Core Insights - The rapid evolution of robotic movement capabilities is highlighted, but understanding complex tasks remains challenging for robots [1] - The introduction of the "WoW" embodied world model by a Chinese research team represents a significant advancement in robotic intelligence [1] Group 1: Technological Advancements - The "WoW" embodied world model allows robots to simulate human-like thinking and decision-making, generating future prediction videos that align with physical laws [5] - The model enables robots to connect imagined movements with real-world execution, enhancing their interaction with the environment [5] Group 2: Data Collection and Training - The research team has collected millions of real interaction data points to ensure the world model can operate effectively in diverse real-world scenarios [8] - The model is designed to adapt to various types of robots, including humanoid and robotic arms, and can be applied across multiple settings such as homes, supermarkets, and logistics [10] Group 3: Open Access and Applications - The "WoW" model is open to global researchers and developers, facilitating broader applications and innovations in robotics [10] - It can accurately simulate extreme scenarios, such as water spilling on a computer, providing crucial data for training that is difficult to obtain through real-world testing [10]

当前，机器人的运动能力正在迅速进化，有些已经可以很轻松地完成后空翻、跑步等动作。但是，相比完成一个后空翻，让它"理解"面前的水杯为什么倒满 水后会洒出来，就更难了。 日前，我国科研团队开源出一个名叫WoW（读作"哇哦"）的具身世界模型，它让机器人可以像人类一样，进化出更好的想象力与执行力。怎么理解具身世 界模型？它如何让机器人更聪明？ 总台记者 袁嘉忆：在北京人形机器人创新中心，各种形态的机器人本体正在进行具身智能数据采集和动作模型训练。这台"天工"机器人正在自主地1∶1复刻 视频中的动作姿态，而这个视频就是机器人在行动之前"想象出来"的预演画面，可以用来指导它与真实世界的交互。这样从想象预演到动作执行的"知行合 一"的能力，依托的就是由科研团队自主研发的具身世界模型。 WoW具身世界模型项目负责人 池晓威：机器人在推倒这个杯子的时候，我们人类会本能地预测到这个杯子要飞出去、要倒掉，所以我会去进行这个接杯子 的动作。世界模型本质上就是AI模拟人类思考和决策的时候，去进行想象和预测的这样一个模型，它需要去生成符合物理规律的未来预测视频，帮助机器 人真的去把想象当中的运动轨迹变到真实世界当中执行出来，从而把想象跟 ...

Core Insights - 2024 is anticipated to be the year of mass production for humanoid robots, while 2025 is expected to mark their commercialization [1] - There has been a surge in humanoid robot orders in the latter half of this year, with significant contracts awarded to companies like UBTECH and ZhiYuan Robotics [1][2] Group 1: Market Trends - Humanoid robots are increasingly being commercialized in various sectors, including data collection, automotive manufacturing, and 3C manufacturing [2][4] - The industry is focusing on eight major application scenarios: industrial manufacturing, logistics sorting, security inspection, commercial cleaning, data collection training, scientific research education, entertainment, and reception [4][6] Group 2: Implementation Challenges - The path to commercialization is seen as a gradual process, starting from scenarios that do not require physical interaction to more complex environments [3][4] - The difficulty of implementing humanoid robots in home settings is highlighted, with challenges related to cost, safety, and task complexity [7][8] Group 3: Technological Barriers - Key technological barriers include insufficient performance of core hardware, such as sensors, which are crucial for the effective operation of humanoid robots [8][9] - The need for advanced tactile sensors and high degrees of freedom in robotic hands is emphasized for achieving fine manipulation [8][9] Group 4: Industry Collaboration - The industry is promoting collaborative innovation through open-source large models and datasets, which are essential for training humanoid robots in diverse scenarios [9] - Companies are developing specific small models tailored to particular applications to enhance performance beyond a basic level [9]

Core Viewpoint - The Robot ETF (562500) is experiencing a strong performance, leading its category with a 0.81% increase, indicating robust investor interest and market activity in the robotics sector [1][2]. Group 1: ETF Performance - The Robot ETF (562500) has a market size exceeding 20 billion, making it the only ETF of its kind in the market with such scale, covering various segments including humanoid robots, industrial robots, and service robots [2]. - As of 10:08 AM today, the ETF's trading volume reached 291 million, with a volume ratio of 1.26, indicating active trading [1]. - In the current trading session, 52 stocks within the ETF rose while 21 fell, showcasing significant structural differentiation among holdings [1]. Group 2: Market Trends and Developments - Recent developments include the Beijing Humanoid Robot Innovation Center open-sourcing parts of the WoW embodied world model, which lowers the entry barrier for global researchers and accelerates the integration of embodied intelligent robots into daily life [1]. - Historical trends suggest that each technological wave, such as the rise of general artificial intelligence since 2020, tends to create new smart terminals, with humanoid robots poised to enter a golden development period similar to that of new energy vehicles in the coming years [1].

