科技云报到原创。 从ChatGPT的语言魔法到DALL·E的图像创造，虚拟世界的AI已无处不在。但当我们把目光转向现实世界，一个新名词正引发热议：具身智能。 具身智能被誉为"AI的下一站"，让机器不仅会"思考"，还能"行动"——从工厂的搬运机器人到养老院的陪伴助手，甚至是太空探索的机械先锋。马斯克宣 称，特斯拉的Optimus将是"通用智能的载体"；谷歌DeepMind则押注机器人来破解物理世界的难题。 然而，现实远比愿景骨感。2025年，具身智能的商用化仍步履蹒跚：成本高企、效率低下、数据瓶颈频现。波士顿动力Spot在巡检场景表现不俗，但价格高 达20万元；优必选的Walker搬运箱子耗时数分钟，远逊于人类工人。 如今，技术已将其推至产业变革的商业化临界点。但我们仍必须回答三个关键问题：需要怎样的颠覆性创新，才能赋予具身智能真正的泛化行动能力？又如 何跨越从"技术可行"到"商业成功"的鸿沟，找到其不可替代的价值锚点？当"数字认知"的潜力被空前释放，我们如何将AI其转化为"物理世界"的真实生产 力？ 具身智能的价值最终要靠场景落地来体现，目前多个传统行业已经尝到了技术红利。制造业、医疗健康、零售业、能源行业、 ...

马斯克刷屏。 10月4日，特斯拉公司首席执行官埃隆·马斯克在社交平台发布了一段该公司开发的人形机器人"擎天柱 （Optimus）"学习功夫的视频。 视频显示，"擎天柱"正与专业人士学习功夫，它展示了用于自卫的武术技巧，且并非单纯模仿专业人士 的动作，而是精准应对每个招式并进行反击。有网友提问"擎天柱"是由远程操控还是人工智能（AI）驱 动的。马斯克回应称，"AI，而非遥控操作"。 据了解，"擎天柱"是特斯拉的人形机器人项目，数年前首次对外公布，而在过去一年中，该项目的重要 性显著提升，也成为特斯拉更核心的关注焦点。 9月7日，特斯拉开通"TeslaAI"的官方微博，认证为"特斯拉官方账号"，并发布首条微博展示其人形机器 人。当天，特斯拉发布了两张人形机器人的照片，并配文："我一直在努力改善自己的身材"。 9月30日，"TeslaAI"再发文，特斯拉正在努力扩大人形机器人的规模，计划2025年底推出第三代，并在 2026年开始量产。马斯克预计2030年前将年产100万台。 9月初，特斯拉发布了"宏图计划第四篇章"（Master Plan Part IV，简称"宏图计划4.0"），预示着特斯拉 未来发展规划的重大 ...

马斯克刷屏。 10月4日，特斯拉公司首席执行官埃隆·马斯克在社交平台发布了一段该公司开发的人形机器人"擎天柱 （Optimus）"学习功夫的视频。 视频显示，"擎天柱"正与专业人士学习功夫，它展示了用于自卫的武术技巧，且并非单纯模仿专业人士 的动作，而是精准应对每个招式并进行反击。有网友提问"擎天柱"是由远程操控还是人工智能（AI）驱 动的。马斯克回应称，"AI，而非遥控操作"。 据了解，"擎天柱"是特斯拉的人形机器人项目，数年前首次对外公布，而在过去一年中，该项目的重要 性显著提升，也成为特斯拉更核心的关注焦点。 9月7日，特斯拉开通"TeslaAI"的官方微博，认证为"特斯拉官方账号"，并发布首条微博展示其人形机器 人。当天，特斯拉发布了两张人形机器人的照片，并配文："我一直在努力改善自己的身材"。 9月30日，"TeslaAI"再发文，特斯拉正在努力扩大人形机器人的规模，计划2025年底推出第三代，并在 2026年开始量产。马斯克预计2030年前将年产100万台。 9月初，特斯拉发布了"宏图计划第四篇章"（Master Plan Part IV，简称"宏图计划4.0"），预示着特斯拉 未来发展规划的重大 ...

