pith. sign in

arxiv: 2606.27504 · v1 · pith:BLTRYIPYnew · submitted 2026-06-25 · 💻 cs.CV

ReWorld: Learning Better Representations for World Action Models

Tianze Xia , Lijun Zhou , Kaixin Xiong , Jingfeng Yao , Yu Zhu , Zhenxin Zhu , Bing Wang , Guang Chen
show 4 more authors
Hangjun Ye Wenyu Liu Haiyang Sun Xinggang Wang
This is my paper

Pith reviewed 2026-06-29 01:55 UTC · model grok-4.3

classification 💻 cs.CV
keywords representation learningworld action modelsautonomous drivingvideo generationplanningdiffusion transformers
0
0 comments X

The pith

ReWorld improves world action models by directly supervising intermediate representations with three targeted losses.

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

World action models for autonomous driving currently train only on final outputs, leaving the intermediate representations that encode world knowledge to form as indirect byproducts. ReWorld makes those representations explicit optimization targets by adding future-predictive supervision to the Video DiT, cross-modal alignment between video and action modules, and hard-negative supervision to sharpen safety-critical distinctions. The resulting models produce higher-quality generated videos, stronger closed-loop planning scores, and faster convergence from scratch on nuScenes and NAVSIM benchmarks. A reader would care because the changes occur inside the standard training loop and require no later reinforcement learning or post-processing.

Core claim

ReWorld is a representation learning framework that treats intermediate representations inside the Video DiT and Action DiT as direct targets of optimization rather than byproducts. It applies future-predictive supervision to the generation module's hidden states, aligns the planning module's states cross-modally with the video world representation, and further shapes them with hard-negative supervision around safety-critical boundaries. This addresses the gap left by output-only supervision and produces the measured gains in generation and planning.

What carries the argument

Three complementary supervision signals applied directly to intermediate representations: future-predictive loss on Video DiT, cross-modal alignment on Action DiT, and hard-negative loss for safety discrimination.

If this is right

  • Fine-tuned video generation improves by 23.9 percent in FVD from 81.3 to 61.9.
  • Closed-loop planning score rises from 89.1 to 90.4 without reinforcement learning or post-processing.
  • From-scratch training converges approximately twice as fast.

Where Pith is reading between the lines

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

  • The same intermediate-supervision pattern could be tested on world models outside autonomous driving.
  • Existing representation-learning techniques may need modification to address cross-modal and safety-boundary requirements specific to action-conditioned driving models.

Load-bearing premise

The reported performance gains are produced by the three new intermediate-representation losses rather than by unmentioned differences in architecture, data, or training schedule.

What would settle it

An experiment that adds only the three ReWorld losses to an otherwise identical baseline world action model and measures the resulting change in FVD and PDMS.

Figures

Figures reproduced from arXiv: 2606.27504 by Bing Wang, Guang Chen, Haiyang Sun, Hangjun Ye, Jingfeng Yao, Kaixin Xiong, Lijun Zhou, Tianze Xia, Wenyu Liu, Xinggang Wang, Yu Zhu, Zhenxin Zhu.

Figure 1
Figure 1. Figure 1: Overview of the ReWorld framework. ReWorld trains a chained world-action model through three stages. Stage 1 trains the Video DiT with the generation loss and an intermediate￾guidance loss, which supervises auxiliary heads on selected blocks to predict the flow-matching ve￾locity target, making intermediate representations future-predictive. Stage 2 freezes the Video DiT and trains the Action DiT with traj… view at source ↗
Figure 2
Figure 2. Figure 2: Intermediate-supervised inference and accelerated convergence of ReWorld. (a) Dur￾ing sampling, ReWorld exploits the discrepancy between the intermediate prediction vi and the final prediction vf to form a corrected velocity vw, which is used by the scheduler to advance the denoising trajectory. (b) ReWorld achieves faster convergence than Vanilla Flow Matching by ap￾proximately 2× without using any extern… view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparison with DriveLaW for video generation [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
read the original abstract

