2025.08. 26 Deepu Talla表示，现在有很多人形机器人客户在一台机器上同时使用2个或2个以上的Orin，很多汽 车客户也在同时使用2个甚至4个Orin。算力提升后，一些工作负载可以在同一个Jetson Thor上运 行。至于机器人为何需要边缘端运算、不能只依靠云端计算，Deepu Talla表示，云端计算有延迟情 况，机器人有时也可能连接不到云、无法做出实时决策，这些都是机器人不可接受的。 据介绍，Jetson平台已有220万名开发者，有7000多家公司在使用Orin。一些中国公司已在使用 Orin，并准备使用Thor，这些公司包括智元机器人、众擎机器人、银河通用、优必选、宇树科技 等 。 从 价 格 上 看 ， Jetson Thor 开 发 套 件 售 价 3499 美 元 ， 生 产 模 块 Jetson T5000 （ 2070 TFLOPS）售价2999美元。 Deepu Talla表示，在物理AI和机器人领域，当前有巨大的机遇，"我们正处在一个激动人心的时 刻，通用机器人时代正在到来。有两类技术对于通用机器人十分重要，一是生成式AI和推理式AI， 二是模拟技术。" 本文字数：15 ...

英伟达(NVDA.US)周一正式推出其专为机器人应用设计的计算平台 Jetson Thor，同步推出开发者套件 Jetson AGX Thor和量产模组Jetson T5000。新的Jetson AGX Thor开发者套件起售价为3499美元。 Jetson Thor基于英伟达Blackwell架构打造，相比前代Jetson Orin产品实现了显著性能提升。该平台AI计 算性能提高至前代的7.5倍、即较前代提高了6.5倍，CPU性能提升至前代的3.1倍，内存容量增加1倍， 达到128GB。 据悉，这一性能飞跃使机器人开发者能够在边缘端处理高速传感器数据并执行视觉推理，而此前这类工 作流因速度限制无法在动态真实环境中运行。该平台专为生成式推理模型打造，可支持由大型 transformer模型、视觉语言模型驱动的下一代物理AI智能体在边缘端实时运行，最大限度减少对云端依 赖。 英伟达首席执行官黄仁勋表示："Jetson Thor 专为全球数百万开发者打造，助力他们构建可与物理世界 交互、甚至改变物理世界的机器人系统。它具备无与伦比的性能与能效，还能够在边缘设备上同时运行 多个生成式 AI 模型。作为一款卓越的超级 ...

梦晨 发自 凹非寺 量子位 | 公众号 QbitAI 英伟达直接把服务器级别的算力塞进了机器人体内。 全新的机器人计算平台 Jetson Thor 正式发售，基于最新的Blackwell GPU架构，AI算力直接飙升到2070 TFLOPS，比上一代Jetson Orin提高至整整 7.5倍 ，同时能效提高至 3.5倍 。 128GB的超大内存 配置，在边缘计算设备里是前所未有。 在宣传片中，黄仁勋把他当成送给所有机器人的礼物： 黄仁勋表示： Jetson Thor助力全球百万开发者构建可与物理世界交互、甚至改变物理世界的机器人系统。具备无与伦比的性能与能效，还能够在边 缘设备上同时运行多个AI模型。 作为一款卓越的超级计算机，Jetson Thor正在推动物理AI与通用机器人时代真正到来。 在软件优化上，英伟达也下了功夫。 Jetson Thor基于Jetson软件栈优化，支持所有流行的生成式AI框架和AI推理模型，包括主流语言模型Qwen、DeepSeek以及视觉语言动作 模型到和视觉语言模型等。 Jetson AGX THOR开发者套件美国售价3499美元 （约25000人民币，但中国售价尚未公布） 。 ...

北京时间8月25日晚间，英伟达宣布新一代机器人计算平台Jetson Thor开发套件（developer kit）和生产模块（production modules）上市。该计算平台于去年3 月发布，此次上市意味着市面上可用的英伟达机器人计算平台算力再次提升。 Deepu Talla表示，现在有很多人形机器人客户在一台机器上同时使用2个或2个以上的Orin，很多汽车客户也在同时使用2个甚至4个Orin。算力提升后，一些 工作负载可以在同一个Jetson Thor上运行。至于机器人为何需要边缘端运算、不能只依靠云端计算，Deepu Talla表示，云端计算有延迟情况，机器人有时也 可能连接不到云、无法做出实时决策，这些都是机器人不可接受的。 据介绍，Jetson平台已有220万名开发者，有7000多家公司在使用Orin。一些中国公司已在使用Orin，并准备使用Thor，这些公司包括智元机器人、众擎机器 人、银河通用、优必选、宇树科技等。从价格上看，Jetson Thor开发套件售价3499美元，生产模块Jetson T5000（2070 TFLOPS）售价2999美元。 据介绍，作为机器人"大脑"，Jetson ...