Core Insights - The Fourth Global Digital Trade Expo showcased various embodied intelligence products, highlighting the transition of technology from mere exhibits to practical tools for daily life, entertainment, and public services [1][2][3] Group 1: Technological Innovations - Strong Brain Technology presented intelligent bionic hands and legs that utilize sensors to collect weak nerve and muscle signals, enabling users to walk more naturally across different terrains [1][2] - Tai Xi Intelligent introduced a wearable exoskeleton that assists users in climbing and carrying loads by providing real-time support based on gait monitoring [1][2] - Kuangxing Technology showcased four-legged inspection robots and drones that enhance safety by identifying hazards in various environments, achieving full-process intelligence from data collection to problem diagnosis [2][3] Group 2: Applications in Public and Consumer Services - Star Species Robotics unveiled the world's first human-machine collaborative cleaning robot capable of performing 3D space cleaning tasks autonomously, addressing labor shortages in public cleaning [3][4] - Yiqi Future introduced the R1 intelligent therapy robot, which can perform various therapeutic tasks while allowing therapists to manage multiple units simultaneously, enhancing service efficiency [4] - Luka Intelligent Technology's reading robot has sold over 10 million units globally, providing interactive reading experiences for children and supporting educational needs [4][5] Group 3: Future Prospects and Research Directions - Companies are exploring future applications of embodied intelligence, including AI lunar robots designed for tasks on the moon, such as space payment and wireless charging [6][7] - The Hangzhou Human-Robot Innovation Center is focusing on developing a comprehensive robotic brain system to enhance robots' cognitive and operational capabilities in unstructured environments [7]

Core Viewpoint - The robotics industry in China is experiencing significant growth, with an average growth rate of 50% to 100% among related companies in 2023 [1]. Group 1: Company Developments - The founder and CEO of Yushu Technology, Wang Xingxing, announced that the company has iterated its robot algorithms multiple times this year and plans to release a 1.8-meter tall humanoid robot in the second half of the year [1]. - Yushu Technology's robot, the Yushu H1, has shown improvements in performance, winning multiple championships at the 2025 World Humanoid Robot Sports Competition, including 1500m, 400m, 100m obstacle races, and 4x100m relay [1]. - The Yushu H1 can reach speeds of up to 3.3 meters per second, with the potential to exceed 5 meters per second, making it the fastest robot globally [1]. Group 2: Product Launches - The second humanoid robot, Unitree G1, is set to be released in 2024 and was the best-selling robot globally from last year to this year [2]. - The newly launched humanoid robot, Unitree R1, is lighter and has a starting price of 39,900 yuan, with some orders already accepted despite not yet being mass-produced [2].

Core Insights - The CEO of Yushu Technology, Wang Xingxing, announced that the company has iterated its robot algorithms several times this year and plans to release a 1.8-meter tall humanoid robot in the second half of the year [1][2] - The domestic robotics industry has experienced significant growth, with an average growth rate of 50% to 100% among intelligent robot-related companies [1][2] Group 1: Technological Advancements - Yushu Technology's robot algorithms have undergone multiple iterations, improving stability and flexibility, with a new "anti-gravity mode" allowing robots to recover from disturbances [1] - The company has developed new technology enabling robots to perform various dance and martial arts movements, enhancing their performance capabilities [1] Group 2: Product Launches and Achievements - The Yushu H1 robot achieved multiple championships in the 2025 World Humanoid Robot Sports Competition, showcasing its speed and performance [2] - The H1 robot can reach speeds of up to 5 meters per second, making it the fastest robot globally, with expectations to break records again in the near future [2] - Yushu Technology plans to release the Unitree G1 in 2024, which has been the best-selling robot globally from last year to this year, and the new Unitree R1, priced at 39,000 yuan, is currently accepting orders [2]

Group 1 - The fourth Global Digital Trade Expo opened in Hangzhou, highlighting AI and robotics as major attractions [1][4] - The "Smart Future" exhibition area featured six companies from Hangzhou showcasing their technological achievements in a scenario-based manner [3] - Companies like Strong Brain Technology presented intelligent bionic hands and legs, which utilize sensors to interpret users' neural signals for improved mobility [4][5] Group 2 - Yushutech's Unitree G1 demonstrated skills such as dancing and boxing, with the CEO emphasizing the significant opportunities in AI and robotics for various sectors [4] - Tai Xi Intelligent showcased wearable exoskeleton products designed to assist users in carrying loads while climbing or walking, with a market price around 5,000 to 6,000 yuan [5] - Star Species Robotics introduced the world's first "human-machine collaboration" cleaning robot, capable of performing complex cleaning tasks autonomously in 3D spaces [6][7] Group 3 - The VR technology-supported entertainment experiences attracted audience interest, allowing users to interact with various dinosaur species in a simulated environment [7] - The project aims to enhance user engagement through open and exploratory experiences, particularly appealing to children and fostering social sharing [7]

