Core Viewpoint - The 2025 World Humanoid Robot Games successfully concluded on August 17, featuring 280 teams from 16 countries competing in 26 events over three days, marking the first comprehensive competition for humanoid robots globally [1]. Group 1: Competition Overview - The event included a total of 487 matches across various categories such as competitive events, performance events, scenario events, and peripheral events [1]. - Notable performances included the championship in the 100 meters by the team "Tian Gong" with the robot "Tian Gong Ultra" achieving a time of over 21.50 seconds [1]. Group 2: Event Results - In the 400 meters, "Ling Yi Technology" with the robot "Yu Shu H1" secured the championship with a time of 6 minutes and 34 seconds [2]. - The 1500 meters saw "Yu Shu Technology" win the championship with a time of 1 minute and 48.78 seconds [2]. - The 4x100 meters relay was won by "Gao Yi Technology" with a time of 33.71 seconds using the robot "Yu Shu G1" [2]. Group 3: Special Awards and Recognitions - The "Best Robot" award was given to the "Wolf Pack Team" [4]. - The "Popularity Award" was awarded to "Chiweng" with the robot "Yu Shu G1" [4]. - The "Best Interaction Award" in the table tennis category went to the "Exploration Extreme Team" [5]. Group 4: Industry Participants - Various companies participated in the event, including "Yu Shu," "Tian Gong," and "Ling Yi Technology," showcasing advancements in humanoid robotics [12][13]. - The event highlighted the growing interest and investment in humanoid robotics, with numerous teams representing educational institutions and tech companies [12][13].

人形机器人与具身智能是当之无愧的时代风口，在刚刚落下帷幕的世界人形机器人运动会上，宇树科技、北京 人形机器人创新中心、松延动力、星动纪元等人形机器人头部企业齐聚赛场展开激烈比拼， 16 个国家的 280 支参赛队伍带来 500 余台人形机器人同台竞技，数十家企业展示惊人绝活， 26 位金牌得主创下了多项 人形机器人界的世界纪录。 以赛为媒，这场人形机器人盛会，全面检验了人形机器人在运动控制、智能决策、环境感知、场景落地等方面 的能力，无疑是对人形机器人综合实力的一次大阅兵。 而紧随其后的 11 月，我们即将迎来下一场人形机器人与具身智能的关键赛事——第二届中关村具身智能机 器人应用大赛。 不同于本次运动会设置竞技赛、表演赛、场景赛、外围赛四大类赛事，第二届中关村具身智能机器人应用大赛 聚焦落地 ，希望真正展现一个更加切实的主题： 随着人形机器人与具身智能技术迭代与应用落地的双重突 破，到底何时、何地、以何种方式落场景，助力人类生产生活？ 我们相信，随着这场赛事拉开帷幕，有望让行业更进一步，真刀真枪展示机器人技术实际落地的可能性， 为 投资机构提供真实考核样本，为下游客户提供采购选型参考，成为具身智能机器人技术 ...

⼀、 端到端离不开的轨迹预测 端到端量产以来，很多规划控制和轨迹预测放同学都很焦虑，都想着转行做感知模型，怕自己过两年失业。但 这一年多以来，据自动驾驶之心了解，一段式端到端上车的并不多，很多公司依然沿用二段式端到端或者模块 化的方法，轨迹预测或者说联合预测仍然是量产使用最多的算法， 依然是许多公司和机构研究的热点。但更 进一步，其实轨迹预测的模型和感知模型融合在一起训练，其实就是所谓的端到端，因此 相关的会议和期刊 依然有较大量的工作产出。 自动驾驶之心针对目前比较火的基于扩散模型的多智能体轨迹预测方法研究展开了首个1v6小班课！本课题聚 焦于"基于扩散模型的多智能体轨迹预测方法"。多智能体轨迹预测旨在根据多个交互主体的历史轨迹，预测其 未来运动轨迹，这在自动驾驶、智能监控和机器人导航等场景中至关重要。然而，由于人的行为具有不确定性 和多模态性，预测任务十分困难。传统方法通常依赖循环神经网络、卷积网络或图神经网络建模社会交互，而 生成模型（如GAN和CVAE）虽然可以模拟多模态分布，但效率不高。 扩散模型是一类通过逐步去噪实现复杂分布生成的新型模型，近年来在图像生成等领域取得了重大突破。研究 者发现将扩散模 ...

