Pith. sign in

REVIEW 5 major objections 6 minor 3 cited by

IRL-VLA: Training an Vision-Language-Action Policy via Reward World Model

T0 review · 5 major / 6 minor · reviewed 2026-08-05 · deepseek-v4-flash

Pith's one-line read A simulator-free reward world model lets reinforcement learning fine-tune a VLA driving policy, and the result beats imitation-only training on NAVSIM v2.

desk verdict A plausible three-stage recipe for closed-loop VLA training via a learned reward model; the central result is externally benchmarked but the reward model is unvalidated, so the 74.9 EPDMS gain could partly be reward hacking. read the letter →

arxiv 2508.06571 v3 pith:47NZQJKU submitted 2025-08-07 cs.AI cs.CVcs.RO

classification cs.AIcs.CVcs.RO
keywords vision-language-actionclosed-loopreinforcementlearningrewardworldmodelinversediffusionpolicyautonomousdrivingNAVSIMPPO
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

The paper tries to show that a vision-language-action (VLA) driving model can be improved by closed-loop reinforcement learning without a simulator, by replacing simulator-based reward computation with a lightweight learned Reward World Model (RWM). The pipeline first imitates human driving with a diffusion-based VLA planner, then trains the RWM on offline trajectories and their EPDMS sub-scores, and finally fine-tunes the policy with PPO using RWM rewards plus a behavior-cloning regularizer. On the NAVSIM navhard-real split, the RL-tuned model reaches 74.9 EPDMS versus 74.4 for the same model after imitation pretraining, and the approach reports first runner-up placement in the challenge benchmark. If correct, this matters because it offers a scalable path to closed-loop VLA training that avoids high-fidelity sensor simulation and the sim-to-real gap.

What carries the argument

The load-bearing object is the Reward World Model (RWM): an inverse-RL trained network that maps each trajectory's BEV-space features to per-metric predictions, $$\hat{r}_m = \mathrm{MLP}_m(f_{\mathrm{traj}}),\quad \hat{r}_{\mathrm{epdms}}=\sum_m w_m \hat{r}_m,$$ with $m \in \{\text{NC, DAC, DDC, TLC, EP, TTC, LK, HC}\}$ and weights from EPDMS. It replaces simulator-based scoring during PPO. The second mechanism is the diffusion-policy-as-MDP view: each denoising step is a transition modeled by a Gaussian policy $\pi_\theta(x_{t-1}|x_t,c)$, whose joint log-likelihood forms the PPO objective with KL-to-reference and an imitation term.

What would settle it

Compare RWM-predicted EPDMS with true simulator EPDMS on trajectories sampled from the final PPO policy; if predictions are systematically optimistic, or if per-metric prediction error grows across PPO iterations, the RL gain is an artifact of the learned reward. A sharper version: retrain the RWM with EC included and check whether the 74.9 navhard-real EPDMS gain survives.

Watch

Extended reading notes

Core claim

IRL-VLA is a three-stage method for training a VLA driving policy in closed loop without a simulator. Stage one is imitation pretraining of a VLA whose planner is a conditional diffusion model over anchored trajectories, conditioned on semantic (VLM) and 3D (BEV, detection, map) reasoning. Stage two trains a lightweight Reward World Model: for each of eight EPDMS sub-metrics it predicts the sub-score from BEV trajectory features, and the sub-scores are weighted into a single reward. The training data deliberately includes intermediate diffusion steps, K-means-sampled trajectory patterns with K from 32 to 8192, and multiple ego poses to cover diverse behavior. Stage three fine-tunes the diffu

Load-bearing premise

The learned Reward World Model, trained once on offline trajectories and scores, stays accurate on the different trajectories the policy visits during PPO; if it over-scores those, PPO will maximize the model's mistakes rather than driving quality.

Editorial extensions

If this is right

  • Closed-loop RL for VLA policies is feasible without high-fidelity sensor simulation: the RWM supplies real-time per-trajectory rewards, so PPO no longer needs rendering or physics simulation.
  • RL fine-tuning changes the trade-off surface: Ego Progress rises from 83.9 to 96.2 while most safety and comfort sub-scores fall slightly, yet net EPDMS improves from 74.4 to 74.9 on navhard-real.
  • The balance between RL and imitation matters: imitation weight 0.5 gives 74.9, while 1.0 and 0.1 give 73.9 and 73.4, so the diffusion policy needs the behavior-cloning term for stable optimization.
  • Because the RWM is trained offline on diverse samples (intermediate diffusion steps, K-means trajectory modes, multiple ego poses), the same reward infrastructure can be reused across many scenes without rerunning a simulator per policy update.

