InternW0-Δ is a unified World Action Model that integrates pretrained visual dynamics, scene semantics, 4D geometry, and motion priors within a Mixture-of-Transformers framework to generate robot actions. It leverages a frozen VLM for semantic guidance, a 4D foundation model for geometric priors, and introduces Causal Imprint to learn future-relevant scene changes without future-video rollout. The model is pretrained on a newly curated 20K‑hour heterogeneous corpus of robot and human demonstrations, achieving superior performance on simulation benchmarks and real‑robot platforms.
By Xingyu Miao, Zizun Li, Baole Fang, Kaiwen Song, Tenghui Wang, Hanxue Zhang, Yating Wang, Xudong Li, Yuping He, Xueyuan Wei, Chao Gao, Xijie Yang, Yingxiang Xu, Kerui Ren, Wenqi Guo, Jianjun Zhou, Xinzhe Wang, Weiguang Zhao, Ni Yang, Zetao Cai, Yufei Xue, Hengjie Li, Zeyu He, Yuanzhen Zhou, Rong Fu, Jianyang Zhang, Siwei Cui, Fuxian Huang, Yunsong Zhou, Xing Gao, Yifei Yao, Qiaojun Yu, Kailin Li, Ming Zhou, Mu Huang, Xinyue Li, Wenze Cui, Bingqi Jiang, Xueyue Zhu, Junting Dong, Haoyu Guo, Tao Lu, Mulin Yu, Bowen Zhou, Bin Zhao, Tianfan Xue, Weinan Zhang, Chunhua Shen
The paper introduces NAVA-WAM, a method that learns action priors directly from observation-only videos to pretrain action policies for robots. It uses a two-stage training process: first, it pretrains on videos with future‑video flow‑matching supervision to learn action‑relevant priors, and second, it fine‑tunes with action‑labeled demonstrations for robot control. Experiments show that NAVA‑WAM outperforms previous methods in both in‑distribution and out‑of‑distribution scenarios, achieving strong action‑label efficiency and real‑robot generalization.
By Zhaochong An, Fei Zhang, Menglin Jia, Duncan Frost, Zijian Zhou, Yikai Wang, Xudong Wang, Aditya Patel, Belinda Zeng, Tao Xiang, Serge Belongie, Amir Bar, Sen He
Being‑M0.7 is a latent world‑action model that learns from a large mixed‑modality human dataset (over 10,000 hours of video‑only, motion‑only, and paired video‑motion data) to predict future latent visual states and motion. The model is pre‑trained on human data, adapted to robot viewpoints during mid‑training, and then combined with an action expert via gated cross‑attention during post‑training to produce executable whole‑body commands. In experiments, Being‑M0.7 outperforms baselines on the SIMPLE benchmark and matches the best baseline on real‑world Unitree G1 loco‑manipulation tasks.
By Junpeng Yue, Boyuan Li, Yuxuan Wang, Zepeng Wang, Yuhui Fu, Feiyang Xie, Yu Zhang, Jing Zhang, Xianqi Zhang, Weibo Li, Xiaofei Zheng, Yuming Fang, Jiangxing Wang, Zongqing Lu
arXiv:2609.38059v1 Announce Type: cross
Abstract: Real-world robot learning is constrained by the cost of collecting experience and evaluating candidate behaviors. Video generation models offer a sca...
By Shenghe Zheng, Wenbo Li, Jiyao Zhang, Bin Xia, Haoyang Huang, Nan Duan, Jiaya Jia
arXiv:2610.27851v1 Announce Type: cross
Abstract: This paper shows that JEPA-style latent world models (LWMs) can learn across diverse domains and data sources by adopting the language-conditioned in...
By Junha Song, Geonmo Gu, Jeongeun Park, Heejeong Nam, Lucas Maes, Byeongho Heo, Dongyoon Han, Sangdoo Yun
KnowDemo is a framework that generates diverse robot demonstrations from human videos by leveraging structured manipulation knowledge. It uses a vision‑language model to extract task requirements and permissible execution variations, then resolves these against target‑scene entities to guide candidate generation and screening before motion planning. The resulting demonstrations feature multimodal behavior, alternative contact strategies, and valid subtask orders, and have been shown to improve planning success and enable sim‑to‑real policy transfer across three tasks.