World Action Models (WAMs) model future environment evolution under action conditioning, offering a scalable paradigm for autonomous driving. However, existing approaches focus largely on model architecture design, and how a WAM can efficiently learn better world representations for planning remains underexplored. To address this gap, we propose ReWorld, the first representation learning framework specifically designed for autonomous-driving world action models. In WAMs, standard training supervises only the output ends of the generation and planning modules, leaving the intermediate representations that carry world knowledge to be shaped only indirectly, as byproducts of fitting these outputs. The core idea of ReWorld is to treat intermediate representations as direct targets of optimization, shaping them along three complementary dimensions. On the Video DiT responsible for generation, we impose future-predictive supervision on its intermediate representations. On the Action DiT responsible for planning, we first align its intermediate representations cross-modally with the video world representation, then further shape them to be discriminative around safety-critical boundaries via hard-negative supervision. In addition, we systematically analyze the effectiveness of existing representation learning methods in video generation world models, and discuss why their performance is limited on this task. Experiments on nuScenes and NAVSIM show that ReWorld improves fine-tuned video generation by 23.9% in FVD (81.3 to 61.9), raises closed-loop PDMS from 89.1 to 90.4 without any post-training such as RL or post-processing, and accelerates from-scratch convergence by approximately 2x.

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 / 1 minor

Summary. The paper proposes ReWorld, a representation learning framework for World Action Models (WAMs) in autonomous driving. It introduces three intermediate-representation losses: future-predictive supervision on Video DiT intermediates, cross-modal alignment on Action DiT intermediates, and hard-negative supervision around safety-critical boundaries. Experiments on nuScenes and NAVSIM report a 23.9% FVD reduction in fine-tuned video generation (81.3 to 61.9), PDMS improvement from 89.1 to 90.4 in closed-loop planning without post-training, and ~2x faster from-scratch convergence.

Significance. If the gains are causally due to the proposed losses rather than uncontrolled factors, the work would demonstrate a practical way to directly optimize intermediate representations in world models, improving both generation and planning without RL or post-processing. This could influence scalable training of action-conditioned video models for driving.

major comments (2)
  1. [Experiments] Experiments section: The abstract directly attributes the FVD (81.3→61.9), PDMS (89.1→90.4), and convergence improvements to the three new losses, yet no ablation experiments are described that re-train the baseline WAM under identical architecture, data, optimizer, and schedule with only those losses removed. This control is required to support the causal claim.
  2. [Method] Method section: The implementation details of the future-predictive supervision (Video DiT), cross-modal alignment, and hard-negative supervision (Action DiT) must include explicit loss equations, layer selection, and weighting coefficients; without them the reproducibility of the reported deltas cannot be assessed.
minor comments (1)
  1. [Abstract] The abstract states that existing representation learning methods were systematically analyzed for limitations on this task; a dedicated table or subsection summarizing those findings and why they underperform would improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We will revise the manuscript to address the concerns about missing ablations and implementation details.

read point-by-point responses

  1. Referee: [Experiments] Experiments section: The abstract directly attributes the FVD (81.3→61.9), PDMS (89.1→90.4), and convergence improvements to the three new losses, yet no ablation experiments are described that re-train the baseline WAM under identical architecture, data, optimizer, and schedule with only those losses removed. This control is required to support the causal claim.

    Authors: We agree that controlled ablations isolating the contribution of the three proposed losses are required to substantiate the causal claims in the abstract. In the revised manuscript we will add experiments that retrain the baseline WAM from scratch under identical architecture, data, optimizer, and schedule while removing only the new losses. revision: yes

  2. Referee: [Method] Method section: The implementation details of the future-predictive supervision (Video DiT), cross-modal alignment, and hard-negative supervision (Action DiT) must include explicit loss equations, layer selection, and weighting coefficients; without them the reproducibility of the reported deltas cannot be assessed.

