 · methodology

discussion (0)

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

Forward citations

Cited by 4 Pith papers

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

  1. MDrive: Benchmarking Closed-Loop Cooperative Driving for End-to-End Multi-agent Systems

    cs.RO 2026-05 unverdicted novelty 7.0

    MDrive benchmark shows multi-agent cooperative driving systems generally outperform single-agent ones in closed-loop settings but perception sharing does not always improve planning and negotiation can harm performanc...

  2. Beyond Imitation: Learning Safe End-to-End Autonomous Driving from Hard Negatives

    cs.RO 2026-05 unverdicted novelty 6.0

    BeyondDrive augments imitation learning with synthesized safety-critical negative trajectories and a repulsive loss to improve safety in autonomous driving, reporting 89.7 PDMS on NAVSIMv1 and generalization to other models.

  3. BridgeSim: Unveiling the OL-CL Gap in End-to-End Autonomous Driving

    cs.RO 2026-04 unverdicted novelty 6.0

    The primary OL-CL gap in end-to-end autonomous driving arises from objective mismatch creating structural inability to model reactive behaviors, which a test-time adaptation method can mitigate.

  4. Do Open-Loop Metrics Predict Closed-Loop Driving? A Cross-Benchmark Correlation Study of NAVSIM and Bench2Drive

    cs.RO 2026-04 conditional novelty 4.0

    Cross-benchmark analysis of 8 methods shows NAVSIM PDM Score correlates with Bench2Drive Driving Score at Spearman ρ=0.90, with Ego Progress as the strongest single predictor and a simpler 3-metric formula matching th...

Reference graph

Works this paper leans on

64 extracted references · 64 canonical work pages · cited by 4 Pith papers · 2 internal anchors

  1. [1]

    Bench2drive leaderboard.URL: https: //github.com/autonomousvision/Bench2Drive-Leaderboard,

    autonomousvision. Bench2drive leaderboard.URL: https: //github.com/autonomousvision/Bench2Drive-Leaderboard,

    work page

  2. [2]

    Pdm-lite: A rule-based planner for carla leaderboard 2.0.URL: https://github.com/OpenDriveLab/ DriveLM/blob/DriveLM-CARLA/pdm lite/docs/report.pdf,

    Jens Beißwenger. Pdm-lite: A rule-based planner for carla leaderboard 2.0.URL: https://github.com/OpenDriveLab/ DriveLM/blob/DriveLM-CARLA/pdm lite/docs/report.pdf,

    work page

  3. [3]

    End to End Learning for Self-Driving Cars

    Mariusz Bojarski, Davide Del Testa, Daniel Dworakowski, Bernhard Firner, Beat Flepp, Prasoon Goyal, Lawrence D. Jackel, Mathew Monfort, Urs Muller, Jiakai Zhang, Xin Zhang, Jake Zhao, and Karol Zieba. End to end learning for self-driving cars.arXiv.org, 1604.07316, 2016. 2

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  4. [4]

    Pseudo-simulation for autonomous driving.Proc

    Wei Cao, Marcel Hallgarten, Tianyu Li, Daniel Dauner, Xunjiang Gu, Caojun Wang, Yakov Miron, Marco Aiello, Hongyang Li, Igor Gilitschenski, et al. Pseudo-simulation for autonomous driving.Proc. Conf. on Robot Learning (CoRL), 2025. 3, 6, 7, 8

    work page 2025

  5. [5]

    Learning from all vehi- cles

    Dian Chen and Philipp Kr ¨ahenb¨uhl. Learning from all vehi- cles. InProc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2022. 2

    work page 2022

  6. [6]

    Learning by cheating

    Dian Chen, Brady Zhou, Vladlen Koltun, and Philipp Kr¨ahenb¨uhl. Learning by cheating. InProc. Conf. on Robot Learning (CoRL), 2019. 2, 3

    work page 2019

  7. [7]

    Learn- ing to drive from a world on rails

    Dian Chen, Vladlen Koltun, and Philipp Kr ¨ahenb¨uhl. Learn- ing to drive from a world on rails. InProc. of the IEEE International Conf. on Computer Vision (ICCV), 2021. 2

    work page 2021

  8. [8]

    End-to-end autonomous driving: Challenges and frontiers.Transactions on Pattern Analysis and Machine Intelligence (T-PAMI), 2024