Reading between the lines

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

  • A testable consequence of excluding EC from the RWM: the RL stage is blind to Extended Comfort, and the reported table shows EC dropping from 76.0 to 72.4 after RL; including EC in the reward would probably shrink the net EPDMS gain.
  • The paper does not report any distribution-shift validation of the RWM; measuring RWM prediction error on post-PPO trajectories would decide whether the improvement is a real policy gain or exploitation of reward-model regression error.
  • The RWM could also be used at inference time as a differentiable guide for diffusion denoising (reward-guided trajectory scoring), giving a cheaper way to exploit the same learned reward without further policy updates.
  • If the RWM stays unbiased for other sensor modalities, the same BEV-trajectory-feature interface should let IRL-VLA extend to LiDAR- or radar-only inputs; that is a direct experiment suggested by the architecture.
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

5 major / 6 minor

Summary. The paper proposes IRL-VLA, a three-stage autonomous-driving framework: (1) imitation-learning pretraining of a VLA policy that combines semantic reasoning, 3D/BEV reasoning, and a diffusion-based planner; (2) training a lightweight Reward World Model (RWM) by inverse reinforcement learning to predict EPDMS sub-scores from sensor features and planned trajectories; (3) fine-tuning the diffusion VLA policy with PPO using RWM rewards, with an added imitation term and a KL regularizer. The authors report 45.0 EPDMS as first runner-up in the CVPR 2025 NAVSIM v2 challenge and 74.9 EPDMS on the navhard-real benchmark, claiming the first closed-loop VLA trained via end-to-end RL without a simulator.

Significance. If the learned RWM is an accurate and distribution-robust proxy for the EPDMS simulator, replacing sensor-rendering simulation with a lightweight reward model is a meaningful step toward scalable closed-loop training for large VLA policies. The evaluation is anchored to the external NAVSIM benchmark, so the final numbers are not produced by the authors' own machinery; this keeps the circularity risk low. The idea of applying PPO to the diffusion denoising chain with a learned reward model is also timely. However, the central empirical claim rests on a 0.5 EPDMS gain over the authors' own imitation baseline (74.4 to 74.9), and the paper does not validate the RWM on the policy's actual rollout distribution. Several load-bearing implementation details are also underspecified.