    Authors: We will expand the Method section to include the explicit loss equations for future-predictive supervision, cross-modal alignment, and hard-negative supervision, together with the chosen DiT layers and the numerical weighting coefficients applied in the combined objective. revision: yes

Circularity Check

0 steps flagged

No circularity in derivation chain

full rationale

The paper introduces ReWorld as a representation learning framework adding three explicit intermediate losses (future-predictive on Video DiT, cross-modal alignment plus hard-negative on Action DiT) and reports empirical metric improvements on nuScenes/NAVSIM. No equations, self-citations, or uniqueness theorems are present in the supplied text. The performance deltas are presented as experimental outcomes rather than any mathematical prediction or quantity that reduces by construction to the inputs or to a fitted parameter. The central claim therefore remains an independent empirical assertion whose validity depends on experimental controls, not on definitional equivalence or self-referential fitting.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only input supplies no equations, no training objectives, and no architectural diagrams, so free parameters, axioms, and invented entities cannot be identified or counted.

pith-pipeline@v0.9.1-grok · 5842 in / 1207 out tokens · 29397 ms · 2026-06-29T01:55:59.736165+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

34 extracted references · 19 linked inside Pith

  1. [1]

    Cosmos world foundation model platform for physical ai.arXiv preprint arXiv:2501.03575,

    Niket Agarwal, Arslan Ali, Maciej Bala, Yogesh Balaji, Erik Barker, Tiffany Cai, Prithvijit Chat- topadhyay, Yongxin Chen, Yin Cui, Yifan Ding, et al. Cosmos world foundation model platform for physical ai.arXiv preprint arXiv:2501.03575,

    Pith/arXiv arXiv

  2. [2]

    V-jepa 2: Self-supervised video models enable understanding, prediction and planning.arXiv preprint arXiv:2506.09985,

    Mido Assran, Adrien Bardes, David Fan, Quentin Garrido, Russell Howes, Matthew Muckley, Am- mar Rizvi, Claire Roberts, Koustuv Sinha, Artem Zholus, et al. V-jepa 2: Self-supervised video models enable understanding, prediction and planning.arXiv preprint arXiv:2506.09985,

    Pith/arXiv arXiv

  3. [3]

    Latent forcing: Reordering the diffusion trajectory for pixel-space image generation.arXiv preprint arXiv:2602.11401,

    Alan Baade, Eric Ryan Chan, Kyle Sargent, Changan Chen, Justin Johnson, Ehsan Adeli, and Li Fei- Fei. Latent forcing: Reordering the diffusion trajectory for pixel-space image generation.arXiv preprint arXiv:2602.11401,

    arXiv

  4. [4]

    Vavim and vavam: Autonomous driving through video generative modeling.arXiv preprint arXiv:2502.15672,

    Florent Bartoccioni, Elias Ramzi, Victor Besnier, Shashanka Venkataramanan, Tuan-Hung Vu, Yihong Xu, Loick Chambon, Spyros Gidaris, Serkan Odabas, David Hurych, et al. Vavim and vavam: Autonomous driving through video generative modeling.arXiv preprint arXiv:2502.15672,

    arXiv

  5. [5]

    nuplan: A closed-loop ml-based planning benchmark for autonomous vehicles.arXiv preprint arXiv:2106.11810,

    Holger Caesar, Juraj Kabzan, Kok Seang Tan, Whye Kit Fong, Eric Wolff, Alex Lang, Luke Fletcher, Oscar Beijbom, and Sammy Omari. nuplan: A closed-loop ml-based planning benchmark for autonomous vehicles.arXiv preprint arXiv:2106.11810,

    Pith/arXiv arXiv

  6. [6]

    Self-supervised flow matching for scalable multi-modal synthesis

    Hila Chefer, Patrick Esser, Dominik Lorenz, Dustin Podell, Vikash Raja, Vinh Tong, Antonio Tor- ralba, and Robin Rombach. Self-supervised flow matching for scalable multi-modal synthesis. arXiv preprint arXiv:2603.06507,

    arXiv

  7. [7]

    Aligning visual foundation encoders to tokenizers for diffusion models

    Bowei Chen, Sai Bi, Hao Tan, He Zhang, Tianyuan Zhang, Zhengqi Li, Yuanjun Xiong, Jianming Zhang, and Kai Zhang. Aligning visual foundation encoders to tokenizers for diffusion models. InThe Fourteenth International Conference on Learning Representations, 2025a. Shaoyu Chen, Bo Jiang, Hao Gao, Bencheng Liao, Qing Xu, Qian Zhang, Chang Huang, Wenyu Liu, an...