    Li Chen, Penghao Wu, Kashyap Chitta, Bernhard Jaeger, An- dreas Geiger, and Hongyang Li. End-to-end autonomous driving: Challenges and frontiers.Transactions on Pattern Analysis and Machine Intelligence (T-PAMI), 2024. 2

    work page 2024

  9. [9]

    Neat: Neural attention fields for end-to-end autonomous driving

    Kashyap Chitta, Aditya Prakash, and Andreas Geiger. Neat: Neural attention fields for end-to-end autonomous driving. InProc. of the IEEE International Conf. on Computer Vision (ICCV), 2021. 2

    work page 2021

  10. [10]

    Transfuser: Imitation with transformer-based sensor fusion for autonomous driv- ing.Transactions on Pattern Analysis and Machine Intelli- gence (T-PAMI), 2023

    Kashyap Chitta, Aditya Prakash, Bernhard Jaeger, Zehao Yu, Katrin Renz, and Andreas Geiger. Transfuser: Imitation with transformer-based sensor fusion for autonomous driv- ing.Transactions on Pattern Analysis and Machine Intelli- gence (T-PAMI), 2023. 2, 4, 6, 7, 8

    work page 2023

  11. [11]

    Sledge: Synthesizing driving environments with generative models and rule-based traffic

    Kashyap Chitta, Daniel Dauner, and Andreas Geiger. Sledge: Synthesizing driving environments with generative models and rule-based traffic. InProc. of the European Conf. on Computer Vision (ECCV), 2024. 6

    work page 2024

  12. [12]

    Empirical evaluation of gated recurrent neu- ral networks on sequence modeling, 2014

    Junyoung Chung, Caglar Gulcehre, KyungHyun Cho, and Yoshua Bengio. Empirical evaluation of gated recurrent neu- ral networks on sequence modeling, 2014. 4

    work page 2014

  13. [13]

    Lopez, Vladlen Koltun, and Alexey Dosovitskiy

    Felipe Codevilla, Antonio M. Lopez, Vladlen Koltun, and Alexey Dosovitskiy. On offline evaluation of vision-based driving models. InProc. of the European Conf. on Computer Vision (ECCV), 2018. 2, 3

    work page 2018

  14. [14]

    End-to-end driving via conditional imitation learning

    Felipe Codevilla, Matthias Miiller, Antonio L ´opez, Vladlen Koltun, and Alexey Dosovitskiy. End-to-end driving via conditional imitation learning. InProc. IEEE International Conf. on Robotics and Automation (ICRA), 2018. 2, 3

    work page 2018

  15. [15]

    L ´opez, and Adrien Gaidon

    Felipe Codevilla, Eder Santana, Antonio M. L ´opez, and Adrien Gaidon. Exploring the limitations of behavior cloning for autonomous driving. InProc. of the IEEE In- ternational Conf. on Computer Vision (ICCV), 2019. 2

    work page 2019

  16. [16]

    Robust autonomy emerges from self-play

    Marco Cusumano-Towner, David Hafner, Alex Hertzberg, Brody Huval, Aleksei Petrenko, Eugene Vinitsky, Erik Wi- jmans, Taylor Killian, Stuart Bowers, Ozan Sener, Philipp Kr¨ahenb¨uhl, and Vladlen Koltun. Robust autonomy emerges from self-play. InProc. of the International Conf. on Ma- chine learning (ICML), 2025. 3

    work page 2025

  17. [17]

    Parting with misconceptions about learning- based vehicle motion planning

    Daniel Dauner, Marcel Hallgarten, Andreas Geiger, and Kashyap Chitta. Parting with misconceptions about learning- based vehicle motion planning. InProc. Conf. on Robot Learning (CoRL), 2023. 3

    work page 2023

  18. [18]

    Navsim: Data-driven non-reactive autonomous vehicle simulation and benchmarking.Advances in Neural Information Processing Systems (NeurIPS), 2024

    Daniel Dauner, Marcel Hallgarten, Tianyu Li, Xinshuo Weng, Zhiyu Huang, Zetong Yang, Hongyang Li, Igor Gilitschenski, Boris Ivanovic, Marco Pavone, et al. Navsim: Data-driven non-reactive autonomous vehicle simulation and benchmarking.Advances in Neural Information Processing Systems (NeurIPS), 2024. 2, 3, 6, 7, 8

    work page 2024

  19. [19]

    CARLA: An open urban driving simulator

    Alexey Dosovitskiy, German Ros, Felipe Codevilla, Antonio Lopez, and Vladlen Koltun. CARLA: An open urban driving simulator. InProc. Conf. on Robot Learning (CoRL), 2017. 2, 3

    work page 2017

  20. [20]