major comments (5)
  1. [Sec. 3.3, 3.4; Table 1] The RWM is the sole reward source for PPO (Eq. 5), yet no validation accuracy, calibration curve, or distribution-shift check is reported. If the RWM over-scores trajectories visited after PPO updates, the policy can exploit regression error rather than improve true EPDMS. Table 1 shows a pattern consistent with such exploitation: EP rises 83.9→96.2 while NC (98.3→96.9), DAC (92.4→91.3), TTC (97.1→96.2), and EC (76.0→72.4) all decline, with total EPDMS improving by only 0.5. Please report per-metric RWM accuracy on held-out navtrain data and on rollouts before/after RL, and show how RWM-predicted reward tracks true EPDMS during PPO training. Also justify excluding EC from RWM while the final EPDMS includes it.
  2. [Sec. 3.4, Algorithm 1] Algorithm 1 line 6 invokes an unexplained transition model \hat{T}(s_t,a_t); no definition, training procedure, or ablation for \hat{T} appears anywhere in the paper. Section 3.3 only defines the reward predictor, not a dynamics model. The critic/value function V_phi is also unspecified, and Sec. 3.4 refers to a 'critical network' rather than a critic. Without these components, the RL loop cannot be reproduced, and the claim of 'end-to-end reinforcement learning including sensor input' is not verifiable.
  3. [Abstract, Sec. 1, Sec. 4] The abstract and Contribution 1 say the VLA is trained 'without relying on simulator during training,' but Sec. 4 states that the RWM is trained 'leveraging expert demonstrations and simulator feedbacks.' The actual contribution is avoiding sensor rendering / high-fidelity simulation at RL time, not avoiding simulator feedback altogether. Please restate the claim precisely and list exactly which simulator computations are needed: RWM label generation, the two-simulation EC metric, and final evaluation.
  4. [Abstract, Sec. 1, Table 1] The headline '45.0 EDPMS as 1st runner up' is not reconciled with Table 1, which reports 74.9 EPDMS on navhard-real. If 45.0 is from a different challenge split/protocol, that protocol must be described; as written, the primary advertised result cannot be checked against the paper's own table. The authors should clearly state the relationship between the two numbers.
  5. [Table 1, Table 3] All EPDMS scores are reported as single runs with no standard deviations or confidence intervals. The central RL improvement is 74.4→74.9 (0.5 points) and the best-vs-second w_IL gap in Table 3 is 1.0 point; these differences are within typical run-to-run variation for such benchmarks. Please provide multiple seeds / error bars, at least for IRL-VLA-PT, IRL-VLA-RL, and the w_IL ablation, so the reader can judge whether the claimed improvement is statistically meaningful.
minor comments (6)
  1. [Abstract] Typo: 'guidence' should be 'guidance.'
  2. [Sec. 3.3, Sec. 4] The RWM loss is written as a weighted L1 error over all metrics, but Sec. 4 says BCE is used for binary metrics, MSE for [0,1] metrics, and cross-entropy for {0,0.5,1} metrics. Reconcile the two descriptions.
  3. [Sec. 3.4, Eq. (5)] The KL term DKL(pi_theta || pi_ref) uses pi_ref, which is not defined before its first use. Also, the discount gamma over denoising steps 'mitigating instability in early denoising steps' is unclear because the summation index t=1..tau is not tied to the denoising direction.
  4. [Sec. 3.4, Algorithm 1] Variable T is overloaded: for t = 1 to T for trajectory collection, while l is already defined as trajectory length. Rename to avoid confusion.
  5. [Fig. 2] The caption references subfigures (e), (f), and (g), but the text and figure do not consistently describe what (g) contains; check numbering and ensure all panels are referenced.
  6. [References] Reference [27] is missing venue/arXiv number; several other references are duplicated (e.g., DiffusionDrive appears as [20] and [21]).

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: final EPDMS is computed by the external NAVSIM simulator, not by the learned Reward World Model; the RWM is a supervised proxy used for training, and the reported improvement is an empirical outcome.

full rationale

The paper's load-bearing empirical claim is the EPDMS score on NAVSIM navhard-real and the CVPR2025 challenge result. These are computed by the external NAVSIM non-reactive simulator and the EPDMS metric, not by the authors' Reward World Model (RWM). The RWM is trained by supervised regression to predict NAVSIM EPDMS sub-scores (Sec. 3.3, L_Repdms) and is then used as a reward signal in PPO (Sec. 3.4), but the final evaluation re-runs the policy in NAVSIM and computes EPDMS independently. Optimizing a learned proxy for an external scorer is not circular: the improvement from 74.4 to 74.9 is not forced by construction, and several sub-scores decline (NC 98.3 to 96.9, EC 76.0 to 72.4), which is inconsistent with the final number being an artifact of the training objective. No load-bearing argument reduces to a self-citation chain. The same-author citation DiffVLA [13] appears in related work as a prior imitation-learning VLA and is not used to justify the RWM or the RL derivation; DiffSemanticFusion [27] is likewise a related-work mention. No uniqueness theorem from the authors' prior work is invoked. There are, however, non-circular correctness concerns: Sec. 4 states the RWM is trained 'leveraging expert demonstrations and simulator feedbacks,' which qualifies the abstract's claim of training 'without relying on simulator'; Sec. 3.3 says RWM data are collected from 'each scene in the NA VSIM dataset' without explicitly restricting to navtrain, which could imply navhard leakage; and the paper does not report RWM validation accuracy or distribution-shift checks. These are benchmark-integrity and robustness risks, not circular reductions, so they do not raise the circularity score above the minor-self-citation level.