本报讯 （记者贾丽）继DeepSeek在大语言模型开源促进行业发展后，北京人形机器人创新中心（以下 简称"北京人形"）再次打破边界并开源了全新的世界模型架构，提出了一个让机器人真正"看见、理解 并行动于世界"的具身世界模型——WoW（World-Omniscient World Model），旨在帮助具身智能机器人 快速学习掌握各项技能，助力行业打造"最好用"的机器人。一经发布，该模型便受到学术界和产业界的 广泛关注。 同时，WoW不是在记忆训练场景，而是在学习"物理规律的抽象本质"，具备跨机器人形态泛化、任务 泛化、场景泛化全方位能力，这类"视觉+物理"的泛化能力，是通向具身智能的关键指标。 WoW具身世界模型遵循SOPHIA范式，让模型越看越准，越生成越真实。SOPHIA自反范式是指，业内 首次提出SOPHIA框架，让世界模型"自己教自己"。 据了解，WoW具身世界模型可以实现视频生成和机器人动作闭环，意味着AI不再停留在"想象中"，而 能真正"动手"去验证自己的理解，标志着真正实现从生成到执行的跨越。 北京人形提出了专测"物理一致性与因果推理"的新基准WoWBench，也是全球首个针对具身世界模型的 综 ...

Core Insights - The article discusses the advancements in robotics, particularly focusing on the new embodied world model called WoW (World-Omniscient World Model) developed by the Beijing Humanoid Robot Innovation Center, which allows robots to understand and interact with the physical world more effectively [2][4][51]. Group 1: Model Development and Features - WoW represents a significant upgrade in visual models, integrating vision, action, physical perception, and reasoning into a unified framework, enabling robots to learn the physical laws of the world through interaction [4][5]. - The model has gained widespread attention from both academia and industry, with endorsements from notable organizations like Huggingface and Stanford, indicating its leading position in the field of embodied world models [3][4]. - WoW consists of four core components that allow it to predict future scenarios, deduce physical evolution, and reconstruct dynamic causal chains based on historical data [10][12]. Group 2: Performance and Evaluation - WoW has shown superior performance in simulating robotic operations, particularly in physical reasoning and temporal consistency compared to its predecessor, Sora 2 [5][12]. - The model was trained on a dataset of 8 million robot interaction trajectories, refining it down to 2 million high-quality training samples, which significantly improved its physical consistency and generative stability as the model size increased from 1.3 billion to 14 billion parameters [12][36]. - WoWBench, a comprehensive benchmark for evaluating embodied world models, assesses capabilities across perception, reasoning, decision-making, and execution, ensuring alignment with human cognitive performance [29][31]. Group 3: Practical Applications and Future Prospects - The open-source nature of WoW allows global researchers to replicate results and further develop applications, lowering the entry barrier for research in world models and accelerating the integration of embodied intelligent robots into various sectors [42][43]. - WoW's capabilities enable it to generate synthetic samples from limited real data, facilitating a self-cycling process of "imagination - generation - re-labeling - transfer," enhancing the AI's ability to perform complex tasks in real-world environments [53][56]. - The advancements demonstrated by WoW, including its success in various robotic competitions, highlight its potential to redefine the landscape of humanoid robotics and embodied intelligence [56][57].

机器之心发布 机器之心编辑部 如果说 GPT 系列让 AI 理解语言，Sora 系列让 AI 生成视觉世界，那么 WoW 正在尝试让 AI 建模物理世界。 在「具身智能」与「世界模型」成为新一轮 AI 竞赛关键词的当下，来自 北京人形机器人创新中心、北京大学多媒体信息处理国家重点实验室、香港科技大 学的中国团队 开源了全新的世界模型架构。 该团队提出了一个让机器真正 "看见、理解并行动于世界" 的世界模型 —— WoW（World-Omniscient World Model, 意图让 AI 学会 "做" —— 通过身 体与世界互动来学习因果与物理，致力于助力行业打造 "最好用" 的具身智能机器人。 一经发布，受到学术界产业界关注关注，其中 Huggingface 留言："Excellent work" 催更开源，斯坦福具身智能大佬，PI 创始人 Chelsea Finn & 清华 合作文章引用 WoW 具身世界模型技术报告。 不是看图说话，而是动手理解世界：WoW 模型揭秘 真正具备物理理解的世界模型，必须建立在与现实世界广泛且因果丰富的交互与反馈之上。 人类通过与世界的主动互动，逐渐发展出对 直觉物理 的 ...