    Plant 2.0: Exposing biases and structural flaws in closed-loop driv- ing.arXiv.org, 2025

    Simon Gerstenecker, Andreas Geiger, and Katrin Renz. Plant 2.0: Exposing biases and structural flaws in closed-loop driv- ing.arXiv.org, 2025. 4

    work page 2025

  21. [21]

    Co-Reyes, Rishabh Agarwal, Re- becca Roelofs, Yao Lu, Nico Montali, Paul Mougin, Zoey Yang, Brandyn White, Aleksandra Faust, Rowan McAllister, Dragomir Anguelov, and Benjamin Sapp

    Cole Gulino, Justin Fu, Wenjie Luo, George Tucker, Eli Bronstein, Yiren Lu, Jean Harb, Xinlei Pan, Yan Wang, Xiangyu Chen, John D. Co-Reyes, Rishabh Agarwal, Re- becca Roelofs, Yao Lu, Nico Montali, Paul Mougin, Zoey Yang, Brandyn White, Aleksandra Faust, Rowan McAllister, Dragomir Anguelov, and Benjamin Sapp. Waymax: An ac- celerated, data-driven simulat...

    work page 2023

  22. [22]

    Planning-oriented autonomous driving

    Yihan Hu, Jiazhi Yang, Li Chen, Keyu Li, Chonghao Sima, Xizhou Zhu, Siqi Chai, Senyao Du, Tianwei Lin, et al. Planning-oriented autonomous driving. InProc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2023. 3, 6

    work page 2023

  23. [23]

    Expert drivers for autonomous driving

    Bernhard Jaeger. Expert drivers for autonomous driving. Master’s thesis, University of T¨ubingen, 2021. 2, 3

    work page 2021

  24. [24]

    Transfuser versions.URL: https://github

    Bernhard Jaeger. Transfuser versions.URL: https://github. com / autonomousvision / carlagarage / blob / leaderboard2 / docs/history.md, 2024. 3

    work page 2024

  25. [25]

    Hid- den biases of end-to-end driving models

    Bernhard Jaeger, Kashyap Chitta, and Andreas Geiger. Hid- den biases of end-to-end driving models. InProc. of the IEEE International Conf. on Computer Vision (ICCV), 2023. 2, 3, 4

    work page 2023

  26. [26]

    Carl: Learning scalable planning policies with simple rewards

    Bernhard Jaeger, Daniel Dauner, Jens Beißwenger, Simon Gerstenecker, Kashyap Chitta, and Andreas Geiger. Carl: Learning scalable planning policies with simple rewards. Proc. Conf. on Robot Learning (CoRL), 2025. 3

    work page 2025

  27. [27]

    Bench2drive: Towards multi-ability benchmark- ing of closed-loop end-to-end autonomous driving

    Xiaosong Jia, Zhenjie Yang, Qifeng Li, Zhiyuan Zhang, and Junchi Yan. Bench2drive: Towards multi-ability benchmark- ing of closed-loop end-to-end autonomous driving. InAd- vances in Neural Information Processing Systems (NeurIPS),

    work page

  28. [28]

    Drivetransformer: Unified transformer for scalable end-to- end autonomous driving

    Xiaosong Jia, Junqi You, Zhiyuan Zhang, and Junchi Yan. Drivetransformer: Unified transformer for scalable end-to- end autonomous driving. InProc. of the International Conf. on Learning Representations (ICLR). OpenReview.net,

    work page

  29. [29]

    Ivanovic, and Marco Pavone

    Peter Karkus, Maximilian Igl, Yuxiao Chen, Kashyap Chitta, Jef Packer, Bertrand Douillard, Ran Tian, Alexan- der Naumann, Guillermo Garcia-Cobo, Shuhan Tan, Alperen De˘girmenci, Alexander Popov, Nikolai Smolyanskiy, Urs Muller, B. Ivanovic, and Marco Pavone. Beyond behav- ior cloning in autonomous driving: a survey of closed-loop training techniques, 2025. 8

    work page 2025

  30. [30]

    Towards learning- based planning: The nuplan benchmark for real-world au- tonomous driving

    Napat Karnchanachari, Dimitris Geromichalos, Kok Seang Tan, Nanxiang Li, Christopher Eriksen, Shakiba Yaghoubi, Noushin Mehdipour, Gianmarco Bernasconi, Whye Kit Fong, Yiluan Guo, and Holger Caesar. Towards learning- based planning: The nuplan benchmark for real-world au- tonomous driving. InProc. IEEE International Conf. on Robotics and Automation (ICRA)...