Assumptions & free parameters 4 free parameters · 5 assumptions · 1 invented entities

The central claim rests on the EPDMS score definition and metric weights from prior benchmark work (refs [3,6]), on NAVSIM's non-reactive simulation as a valid training and evaluation proxy, and on the unvalidated generalization of the RWM to the RL policy's trajectory distribution. The imitation loss weight w_IL is tuned on the evaluation set itself. The RWM is the only invented component, and it is a learned regressor with checkable predictions, though the paper does not report that check.

free parameters (4)
  • w_IL (imitation loss weight) = 0.5
    Ablation in Table 3 searches 1.0/0.5/0.1 on navhard-real; 0.5 is chosen because it gives the highest EPDMS and is then reported as the final RL result in Table 1.
  • lambda (IL BCE weight) = not reported
    Appears in L_IL as the balance between the L1 trajectory reconstruction loss and the BCE classification loss (Sec. 3.2); no value or tuning procedure is given.
  • gamma_denoise (discount over denoising steps, Eq. 5) = not reported
    Used as gamma^(t-1) in the PPO objective to down-weight early denoising steps; its value is never stated in Sec. 3.4.
  • RWM metric weights w_m = fixed by EPDMS definition (refs [3,6])
    Not fitted by the authors; inherited from the benchmark scoring rule, but the RL reward is a weighted sum over them, so the central claim inherits these benchmark-determined weights.
assumptions (5)
  • standard math Diffusion model forward/reverse process with fixed covariance sigma_t^2 I (Eq. 3) and the joint log-likelihood factorization (Eq. 4).
    Standard diffusion-policy modeling invoked in Sec. 3.4 to convert denoising steps into an MDP for PPO.
  • domain assumption NAVSIM's non-reactive simulation with EPDMS scoring is a valid proxy for driving quality and a valid RL objective.
    The entire reward signal derives from EPDMS (Sec. 3.3); agents in the simulator do not react to the ego vehicle, so 'closed-loop' excludes reactive traffic behavior.
  • domain assumption The RWM, trained on curated trajectories from the pretrained policy, generalizes to trajectories produced by later RL policy iterations; its regression accuracy is never measured.
    The RL loop optimizes RWM predictions (Algorithm 1, Sec. 3.4); the paper reports no RWM validation metrics or distribution-shift checks.
  • ad hoc to paper The diversity recipe for RWM training data (recording intermediate diffusion steps, K-means clustering with K from 32 to 8192, multiple ego poses) is sufficient for reward accuracy.
    Sec. 3.3; no principled justification links this coverage scheme to RWM error on the RL policy's trajectory distribution.
  • domain assumption Excluding EC from the reward does not bias the RL objective.
    EC is excluded from the RWM (Sec. 3.3) because it needs two simulations per scene, but EC remains part of the evaluated EPDMS, so RL optimizes 8 of 9 benchmark components.
invented entities (1)
  • Reward World Model (RWM) independent evidence
    purpose: A learned regressor that predicts EPDMS sub-scores from camera features and candidate trajectories, replacing simulator-based reward computation during PPO fine-tuning.
    It is a model artifact, not a physical postulate; its predictions are checkable against true NAVSIM scores, though the paper does not report that check.

how reviews work

0 comments
Cite this review

Pith. "Pith review of IRL-VLA: Training an Vision-Language-Action Policy via Reward World Model." pith.science (2026). https://pith.science/paper/47NZQJKU

@misc{pith2026250806571,
  author       = {Pith},
  title        = {Pith review of: IRL-VLA: Training an Vision-Language-Action Policy via Reward World Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/47NZQJKU}},
  note         = {Machine review of arXiv:2508.06571}
}
read the original abstract

Vision-Language-Action (VLA) models have demonstrated potential in autonomous driving. However, two critical challenges hinder their development: (1) Existing VLA architectures are typically based on imitation learning in open-loop setup which tends to capture the recorded behaviors in the dataset, leading to suboptimal and constrained performance, (2) Close-loop training relies heavily on high-fidelity sensor simulation, where domain gaps and computational inefficiencies pose significant barriers. In this paper, we introduce IRL-VLA, a novel close-loop Reinforcement Learning via \textbf{I}nverse \textbf{R}einforcement \textbf{L}earning reward world model with a self-built VLA approach. Our framework proceeds in a three-stage paradigm: In the first stage, we propose a VLA architecture and pretrain the VLA policy via imitation learning. In the second stage, we construct a lightweight reward world model via inverse reinforcement learning to enable efficient close-loop reward computation. To further enhance planning performance, finally, we design specialized reward world model guidence reinforcement learning via PPO(Proximal Policy Optimization) to effectively balance the safety incidents, comfortable driving, and traffic efficiency. Our approach achieves state-of-the-art performance in NAVSIM v2 end-to-end driving benchmark, 1st runner up in CVPR2025 Autonomous Grand Challenge. We hope that our framework will accelerate VLA research in close-loop autonomous driving.

