arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2610.18697cs.ROcs.AIcs.CVcs.LG
收藏

UniW(A)M:面向世界模型与世界-动作模型的统一混合专家框架

UniW(A)M: A Unified Mixture-of-Experts Framework for World and World-Action Models

Tsinghua University(清华大学);Zhongguancun Academy(中关村科学院);East China Normal University(华东师范大学);Manifold AI(流形人工智能)

Jingbo Xu, Jianjie Fang, Zhuohang Li, Zihao Song, Zhaolu Wang, Xvyuan Liu, Xin Zhang, Haisheng Su, Yinzhou Tang, Chen Gao, Wei Wu, Yong Li

AI总结:

UniW(A)M是基于扩散Transformer的混合专家框架,统一动作条件世界建模与WAM变体,经多数据集评估,其真实世界操作成功率达88.3%,OOD成功率提升至76.7%,性能优于基线。

摘要:

世界模型(WM)与世界-动作模型(WAM)分别聚焦于未来世界预测与可执行动作生成。尽管这些模型会从具身轨迹中学习共享的物理动力学与交互语义,但它们通常是孤立训练的,这限制了预测与控制之间的知识复用。因此核心挑战在于开发一种统一学习框架,该框架在共享这些动力学的同时,保留每个范式所需的独特建模能力。我们提出UniW(A)M,一种基于扩散Transformer构建的世界-(动作)模型混合专家框架,它统一了动作条件下的世界建模与多样化WAM变体。共享专家捕获通用动力学,而原子能力专家在各接口间复用,使每个接口保留自身的条件通路。部署时,插件式自适应接口路由器(PAIR)会从目标任务块场景的小子集中学习,为每个执行块选择合适的WAM接口,在冻结主干网络的同时平衡任务成功率与执行效率。我们在LIBERO、RoboTwin 2.0及三个真实世界操作任务上评估UniW(A)M,其平均真实世界成功率达88.3%,而最强基线为76.7%;在WorldArena上,它在全部三项任务中均领先动作条件下的世界建模。AC-WM协同训练对每个动作接口的动作生成均有贡献,且PAIR将分布外(OOD)成功率从63.3%提升至76.7%。

英文摘要:

World models (WMs) and world-action models (WAMs) focus on future world prediction and executable action generation, respectively. Although these models learn shared physical dynamics and interaction semantics from embodied trajectories, they are typically trained in isolation, which restricts knowledge reuse between prediction and control. The central challenge is therefore to develop a unified learning framework that shares these dynamics while preserving the distinct modeling capabilities required by each paradigm. We propose UniW(A)M, a mixture-of-experts world-(action) model built upon diffusion Transformers that unifies action-conditioned world modeling and diverse WAM variants. Shared experts capture common dynamics, while atomic capability experts are reused across interfaces so that each interface keeps its own conditional pathway. At deployment, a plug-in adaptive interface router (PAIR) learns from a small subset of target-task chunk scenarios to select a suitable WAM interface for each execution chunk, balancing task success and execution efficiency while keeping the backbone frozen. We evaluate UniW(A)M on LIBERO, RoboTwin 2.0, and three real-world manipulation tasks, where it achieves 88.3% average real-world success versus 76.7% for the strongest baseline. On WorldArena, it leads action-conditioned world modeling on all three tasks. AC-WM co-training contributes to action generation on every action interface and PAIR raises OOD success from 63.3% to 76.7%.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