By Zhiyuan Gao, Yanxiang Zhan, Mohammad Khoshnazar, Jeroen Sch\"afer, Michael Beetz
PDRVLA is a Vision‑Language‑Action model that incorporates a lightweight Physical Dynamics Module to extract physical tokens from vision‑language features, enabling action generation without online video synthesis. The module predicts future RGB, depth, and optical‑flow features, guided by auxiliary supervision from a frozen pretrained video world model. Trained solely on publicly available data augmented with human‑to‑robot synthesis, subtask annotation, and embodied chain‑of‑thought (ECoT) annotation, PDRVLA achieves high success rates across four simulation benchmarks, with ablations demonstrating the benefits of world‑model supervision and various conditioning strategies.
By Hanqing Wang, Zhaotian Cai, Ruichuan An, Bohan Zeng, Tao Zhang, Yilin Chen, Xu Jiang, Kaiquan Chen, Chongyang Wu, Haoran Li, Fan Yang, Mingan Lin, Wentao Zhang, Zenan Zhou, Youzheng Wu
Video Prediction Policy 2 (VPP2) is a new world action model that improves zero‑shot generalization for both video prediction and action generation. It achieves this by pretraining a large, diverse manipulation video dataset with event‑level supervision, then distilling the model into a single‑step visual planner and adding a mixture‑of‑transformers action module. Experiments show VPP2 outperforms leading baselines on open‑ended video prediction, real‑world zero‑shot manipulation, and several challenging benchmarks.
By Yanjiang Guo, Haodong Yan, Zhide Zhong, Zhongru Zhang, Qingyuan Yang, Qingzhou Lu, Xiaoyu Chen, Yen-Jen Wang, Shuying Deng, Chenghan Yang, Puzhen Yuan, Chenxin Liu, Tun Ban, Xiang Zhu, Yichen Liu, Kun Feng, Haoang Li, Jianyu Chen
CLAP is a cross-embodiment framework for action‑conditioned video generation that can be trained on diverse internet‑scale videos from both humans and robots. It reconciles different action spaces—end‑effector poses, language instructions, and latent actions—using a curriculum that first learns physics priors from unlabeled video and then grounds them in real‑world action spaces for zero‑shot deployment. The resulting models match or exceed state‑of‑the‑art single‑embodiment models in challenging environments and support few‑shot adaptation across a wide range of robot morphologies.
By Kechen Liu, Ola Shorinwa
Generalizable robot manipulation requires policies that can anticipate how visual scenes evolve while executing language instructions. While recent Vision-Language-Action models benefit from large-scale pretraining, their predominantly static pretraining objectives provide limited supervision for physical dynamics and temporal causality, leaving control-relevant knowledge to be learned from downstream robot demonstrations.
WorldFlow is a generative world model that jointly predicts future video frames and query-based 3D trajectories using a unified Mixture-of-Transformers backbone. It represents 3D motion as continuous displacement tokens aligned with the video grid and optimizes a rectified-flow objective to couple explicit geometric dynamics with visual evolution. Trained on 1.44 million heterogeneous samples and evaluated on WorldFlowBench, it achieves the lowest average and final displacement errors among learning-based methods, notably outperforming baselines on spatial reasoning tasks.
By Yuhao Lu, Jiongxu Chen, Hexiang Wei, Jiajin Wu, Juelin Li, Mingjian Liang, Fan Yang, Zhuowei Liang, Zihao Liu, Zhennan Wang, Sheng Wang, Lin Shao
Dynin‑Robotics introduces an omnimodal masked‑diffusion backbone, Dynin‑Omni, that jointly represents language, visual observations, goals, and actions as discrete tokens. By conditioning on different spans, the same model learns action prediction, next‑observation prediction, goal‑state prediction, and trajectory‑to‑instruction reconstruction, enabling test‑time scaling through goal prediction and action‑candidate evaluation. The system, pretrained on 1.33 million trajectories from 48 Open X‑Embodiment datasets, achieves competitive performance on LIBERO, zero‑shot LIBERO‑Plus, and a 78.4 % success rate on a Franka Research 3 robot, while a block‑parallel implementation speeds up action decoding by up to 29.2×.
By Hoeun Lee, Jaeik Kim, Jusang Oh, Jinhyeok Kim, Geon Choi, Hyeonggeun Kim, Jaeyoung Do