Figures

Figures reproduced from arXiv: 2508.06571 by the authors.

Figure 1
Figure 1. Different paradigms of VLA autonomous driving (AD) a).Imitation learning for VLA AD. b). Simulator-based rein￾forcement learning for VLA AD. c). IRL-VLA explores improving high-capacity VLA with scalable reinforcement learning without heavy simulator. 1. Introduction End-to-end autonomous driving has emerged as a sig￾nificant and rapidly growing research area. With the abun￾dance of available human driving demonstra… view at source ↗
Figure 2
Figure 2. Overview of the IRL-VLA Framework. This figure illustrates the three-stage pipeline of our close-loop Reinforcement Learning via Reward World Model framework for Vision-Language-Action (VLA) in autonomous driving. a) Imitation Policy Learning initializes the VLA model as a supervised policy via sensor input and planning trajectories. b) Inverse Environment Learning constructs the Reward World Model (RWM) via pretrai… view at source ↗

Discussion (0). Sign in to comment.

Forward citations

Cited by 3 Pith papers

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

  1. WCog-VLA: A Dual-Level World-Cognitive Vision-Language-Action Model for End-to-End Autonomous Driving

    cs.CV 2026-07 conditional novelty 6.0 of 10

    WCog-VLA couples Game-CoT semantic reasoning with an aligned decoupled diffusion transformer to generate joint multi-agent trajectories and reaches 92.9 PDMS on NAVSIM.

  2. MindDrive: A Vision-Language-Action Model for Autonomous Driving via Online Reinforcement Learning

    cs.CV 2025-12 conditional novelty 5.0 of 10

    An autonomous-driving vision-language model that uses online RL over discrete language actions, with a separate action expert mapping decisions to trajectories, reports DS 78.04 and SR 55.09% on Bench2Drive.

  3. Post-Training in End-to-End Autonomous Driving

    cs.CV 2026-07 unverdicted novelty 4.0 of 10

    Post-training for end-to-end autonomous driving is surveyed and grouped into four supervision-based families to address limits of open-loop imitation.

Reference graph

Works this paper leans on

31 extracted references · 12 canonical work pages · cited by 3 Pith papers

  1. [1]

    Gpt-4 technical report

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report. arXiv preprint arXiv:2303.08774, 2023. 3

  2. [2]

    Training diffusion models with reinforcement learning, 2024

    Kevin Black, Michael Janner, Yilun Du, Ilya Kostrikov, and Sergey Levine. Training diffusion models with reinforcement learning, 2024. 6

  3. [3]

    Pseudo- simulation for autonomous driving

    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. arXiv preprint arXiv:2506.04218, 2025. 4, 5

  4. [4]

    Transfuser: Imitation with transformer-based sensor fusion for au- tonomous driving

    Kashyap Chitta, Aditya Prakash, Bernhard Jaeger, Ze- hao Yu, Katrin Renz, and Andreas Geiger. Transfuser: Imitation with transformer-based sensor fusion for au- tonomous driving. IEEE Transactions on Pattern Anal- ysis and Machine Intelligence, 45(11):12878–12895,

  5. [5]

    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. In Conference on Robot Learning, pages 1268–1281. PMLR, 2023. 4

  6. [6]

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

    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 In- formation Processing Systems, 37:28706–28719, 2025. 4

  7. [7]

    Orion: A holistic end-to-end autonomous driving framework by vision-language instructed action gen- eration

    Haoyu Fu, Diankun Zhang, Zongchuang Zhao, Jian- feng Cui, Dingkang Liang, Chong Zhang, Dingyuan Zhang, Hongwei Xie, Bing Wang, and Xiang Bai. Orion: A holistic end-to-end autonomous driving framework by vision-language instructed action gen- eration. arXiv preprint arXiv:2503.19755, 2025. 2, 3

  8. [8]

    Rad: Training an end-to-end driving policy via large-scale 3dgs-based reinforcement learning

    Hao Gao, Shaoyu Chen, Bo Jiang, Bencheng Liao, Yiang Shi, Xiaoyang Guo, Yuechuan Pu, Haoran Yin, Xiangyu Li, Xinbang Zhang, et al. Rad: Training an end-to-end driving policy via large-scale 3dgs-based reinforcement learning. arXiv preprint arXiv:2502.13144, 2025. 3

Show all 31 references
  1. [9]

    Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning

    Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. arXiv preprint arXiv:2501.12948, 2025. 3

  2. [10]

    Planning-oriented autonomous driving

    Yihan Hu, Jiazhi Yang, Li Chen, Keyu Li, Chonghao Sima, Xizhou Zhu, Siqi Chai, Senyao Du, Tianwei Lin, Wenhai Wang, et al. Planning-oriented autonomous driving. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 17853– 17862, 2023. 1, 2

  3. [11]

    Emma: End-to- end multimodal model for autonomous driving

    Jyh-Jing Hwang, Runsheng Xu, Hubert Lin, Wei-Chih Hung, Jingwei Ji, Kristy Choi, Di Huang, Tong He, Paul Covington, Benjamin Sapp, et al. Emma: End-to- end multimodal model for autonomous driving. arXiv preprint arXiv:2410.23262, 2024. 2

  4. [12]

    Planning with diffusion for flexible behavior synthesis

    Michael Janner, Yilun Du, Joshua B Tenenbaum, and Sergey Levine. Planning with diffusion for flexible behavior synthesis. arXiv preprint arXiv:2205.09991,

  5. [13]

    Diffvla: Vision-language guided diffusion planning for autonomous driving

    Anqing Jiang, Yu Gao, Zhigang Sun, Yiru Wang, Jijun Wang, Jinghao Chai, Qian Cao, Yuweng Heng, Hao Jiang, Yunda Dong, et al. Diffvla: Vision-language guided diffusion planning for autonomous driving. arXiv preprint arXiv:2505.19381, 2025. 3

  6. [14]

    Senna: Bridging large vision- language models and end-to-end autonomous driving

    Bo Jiang, Shaoyu Chen, Bencheng Liao, Xingyu Zhang, Wei Yin, Qian Zhang, Chang Huang, Wenyu Liu, and Xinggang Wang. Senna: Bridging large vision- language models and end-to-end autonomous driving. arXiv preprint arXiv:2410.22313, 2024. 2

  7. [15]

    Vad: Vectorized scene representation for efficient autonomous driving

    Bo Jiang, Shaoyu Chen, Qing Xu, Bencheng Liao, Jiajie Chen, Helong Zhou, Qian Zhang, Wenyu Liu, Chang Huang, and Xinggang Wang. Vad: Vectorized scene representation for efficient autonomous driving. In Proceedings of the IEEE/CVF International Confer- ence on Computer Vision, ...

  8. [16]

    End-to-end driving with online trajectory evaluation via bev world model.arXiv preprint arXiv:2504.01941, 2025

    Yingyan Li, Yuqi Wang, Yang Liu, Jiawei He, Lue Fan, and Zhaoxiang Zhang. End-to-end driving with online trajectory evaluation via bev world model.arXiv preprint arXiv:2504.01941, 2025. 4

  9. [17]

    Recogdrive: A rein- forced cognitive framework for end-to-end autonomous driving

    Yongkang Li, Kaixin Xiong, Xiangyu Guo, Fang Li, Sixu Yan, Gangwei Xu, Lijun Zhou, Long Chen, Haiyang Sun, Bing Wang, et al. Recogdrive: A rein- forced cognitive framework for end-to-end autonomous driving. arXiv preprint arXiv:2506.08052, 2025. 2, 6

  10. [18]

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

    Zhenxin Li, Kailin Li, Shihao Wang, Shiyi Lan, Zhid- ing Yu, Yishen Ji, Zhiqi Li, Ziyue Zhu, Jan Kautz, Zuxuan Wu, et al. Hydra-mdp: End-to-end multimodal planning with multi-target hydra-distillation. arXiv preprint arXiv:2406.06978, 2024. 4

  11. [19]

    Generalized tra- jectory scoring for end-to-end multimodal planning

    Zhenxin Li, Wenhao Yao, Zi Wang, Xinglong Sun, Joshua Chen, Nadine Chang, Maying Shen, Zuxuan Wu, Shiyi Lan, and Jose M Alvarez. Generalized tra- jectory scoring for end-to-end multimodal planning. arXiv preprint arXiv:2506.06664, 2025. 4, 5

  12. [20]