编辑丨周上祺 视频 丨 章启良 8月17日下午，跟随主题曲《光年》的旋律，26个机器人冠军集体站在舞台中间向观众席挥手 致意，2025年世界人形机器人运动会落下帷幕。为期三天的赛事期间，各种亮点片段接连出 圈：自由搏击决赛后机器人"碰瓷"躺倒、1500米赛跑中机器人"撞人逃逸"......人形机器人状况 百出，摔倒、撞人已经不是罕见场面，与此同时，多个人形机器人世界纪录在这里诞生，广受 关注的宇树科技在田径比赛中连续夺魁，在百米障碍赛中把"头"摔掉的天工随后斩获百米冠 军，这场发挥人形机器人极限的比赛，正见证人形机器人如何跌跌撞撞地奔向未来。 魔法原子研发副总裁陈春玉对21世纪经济报道记者表示，"人都会摔倒，更别说处于发展初期 阶段的机器人。重要的是机器人摔倒后能否自主运动起来或恢复后继续跑。只有在摔倒中，大 家才能看到机器人一年比一年进步。" 光谷东智副总经理楼凯齐的想法是"感同身受"。他对21世纪经济报道记者表示，"机器人是比 较精密的设备，即使在比赛前一刻都认为做了相对充分的准备，但现场仍可能出现突发状 况。"楼凯齐表示，"机器人行业或技术的发展有一个过程，不能奢望机器人从一开始就是全能 全知的。它就像 ...

转自：新华财经 "比人类神经的反应极限还要快20ms，人眼几乎察觉不到延迟。"当虹科技技术人员在活动现场介绍， 超远距离远程操控技术真正打破了机器边界。 此外，通信链路稳定也十分重要。依托当虹自研的BlackEye多模态视听大模型，BlackEye Vision系统可 根据带宽情况，将高清视频进行10-100倍压缩，并结合AI画质增强播放技术，在超窄带宽环境下依旧可 以回传清晰画面。即便是在大型赛事、演唱会这种人群密集、网络拥堵的环境，机器人依旧能"传得 回、看得清"。 如果遇到完全断网的紧急情况，系统还能自动切换至卫星通信，实现4G、5G、Wi-Fi、卫星全网络自适 应，真正实现"无需组网""不挑网络"。正如一位民警所说："无论是日常街面巡逻，还是极端条件下的 搜救工作，都能保证机器人听指挥、不断联。" BlackEye Vision还有多模态处理能力，针对一些特定的搜寻任务，不仅要回传高清视频，还要同步回传 三维图像、音频等多种数据，而且必须毫秒级送达。BlackEye Vision同时处理视频、音频、激光雷达点 云、图像、信令、文本等多模态数据，让操控员不仅"看得清"，还能根据多种信息增加"临场感"，让 ...

Core Viewpoint - The article presents UAV-ON, the first large-scale benchmark for open-world object goal navigation with aerial agents, defining over 11,000 navigation tasks across 14 high-fidelity outdoor scenes, emphasizing the need for drones to navigate complex environments autonomously [2][5]. Group 1: Research Background - UAV-ON aims to enhance drone navigation capabilities in diverse real-world environments, addressing the limitations of existing navigation studies that rely heavily on detailed language instructions [2]. - The benchmark includes a set of baseline strategies for drone navigation, such as random strategies, CLIP-based semantic heuristic algorithms, and the proposed aerial object navigation agent (AOA) [2]. Group 2: Environment and Task Definition - The UAV-ON benchmark defines an instance-level object navigation task where drones must navigate to target objects based on semantic instructions [5]. - Drones are equipped with multi-view RGB-D cameras and rely solely on their perception for navigation, without any global positioning signals [6][12]. Group 3: Action Space and Success Conditions - The action space for drones includes parameterized movements such as translation, rotation, and stopping, with each action linked to continuous control parameters [11][14]. - A successful episode is defined as the drone being within a specified distance from the target object at the end of the episode [7]. Group 4: Dataset Analysis and Environment Diversity - The UAV-ON dataset comprises 14 high-fidelity outdoor environments, featuring a variety of natural and man-made landscapes, with a total of 1,270 unique target objects distributed across approximately 9 million square units [15]. - The training set includes 10 diverse outdoor environments generating 10,000 navigation episodes, while the test set consists of 1,000 episodes to evaluate generalization capabilities [15]. Group 5: Experimental Results and Baseline Methods - Various baseline methods were tested, including Random, CLIP-H, AOA-F, and AOA-V, with AOA-V showing the best performance in Oracle success rate but lower in success rate and SPL [16][17]. - The results indicate that all methods have a collision rate exceeding 30%, highlighting a significant gap between current navigation strategies and the safety requirements for real-world drone operations [20]. Group 6: Conclusion and Future Work - UAV-ON serves as a comprehensive benchmark for semantic reasoning, obstacle perception, and target localization challenges in drone navigation [24]. - Future research will focus on enhancing multi-modal perception, prompt-based control, and developing safer, more reliable navigation strategies for autonomous drone operations in complex environments [24].