    Pith/arXiv arXiv

  8. [8]

    Rad: Training an end-to-end driving policy via large-scale 3dgs-based reinforcement learning.arXiv preprint arXiv:2502.13144,

    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,

    arXiv

  9. [9]

    Magicdrive: Street view generation with diverse 3d geometry control.arXiv preprint arXiv:2310.02601,

    Ruiyuan Gao, Kai Chen, Enze Xie, Lanqing Hong, Zhenguo Li, Dit-Yan Yeung, and Qiang Xu. Magicdrive: Street view generation with diverse 3d geometry control.arXiv preprint arXiv:2310.02601,

    arXiv

  10. [10]

    Bridging scene generation and planning: Driving with world model via unifying vision and motion representation.arXiv preprint arXiv:2603.14948,

    Xingtai Gui, Meijie Zhang, Tianyi Yan, Wencheng Han, Jiahao Gong, Feiyang Tan, Cheng-zhong Xu, and Jianbing Shen. Bridging scene generation and planning: Driving with world model via unifying vision and motion representation.arXiv preprint arXiv:2603.14948,

    arXiv

  11. [11]

    Genesis: Multimodal driving scene generation with spatio-temporal and cross-modal consistency.arXiv preprint arXiv:2506.07497,

    Xiangyu Guo, Zhanqian Wu, Kaixin Xiong, Ziyang Xu, Lijun Zhou, Gangwei Xu, Shaoqing Xu, Haiyang Sun, Bing Wang, Guang Chen, et al. Genesis: Multimodal driving scene generation with spatio-temporal and cross-modal consistency.arXiv preprint arXiv:2506.07497,

    arXiv

  12. [12]

    Ltx-video: Realtime video latent diffusion

    Yoav HaCohen, Nisan Chiprut, Benny Brazowski, Daniel Shalem, Dudu Moshe, Eitan Richardson, Eran Levin, Guy Shiran, Nir Zabari, Ori Gordon, et al. Ltx-video: Realtime video latent diffusion. arXiv preprint arXiv:2501.00103,

    Pith/arXiv arXiv

  13. [13]

    Gaia-1: A generative world model for autonomous driving.arXiv preprint arXiv:2309.17080, 2023a

    Anthony Hu, Lloyd Russell, Hudson Yeo, Zak Murez, George Fedoseev, Alex Kendall, Jamie Shot- ton, and Gianluca Corrado. Gaia-1: A generative world model for autonomous driving.arXiv preprint arXiv:2309.17080, 2023a. Yihan Hu, Jiazhi Yang, Li Chen, Keyu Li, Chonghao Sima, Xizhou Zhu, Siqi Chai, Senyao Du, Tianwei Lin, Wenhai Wang, et al. Planning-oriented ...

    Pith/arXiv arXiv

  14. [14]

    Auto-encoding variational bayes.arXiv preprint arXiv:1312.6114,

    Diederik P Kingma and Max Welling. Auto-encoding variational bayes.arXiv preprint arXiv:1312.6114,

    Pith/arXiv arXiv

  15. [15]

    Uniscene: Unified occupancy-centric driving scene gener- ation

    15 Preprint Bohan Li, Jiazhe Guo, Hongsi Liu, Yingshuang Zou, Yikang Ding, Xiwu Chen, Hu Zhu, Feiyang Tan, Chi Zhang, Tiancai Wang, et al. Uniscene: Unified occupancy-centric driving scene gener- ation. InProceedings of the Computer Vision and Pattern Recognition Conference, pp. 11971– 11981, 2025a. Bohan Li, Zhuang Ma, Dalong Du, Baorui Peng, Zhujin Lian...