    work page 2024

  31. [31]

    Gpudrive: Data- driven, multi-agent driving simulation at 1 million FPS

    Saman Kazemkhani, Aarav Pandya, Daphne Cornelisse, Brennan Shacklett, and Eugene Vinitsky. Gpudrive: Data- driven, multi-agent driving simulation at 1 million FPS. In Proc. of the International Conf. on Learning Representations (ICLR), 2025. 6

    work page 2025

  32. [32]

    3d gaussian splatting for real-time radiance field rendering.ACM Transactions on Graphics,

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimk ¨uhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering.ACM Transactions on Graphics,

    work page

  33. [33]

    Metadrive: Composing diverse driving scenarios for generalizable reinforcement learning.IEEE Trans

    Quanyi Li, Zhenghao Peng, Lan Feng, Qihang Zhang, Zhenghai Xue, and Bolei Zhou. Metadrive: Composing diverse driving scenarios for generalizable reinforcement learning.IEEE Trans. on Pattern Analysis and Machine In- telligence (PAMI), 2022. 3

    work page 2022

  34. [34]

    Think2drive: Efficient reinforcement learning by thinking with latent world model for autonomous driving (in CARLA- V2)

    Qifeng Li, Xiaosong Jia, Shaobo Wang, and Junchi Yan. Think2drive: Efficient reinforcement learning by thinking with latent world model for autonomous driving (in CARLA- V2). InProc. of the European Conf. on Computer Vision (ECCV), 2024. 2, 3

    work page 2024

  35. [35]

    Mtgs: Multi-traversal gaussian splatting.arXiv preprint arXiv:2503.12552, 2025

    Tianyu Li, Yihang Qiu, Zhenhua Wu, Carl Lindstr ¨om, Peng Su, Matthias Nießner, and Hongyang Li. MTGS: Multi- traversal gaussian splatting.arXiv.org, 2503.12552, 2025. 7

    work page arXiv 2025

  36. [36]

    ´Alvarez

    Zhiqi Li, Zhiding Yu, Shiyi Lan, Jiahan Li, Jan Kautz, Tong Lu, and Jos´e M. ´Alvarez. Is ego status all you need for open- loop end-to-end autonomous driving? InProc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2024. 3

    work page 2024

  37. [37]

    Diffusiondrive: Trun- cated diffusion model for end-to-end autonomous driving

    Bencheng Liao, Shaoyu Chen, Haoran Yin, Bo Jiang, Cheng Wang, Sixu Yan, Xinbang Zhang, Xiangyu Li, Ying Zhang, Qian Zhang, and Xinggang Wang. Diffusiondrive: Trun- cated diffusion model for end-to-end autonomous driving. In Proc. IEEE Conf. on Computer Vision and Pattern Recogni- tion (CVPR), 2025. 2

    work page 2025

  38. [38]

    Gaussianfusion: Gaussian-based multi-sensor fusion for end-to-end autonomous driving

    Shuai Liu, Quanmin Liang, Zefeng Li, Boyang Li, and Kai Huang. Gaussianfusion: Gaussian-based multi-sensor fusion for end-to-end autonomous driving. InAdvances in Neural Information Processing Systems (NeurIPS), 2025. 2

    work page 2025

  39. [39]

    Neuroncap: Photorealistic closed- loop safety testing for autonomous driving

    William Ljungbergh, Adam Tonderski, Joakim Johnan- der, Holger Caesar, Kalle ˚Astr¨om, Michael Felsberg, and Christoffer Petersson. Neuroncap: Photorealistic closed- loop safety testing for autonomous driving. InProc. of the European Conf. on Computer Vision (ECCV), 2024. 3, 6

    work page 2024

  40. [40]

    Student-informed teacher training, 2025

    Nico Messikommer, Jiaxu Xing, Elie Aljalbout, and Davide Scaramuzza. Student-informed teacher training, 2025. 3

    work page 2025

  41. [41]

    Diamos, Erich Elsen, David Garc´ıa, Boris Ginsburg, Michael Houston, Oleksii Kuchaiev, Ganesh Venkatesh, and Hao Wu