Core Insights - The 2025 World Humanoid Robot Games showcased the progress of humanoid robots, with 280 teams and over 500 robots competing, highlighting the industry's advancements and the importance of iterative testing in technology development [1][2] Group 1: Event Overview - The event featured three categories: competitive events, performance events, and scenario events, assessing the maturity of embodied intelligence technology from multiple dimensions [2] - Humanoid robots demonstrated various capabilities, including sorting materials, cleaning services, dancing, and heavy lifting, emphasizing the importance of real-world applications in driving research and development [2] Group 2: Technological Advancements - The "Walker No. 2" robot from Henan achieved first place in its group for the 1500-meter event, showcasing the effectiveness of interdisciplinary support in hardware, data, algorithms, new materials, and bionic technology [3] - The tendon-driven design of "Walker No. 2" reduced its weight by 40% compared to similar-sized robots, enhancing its agility and energy efficiency through optimized algorithms [3] Group 3: Industry Implications - The event brought together various innovative teams from universities, research institutions, and leading humanoid robot companies, fostering collaboration between academia and industry [3] - Henan's complete industrial chain and strategic emerging industries provide a solid foundation for robot manufacturing, with significant market potential and diverse application scenarios driving industry growth [3]

Group 1 - The first "World Humanoid Robot Games" took place in Beijing, featuring 280 teams from 16 countries, including the U.S. [1] - The competition included various events such as dance battles, martial arts, track and field events (400-meter and 1500-meter races, long jump), and a soccer tournament [1] - Robots used in the competition were manufactured by Chinese companies like Unitree and Booster [1] Group 2 - Guo Tong, a programmer, expressed that robots have stronger joints and core strength, predicting they could replace human soccer stars by 2050 [2] - Unitree, based in Hangzhou, won multiple medals and is seen as a competitor to Tesla's Optimus [2] - Other notable companies included Beijing-based X-Humanoid and Hong Kong-listed Shenzhen firm UBTECH, which also performed impressively in the competition [2]

Robots playing soccer. Maybe not. China's first World Humanoid Robot Games is just the latest in a series of events and programs aimed at promoting humanoid robot technology.The Chinese government is hosting this 3-day event where humanoid athletes show off their skills in martial arts, the 400 meter race. Is it just me or does that guy look tired. and rounds in a soccer tournament.Guo, who programmed one of the futuristic footballers, sees robots replacing his idol, Ronaldo. 2050, he replies. Chinese Olymp ...

Core Insights - The 2025 World Humanoid Robot Games concluded successfully in Beijing, showcasing the integration of technology and sports with 280 teams from 16 countries competing in 26 events, resulting in 26 gold medals awarded [1][2]. Group 1: Technological Advancements - The "Jushen Tiangong Ultra" robot, representing the Beijing Humanoid Robot Innovation Center, won the 100m race autonomously, marking a significant milestone as the first robot to achieve this feat [1][2]. - The robot utilized a fully autonomous navigation system, a significant upgrade from its previous semi-autonomous capabilities, allowing it to understand complex environments and plan movements in real-time [2][3]. - The introduction of the "Wisdom Open Platform" by the Beijing Humanoid Robot Innovation Center aims to enhance robots' decision-making and execution capabilities, transitioning from single-task execution to complex autonomous operations [3][4]. Group 2: Competitive Performance - The "Jushen Tiangong Ultra" demonstrated exceptional stability and control during races, successfully navigating without human intervention, which was validated through its performance in various competitions [4][5]. - Hangzhou Yushu Technology Co., Ltd. achieved remarkable results, winning four gold medals and setting a record speed of 4.78 meters per second with its humanoid robots [4][5]. - The competition served as a platform for testing and showcasing the practical applications of humanoid robots in real-world scenarios, such as material handling and service tasks [6]. Group 3: Industry Implications - The event highlighted the potential for humanoid robots to integrate into everyday life, serving as a bridge to various industries and enhancing productivity across sectors [6]. - The competition provided insights into the reliability and precision of robotic components, guiding future improvements to meet real-world application demands [6]. - The success of the event is expected to attract investment and support for the growth of the humanoid robotics industry, fostering innovation and development [6].