    Diffusiondrive: Trun- cated diffusion model for end-to-end autonomous driv- ing

    Bencheng Liao, Shaoyu Chen, Haoran Yin, Bo Jiang, Cheng Wang, Sixu Yan, Xinbang Zhang, Xiangyu Li, Ying Zhang, Qian Zhang, et al. Diffusiondrive: Trun- cated diffusion model for end-to-end autonomous driv- ing. arXiv preprint arXiv:2411.15139, 2024. 4

  13. [21]

    Diffusiondrive: Trun- cated diffusion model for end-to-end autonomous driv- ing

    Bencheng Liao, Shaoyu Chen, Haoran Yin, Bo Jiang, Cheng Wang, Sixu Yan, Xinbang Zhang, Xiangyu Li, Ying Zhang, Qian Zhang, et al. Diffusiondrive: Trun- cated diffusion model for end-to-end autonomous driv- ing. In Proceedings of the Computer Vision and Pattern Recognition Conf...

  14. [22]

    Diffusion policy policy optimization

    Allen Z Ren, Justin Lidard, Lars L Ankile, Anthony Simeonov, Pulkit Agrawal, Anirudha Majumdar, Ben- jamin Burchfiel, Hongkai Dai, and Max Simchowitz. Diffusion policy policy optimization. arXiv preprint arXiv:2409.00588, 2024. 5

  15. [23]

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

    Katrin Renz, Long Chen, Elahe Arani, and Oleg Sinavski. Simlingo: Vision-only closed-loop au- tonomous driving with language-action alignment. In Proceedings of the Computer Vision and Pattern Recog- nition Conference, pages 11993–12003, 2025. 2, 3

  16. [24]

    High-dimensional continuous control using generalized advantage estima- tion, 2018

    John Schulman, Philipp Moritz, Sergey Levine, Michael Jordan, and Pieter Abbeel. High-dimensional continuous control using generalized advantage estima- tion, 2018. 6

  17. [25]

    Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, Y . K. Li, Y . Wu, and Daya Guo. Deepseekmath: Pushing the limits of mathematical reasoning in open language models, 2024. 6

  18. [26]

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

    Wenchao Sun, Xuewu Lin, Yining Shi, Chuang Zhang, Haoran Wu, and Sifa Zheng. Sparsedrive: End-to-end autonomous driving via sparse scene representation. arXiv preprint arXiv:2405.19620, 2024. 1, 2

  19. [27]

    Diffsemanticfusion: Semantic raster bev fusion for autonomous driving via online hd map diffusion, 2025

    Zhigang Sun, Yiru Wang, Anqing Jiang, Shuo Wang, Yu Gao, Yuwen Heng, Shouyi Zhang, An He, Hao Jiang, Jinhao Chai, Zichong Gu, Wang Jijun, Shichen Tang, Lavdim Halilaj, Juergen Luettin, and Hao Sun. Diffsemanticfusion: Semantic raster bev fusion for autonomous driving via onlin...

  20. [28]

    Effi- cient reinforcement learning for autonomous driving with parameterized skills and priors

    Letian Wang, Jie Liu, Hao Shao, Wenshuo Wang, Ruob- ing Chen, Yu Liu, and Steven L Waslander. Effi- cient reinforcement learning for autonomous driving with parameterized skills and priors. arXiv preprint arXiv:2305.04412, 2023. 3

  21. [29]

    Carplanner: Consistent auto-regressive trajectory plan- ning for large-scale reinforcement learning in au- tonomous driving

    Dongkun Zhang, Jiaming Liang, Ke Guo, Sha Lu, Qi Wang, Rong Xiong, Zhenwei Miao, and Yue Wang. Carplanner: Consistent auto-regressive trajectory plan- ning for large-scale reinforcement learning in au- tonomous driving. In Proceedings of the Computer Vi- sion and Pattern Recog...

  22. [30]

    Accelerating reinforcement learning for autonomous driving using task-agnostic and ego- centric motion skills

    Tong Zhou, Letian Wang, Ruobing Chen, Wenshuo Wang, and Yu Liu. Accelerating reinforcement learning for autonomous driving using task-agnostic and ego- centric motion skills. In 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 11289–11296....

  23. [31]

    Opendrivevla: Towards end-to-end autonomous driving with large vision language action model

    Xingcheng Zhou, Xuyuan Han, Feng Yang, Yunpu Ma, and Alois C Knoll. Opendrivevla: Towards end-to-end autonomous driving with large vision language action model. arXiv preprint arXiv:2503.23463, 2025. 2

Pith tools

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