    Paulius Micikevicius, Sharan Narang, Jonah Alben, Gre- gory F. Diamos, Erich Elsen, David Garc´ıa, Boris Ginsburg, Michael Houston, Oleksii Kuchaiev, Ganesh Venkatesh, and Hao Wu. Mixed precision training. InProc. of the Interna- tional Conf. on Learning Representations (ICLR), 2018. 6

    work page 2018

  42. [42]

    ALVINN: an autonomous land vehicle in a neural network

    Dean Pomerleau. ALVINN: an autonomous land vehicle in a neural network. InAdvances in Neural Information Pro- cessing Systems (NIPS), 1988. 2

    work page 1988

  43. [43]

    Multi- modal fusion transformer for end-to-end autonomous driv- ing

    Aditya Prakash, Kashyap Chitta, and Andreas Geiger. Multi- modal fusion transformer for end-to-end autonomous driv- ing. InProc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2021. 2, 4

    work page 2021

  44. [44]

    Designing network design spaces, 2020

    Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, and Piotr Doll ´ar. Designing network design spaces, 2020. 6

    work page 2020

  45. [45]

    Simlingo: Vision-only closed-loop autonomous driving with language-action alignment

    Katrin Renz, Long Chen, Elahe Arani, and Oleg Sinavski. Simlingo: Vision-only closed-loop autonomous driving with language-action alignment. InProc. IEEE Conf. on Com- puter Vision and Pattern Recognition (CVPR), 2025. 2, 4, 6, 7

    work page 2025

  46. [46]

    GAIA-2: A Controllable Multi-View Generative World Model for Autonomous Driving

    Lloyd Russell, Anthony Hu, Lorenzo Bertoni, George Fe- doseev, Jamie Shotton, Elahe Arani, and Gianluca Corrado. GAIA-2: A controllable multi-view generative world model for autonomous driving.arXiv.org, 2503.20523, 2025. 6

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  47. [47]

    Airsim: High-fidelity visual and physical simula- tion for autonomous vehicles

    Shital Shah, Debadeepta Dey, Chris Lovett, and Ashish Kapoor. Airsim: High-fidelity visual and physical simula- tion for autonomous vehicles. InField and service robotics: Results of the 11th international conference, pages 621–635. Springer, 2017. 3

    work page 2017

  48. [48]

    Safety-enhanced autonomous driving using inter- pretable sensor fusion transformer

    Hao Shao, Letian Wang, RuoBing Chen, Hongsheng Li, and Yu Liu. Safety-enhanced autonomous driving using inter- pretable sensor fusion transformer. InProc. Conf. on Robot Learning (CoRL), 2022. 2

    work page 2022

  49. [49]

    Waslan- der, Hongsheng Li, and Yu Liu

    Hao Shao, Letian Wang, Ruobing Chen, Steven L. Waslan- der, Hongsheng Li, and Yu Liu. Reasonnet: End-to-end driv- ing with temporal and global reasoning. InProc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2023. 2

    work page 2023

  50. [50]

    Pdm-lite dataset for carla leaderboard 2.0.URL: https://huggingface.co/datasets/ autonomousvision/PDM Lite Carla LB2, 2024

    Chonghao Sima, Katrin Renz, Kashyap Chitta, Li Chen, Hanxue Zhang, Chengen Xie, Jens Beißwenger, Ping Luo, Andreas Geiger, and Hongyang Li. Pdm-lite dataset for carla leaderboard 2.0.URL: https://huggingface.co/datasets/ autonomousvision/PDM Lite Carla LB2, 2024. 2

    work page 2024

  51. [51]

    Drivelm: Driving with graph visual question answering

    Chonghao Sima, Katrin Renz, Kashyap Chitta, Li Chen, Hanxue Zhang, Chengen Xie, Jens Beißwenger, Ping Luo, Andreas Geiger, and Hongyang Li. Drivelm: Driving with graph visual question answering. InProc. of the European Conf. on Computer Vision (ECCV), 2024. 2, 3, 4, 6, 7

    work page 2024

  52. [52]

    Hip-ad: Hierarchical and multi-granularity planning 10 with deformable attention for autonomous driving in a single decoder

    Yingqi Tang, Zhuoran Xu, Zhaotie Meng, and Erkang Cheng. Hip-ad: Hierarchical and multi-granularity planning 10 with deformable attention for autonomous driving in a single decoder. InProc. of the IEEE International Conf. on Com- puter Vision (ICCV), 2025. 2, 6, 7

    work page 2025

  53. [53]