Group 1: Digital Trade Expo Highlights - The Fourth Global Digital Trade Expo opened in Hangzhou, showcasing AI and robotics as major attractions [1] - The "Smart Future" theme area featured six companies from Hangzhou, collectively demonstrating their technological achievements in a scenario-based manner [2] Group 2: Innovations in Robotics - Strong Brain Technology introduced a smart bionic leg that collects weak nerve and muscle electrical signals to assist users in walking naturally across different terrains [2] - Unitree Technology showcased its Unitree G1 robot, capable of dancing and performing martial arts, emphasizing the potential of AI and robotics in various sectors [2] - Tai Xi Intelligent presented a wearable exoskeleton that aids users in carrying heavy loads while climbing or walking, utilizing mechatronics technology for real-time support [2][3] Group 3: Market Applications and Collaborations - Tai Xi Intelligent's consumer-grade exoskeleton is priced between 5,000 to 6,000 yuan and is available for rental in tourist areas at 20 to 30 yuan per hour [3] - The company is collaborating with SF Express to deploy its products in logistics centers to assist with cargo handling [3] Group 4: Cleaning Robotics - Star Species Robotics unveiled the world's first "human-robot collaboration" cleaning robot, capable of performing 3D space cleaning tasks autonomously [3] - This robot is equipped with a water tank and cleaning tools, allowing it to identify stains and utilize various cleaning functions, currently deployed in public areas like airports and metro stations [3] Group 5: VR Technology in Entertainment - The expo featured new entertainment consumption scenarios supported by digital and VR technologies, which were well-received by attendees [4] - Time Space Elf showcased a VR experience allowing users to interact with various dinosaur species, enhancing user engagement through immersive technology [5][7]

Core Insights - The article discusses the rapid advancements in humanoid robot technology, particularly focusing on the Visual-Language-Action (VLA) model systems that can perform various household tasks with high reliability and generalization capabilities. However, a significant bottleneck remains due to the lack of high-quality, comprehensive demonstration data for bipedal robots [1][20]. Group 1: TrajBooster Framework - The TrajBooster framework was proposed by research teams from Zhejiang University and Westlake University to address the challenge of data scarcity by utilizing rich operational data from wheeled robots and trajectory redirection technology to enhance the action learning efficiency of bipedal humanoid robots [1][20]. - The core idea of TrajBooster is to use the 6D end-effector trajectory (3D position + 3D rotation) as a universal interface, allowing for "cross-modal" teaching regardless of robot morphology [2][4]. Group 2: Process Overview - The process involves three main stages: 1. Source data extraction from large datasets of wheeled robots, including language instructions, multi-view visual observations, and corresponding 6D end-effector trajectories [4]. 2. Trajectory redirection in a simulated environment to teach the target bipedal robot how to coordinate its joints to follow these trajectories [4][5]. 3. Model training and fine-tuning using minimal real data from the target robot to deploy the model effectively in real-world scenarios [4][9]. Group 3: Model Architecture - The model architecture consists of a hierarchical control model that breaks down complex problems into manageable sub-problems, with an upper layer for inverse kinematics (IK) to control the arms and a lower layer for a hierarchical reinforcement learning (RL) strategy to manage the legs and balance [5][8]. - The management policy acts as a "decision brain" to determine how the robot should move to reach the target position, while the worker policy translates these commands into specific joint actions [8]. Group 4: Training Phases - The training process includes two phases: Post-Pre-Training (PPT) and Post-Training (PT). PPT combines redirected action data with source data to create a new dataset for further pre-training the VLA model, allowing it to understand the action space of the target robot [9][10]. - The PT phase involves collecting only 10 minutes of real remote operation data to fine-tune the model, bridging the gap between simulation and reality, thus significantly reducing data collection costs [11]. Group 5: Experimental Results - Experiments conducted on the Unitree G1 bipedal robot demonstrated that the model trained with PPT outperformed models trained solely on real data, achieving significant performance improvements in tasks such as "grabbing Mickey Mouse" and "organizing toys" [12][15]. - The model's ability to perform zero-shot skill transfer was highlighted, as it successfully completed tasks not seen during training, indicating effective skill inheritance through trajectory transfer [15][16]. - The model also showed enhanced trajectory generalization capabilities, achieving an 80% success rate in novel object placements compared to 0% for models not using PPT, demonstrating a deeper understanding of the action space [16].

While more affordable than other prominent humanoid robots, the Unitree G1 leaks telemetry data, uses static encryption, and can launch attacks, researchers found. https://t.co/Ncdj5GzaEA ...