    Pith/arXiv arXiv

  16. [16]

    Depth anything 3: Recovering the visual space from any views.arXiv preprint arXiv:2511.10647,

    Haotong Lin, Sili Chen, Junhao Liew, Donny Y Chen, Zhenyu Li, Guang Shi, Jiashi Feng, and Bingyi Kang. Depth anything 3: Recovering the visual space from any views.arXiv preprint arXiv:2511.10647,

    Pith/arXiv arXiv

  17. [17]

    Flow matching for generative modeling.arXiv preprint arXiv:2210.02747,

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow matching for generative modeling.arXiv preprint arXiv:2210.02747,

    Pith/arXiv arXiv

  18. [18]

    Driveva: Video action models are zero-shot drivers.arXiv preprint arXiv:2604.04198,

    Mengmeng Liu, Diankun Zhang, Jiuming Liu, Jianfeng Cui, Hongwei Xie, Guang Chen, Hangjun Ye, Michael Ying Yang, Francesco Nex, and Hao Cheng. Driveva: Video action models are zero-shot drivers.arXiv preprint arXiv:2604.04198,

    Pith/arXiv arXiv

  19. [19]

    Flow straight and fast: Learning to generate and transfer data with rectified flow.arXiv preprint arXiv:2209.03003,

    Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow.arXiv preprint arXiv:2209.03003,

    Pith/arXiv arXiv

  20. [20]

    Recondreamer-rl: Enhancing reinforcement learning via diffusion-based scene reconstruction.arXiv preprint arXiv:2508.08170,

    Chaojun Ni, Guosheng Zhao, Xiaofeng Wang, Zheng Zhu, Wenkang Qin, Xinze Chen, Guanghong Jia, Guan Huang, and Wenjun Mei. Recondreamer-rl: Enhancing reinforcement learning via diffusion-based scene reconstruction.arXiv preprint arXiv:2508.08170,

    arXiv

  21. [21]

    Dinov2: Learning robust visual features without supervision.arXiv preprint arXiv:2304.07193,

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision.arXiv preprint arXiv:2304.07193,

    Pith/arXiv arXiv

  22. [22]

    Gaia-2: A controllable multi-view generative world model for autonomous driving.arXiv preprint arXiv:2503.20523,

    Lloyd Russell, Anthony Hu, Lorenzo Bertoni, George Fedoseev, Jamie Shotton, Elahe Arani, and Gianluca Corrado. Gaia-2: A controllable multi-view generative world model for autonomous driving.arXiv preprint arXiv:2503.20523,

    Pith/arXiv arXiv

  23. [23]

    Latent diffusion model without variational autoencoder.arXiv preprint arXiv:2510.15301,

    Minglei Shi, Haolin Wang, Wenzhao Zheng, Ziyang Yuan, Xiaoshi Wu, Xintao Wang, Pengfei Wan, Jie Zhou, and Jiwen Lu. Latent diffusion model without variational autoencoder.arXiv preprint arXiv:2510.15301,

    arXiv

  24. [24]

    What matters for representation alignment: Global information or spatial structure? arXiv preprint arXiv:2512.10794,

    Jaskirat Singh, Xingjian Leng, Zongze Wu, Liang Zheng, Richard Zhang, Eli Shechtman, and Sain- ing Xie. What matters for representation alignment: Global information or spatial structure? arXiv preprint arXiv:2512.10794,

    arXiv

  25. [25]

    Towards accurate generative models of video: A new metric & challenges

    Thomas Unterthiner, Sjoerd Van Steenkiste, Karol Kurach, Raphael Marinier, Marcin Michalski, and Sylvain Gelly. Towards accurate generative models of video: A new metric & challenges. arXiv preprint arXiv:1812.01717,

    Pith/arXiv arXiv

  26. [26]

    Mila: Multi-view intensive-fidelity long-term video generation world model for autonomous driving.arXiv preprint arXiv:2503.15875,

    Haiguang Wang, Daqi Liu, Hongwei Xie, Haisong Liu, Enhui Ma, Kaicheng Yu, Limin Wang, and Bing Wang. Mila: Multi-view intensive-fidelity long-term video generation world model for autonomous driving.arXiv preprint arXiv:2503.15875,

    arXiv

  27. [27]

    Beyond imitation: Learning safe end-to-end autonomous driving from hard negatives.arXiv preprint arXiv:2605.19771, 2026a

    Junli Wang, Zhihua Hua, Xueyi Liu, Zebin Xing, Haochen Tian, Kun Ma, Hangjun Ye, Guang Chen, Long Chen, and Qichao Zhang. Beyond imitation: Learning safe end-to-end autonomous driving from hard negatives.arXiv preprint arXiv:2605.19771, 2026a. Lening Wang, Wenzhao Zheng, Yilong Ren, Han Jiang, Zhiyong Cui, Haiyang Yu, and Jiwen Lu. Occsora: 4d occupancy g...