    Impossibly good experts and how to follow them

    Aaron Walsman, Muru Zhang, Sanjiban Choudhury, Dieter Fox, and Ali Farhadi. Impossibly good experts and how to follow them. InProc. of the International Conf. on Learning Representations (ICLR), 2023. 3

    work page 2023

  54. [54]

    Wilder Lavington, Adam ´Scibior, Mark Schmidt, and Frank Wood

    Andrew Warrington, J. Wilder Lavington, Adam ´Scibior, Mark Schmidt, and Frank Wood. Robust asymmetric learn- ing in pomdps, 2021. 3

    work page 2021

  55. [55]

    Bridging the imitation gap by adaptive insubor- dination, 2021

    Luca Weihs, Unnat Jain, Iou-Jen Liu, Jordi Salvador, Svetlana Lazebnik, Aniruddha Kembhavi, and Alexander Schwing. Bridging the imitation gap by adaptive insubor- dination, 2021. 3

    work page 2021

  56. [56]

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

    Xinshuo Weng, Boris Ivanovic, Yan Wang, Yue Wang, and Marco Pavone. Para-drive: Parallelized architecture for real- time autonomous driving. InProc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2024. 3

    work page 2024

  57. [57]

    Trajectory-guided control prediction for end-to-end autonomous driving: A simple yet strong base- line

    Penghao Wu, Xiaosong Jia, Li Chen, Junchi Yan, Hongyang Li, and Yu Qiao. Trajectory-guided control prediction for end-to-end autonomous driving: A simple yet strong base- line. InAdvances in Neural Information Processing Systems (NeurIPS), 2022. 2

    work page 2022

  58. [58]

    Torcs: The open racing car simulator, 2015

    Bernhard Wymann, Christos Dimitrakakisy, Andrew Sum- nery, Eric Espi ´e, and Christophe Guionneauz. Torcs: The open racing car simulator, 2015. 3

    work page 2015

  59. [59]

    Wod-e2e: Waymo open dataset for end-to-end driving in challenging long-tail scenarios

    Runsheng Xu, Hubert Lin, Wonseok Jeon, Hao Feng, Yu- liang Zou, Liting Sun, John Gorman, Kate Tolstaya, Sarah Tang, Brandyn White, et al. Wod-e2e: Waymo open dataset for end-to-end driving in challenging long-tail scenarios. arXiv.org, 2025. 2, 3, 6, 7

    work page 2025

  60. [60]

    Unisim: A neural closed-loop sensor simulator

    Ze Yang, Yun Chen, Jingkang Wang, Sivabalan Mani- vasagam, Wei-Chiu Ma, Anqi Joyce Yang, and Raquel Ur- tasun. Unisim: A neural closed-loop sensor simulator. In Proc. IEEE Conf. on Computer Vision and Pattern Recogni- tion (CVPR), 2023. 3

    work page 2023

  61. [61]

    Rethinking the open-loop evaluation of end-to- end autonomous driving in nuscenes.arXiv.org, 2023

    Jiang-Tian Zhai, Ze Feng, Jihao Du, Yongqiang Mao, Jiang- Jiang Liu, Zichang Tan, Yifu Zhang, Xiaoqing Ye, and Jing- dong Wang. Rethinking the open-loop evaluation of end-to- end autonomous driving in nuscenes.arXiv.org, 2023. 3

    work page 2023

  62. [62]

    End-to-end urban driving by imitating a reinforcement learning coach

    Zhejun Zhang, Alexander Liniger, Dengxin Dai, Fisher Yu, and Luc Van Gool. End-to-end urban driving by imitating a reinforcement learning coach. InProc. of the IEEE Interna- tional Conf. on Computer Vision (ICCV), 2021. 3

    work page 2021

  63. [63]

    Hugsim: A real-time, photo-realistic and closed-loop simulator for autonomous driving.arXiv.org,

    Hongyu Zhou, Longzhong Lin, Jiabao Wang, Yichong Lu, Dongfeng Bai, Bingbing Liu, Yue Wang, Andreas Geiger, and Yiyi Liao. Hugsim: A real-time, photo-realistic and closed-loop simulator for autonomous driving.arXiv.org,

    work page

  64. [64]

    Hidden biases of end-to-end driving datasets

    Julian Zimmerlin, Jens Beißwenger, Bernhard Jaeger, An- dreas Geiger, and Kashyap Chitta. Hidden biases of end-to- end driving datasets.arXiv.org, 2412.09602, 2024. 2, 3, 5, 6, 7 11

    work page arXiv 2024