    Pith/arXiv arXiv

  28. [28]

    Latent-wam: Latent world action modeling for end-to-end autonomous driving.arXiv preprint arXiv:2603.24581, 2026b

    Linbo Wang, Yupeng Zheng, Qiang Chen, Shiwei Li, Yichen Zhang, Zebin Xing, Qichao Zhang, Xiang Li, Deheng Qian, Pengxuan Yang, et al. Latent-wam: Latent world action modeling for end-to-end autonomous driving.arXiv preprint arXiv:2603.24581, 2026b. Mengmeng Wang, Dengyang Jiang, Liuzhuozheng Li, Yucheng Lin, Guojiang Shen, Xiangjie Kong, Yong Liu, Guang D...

    arXiv

  29. [29]

    Resim: Reliable world simulation for autonomous driving.arXiv preprint arXiv:2506.09981,

    Jiazhi Yang, Kashyap Chitta, Shenyuan Gao, Long Chen, Yuqian Shao, Xiaosong Jia, Hongyang Li, Andreas Geiger, Xiangyu Yue, and Li Chen. Resim: Reliable world simulation for autonomous driving.arXiv preprint arXiv:2506.09981,

    Pith/arXiv arXiv

  30. [30]

    Representation alignment for generation: Training diffusion transformers is easier than you think.arXiv preprint arXiv:2410.06940,

    Sihyun Yu, Sangkyung Kwak, Huiwon Jang, Jongheon Jeong, Jonathan Huang, Jinwoo Shin, and Saining Xie. Representation alignment for generation: Training diffusion transformers is easier than you think.arXiv preprint arXiv:2410.06940,

    Pith/arXiv arXiv

  31. [31]

    Epona: Autoregressive diffusion world model for autonomous driving.arXiv preprint arXiv:2506.24113,

    Kaiwen Zhang, Zhenyu Tang, Xiaotao Hu, Xingang Pan, Xiaoyang Guo, Yuan Liu, Jingwei Huang, Li Yuan, Qian Zhang, Xiao-Xiao Long, et al. Epona: Autoregressive diffusion world model for autonomous driving.arXiv preprint arXiv:2506.24113,

    arXiv

  32. [32]

    Diffusion transformers with repre- sentation autoencoders.arXiv preprint arXiv:2510.11690,

    Boyang Zheng, Nanye Ma, Shengbang Tong, and Saining Xie. Diffusion transformers with repre- sentation autoencoders.arXiv preprint arXiv:2510.11690,

    Pith/arXiv arXiv

  33. [33]

    Occ- world: Learning a 3d occupancy world model for autonomous driving

    Wenzhao Zheng, Weiliang Chen, Yuanhui Huang, Borui Zhang, Yueqi Duan, and Jiwen Lu. Occ- world: Learning a 3d occupancy world model for autonomous driving. InEuropean conference on computer vision, pp. 55–72. Springer, 2024a. Wenzhao Zheng, Ruiqi Song, Xianda Guo, Chenming Zhang, and Long Chen. Genad: Generative end-to-end autonomous driving.arXiv preprin...

    arXiv

  34. [34]

    Guiding a diffusion transformer with the internal dynamics of itself

    18 Preprint Xingyu Zhou, Qifan Li, Xiaobin Hu, Hai Chen, and Shuhang Gu. Guiding a diffusion transformer with the internal dynamics of itself. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 11536–11545, 2026a. Yang Zhou, Xiaofeng Wang, Hao Shao, Letian Wang, Guosheng Zhao, Jiangnan Shao, Jiagang Zhu, Tingdong Yu, ...

    arXiv