Core Viewpoint - The article discusses the evaluation of the PI0-FAST-DROID model in real-world scenarios, highlighting its potential and limitations in robotic operations, particularly in handling new objects and tasks without extensive prior training [4][10][77]. Evaluation Method - The evaluation utilized the π₀-FAST-DROID model, specifically fine-tuned for the DROID robot platform, which includes a Franka Panda robot equipped with cameras [5][10]. - The assessment involved over 300 trials across various tasks, focusing on the model's ability to perform in diverse environments, particularly in a kitchen setting [10][11]. Findings - The model demonstrated a strong prior assumption of reasonable behavior, often producing intelligent actions, but these were not always sufficient to complete tasks [11]. - Prompt engineering was crucial, as variations in task descriptions significantly affected success rates, indicating the need for clear and structured prompts [12][59]. - The model exhibited impressive visual-language understanding and could mimic continuous actions across different scenarios [13][28]. Performance in Complex Scenarios - The model showed robust performance in recognizing and manipulating transparent objects, which is a significant challenge for traditional methods [20][27]. - It maintained focus on tasks despite human movement in the background, suggesting effective prioritization of relevant visual inputs [25]. Limitations - The model faced challenges with semantic ambiguity and often froze during tasks, particularly when it encountered unfamiliar commands or objects [39][42]. - It lacked memory, which hindered its ability to perform multi-step tasks effectively, leading to premature task completion or freezing [43][32]. - The model struggled with precise spatial reasoning, particularly in estimating distances and heights, which resulted in failures during object manipulation tasks [48][50]. Task-Specific Performance - The model's performance varied across different task categories, with notable success in simple tasks but significant challenges in complex operations like pouring liquids and interacting with household appliances [89][91][100]. - For instance, it achieved a 73.3% progress rate in pouring toy items but only 20% when dealing with real liquids, indicating limitations in physical capabilities [90]. Conclusion - The evaluation indicates that while the PI0 model shows promise as a generalist policy in robotic applications, it still requires significant improvements in instruction adherence, fine manipulation, and handling partial observability [77][88].

点击下方 卡片 ，关注" 具身智能 之心 "公众号 作者丨 Jie Wang等 编辑丨具身智能之心 本文只做学术分享，如有侵权，联系删文 >> 点击进入→ 具身智能之心 技术交流群 更多干货，欢迎加入国内首个具身智能全栈学习社区 ： 具身智能之心知识星球 (戳我) ， 这里包含所有你想要的。 blog：https://penn-pal-lab.github.io/Pi0-Experiment-in-the-Wild/ 这是 GRASP Lab 的一篇在复杂真实场景中（in the wild）评估 PI0-FAST-DROID 的工作，这样可以更直观的帮助理解 PI0 这类通用 policy 的目前性能和边界，以 及探索未来可以解决的方向。 当然现在还有更新一代的 PI0.5 方案（但是还没有开源）。 相关资料 ： Droid 数据集 ：https://droid-dataset.github.io/ 引言： 机器人操作领域，一直以来都缺少能够"开箱即用"地处理新物体、新位置和新任务的预训练模型 。机器人专家们往往曾经历过令人沮丧的过程：为了获取一个 机器人 policy，不得不进行繁琐的工程设计和数据收集，结 ...

Core Insights - Magic Atom showcased its entire family of robots and various embodied intelligence solutions at the 2025 World Robot Conference (WRC 2025) [1] - The company has formed strategic partnerships with KuPass and Youka Technology to build a specialized data collection training ground for the automotive sector, aiming to create a large-scale training platform for embodied robots [6][8] Group 1: Product Showcase - The lineup at WRC 2025 included various robots such as the full-size humanoid robot "Xiao Mai," high-dynamic humanoid robot MagicBot Z1, consumer-grade quadruped robot MagicDog, wheeled quadruped robot MagicDog-W, and the newly launched industrial quadruped robot MagicDog Y1 [2] - The performance of the robots, particularly the "Silicon-based Dance Troupe," highlighted their advanced motion control capabilities, with demonstrations of high-dynamic actions like getting up from a fall and navigating obstacles [2] - The industrial quadruped MagicDog Y1 features an IP67-rated sealed body, allowing stable operation in extreme temperatures and complex environments, enhancing its application in industrial inspections and emergency rescues [3] Group 2: Application Scenarios - Magic Atom demonstrated the dual-hand operation of the humanoid robot Gen-1 in an industrial glue dispensing scenario, achieving high efficiency and precision in continuous operation [4] - The humanoid robot also showcased a smart tour guide solution, capable of providing professional explanations and real-time updates through a dedicated app, enhancing user experience [4] - The consumer-grade quadruped MagicDog serves as a companion robot, supporting voice interaction and mobile app control for family entertainment [5] Group 3: Strategic Initiatives - Magic Atom, established in January 2024, focuses on the research and application of general-purpose robots and embodied intelligence technology, with a diverse product range across industrial, commercial, and household scenarios [6] - The company initiated the "Thousand Scenes Co-Creation" plan, collaborating with over 1,000 partners to develop 1,000 application scenarios, forming a robust ecological alliance [7] - The strategic partnership with KuPass and Youka Technology aims to create a leading training ground for embodied intelligence in the automotive sector, addressing challenges in data costs and resource availability [8]

Group 1 - The company, QiaoJieShuWu, a developer of motion control systems for general robots, has completed angel+ and Pre-A financing rounds, raising nearly 100 million yuan in total over three rounds within six months [2][3]. - The new funding will primarily be used for the technical development of a general robot action development platform, product upgrades, and talent recruitment [3][5]. - The core technology team of QiaoJieShuWu comes from prestigious universities, with an average age of 26, indicating a youthful and innovative workforce [4][5]. Group 2 - The company is transitioning its business model from project-based delivery to a combination of licensing, service fees, and subscription models to better meet customer demands in the booming field of embodied intelligence and humanoid robots [5][6]. - QiaoJieShuWu aims to simplify the process of enabling robots to have motion capabilities, making it as easy as assembling a computer, with plans to launch a general robot action development platform that is user-friendly and adaptable [6][7]. - The market for embodied intelligence has seen increased investment activity, with several companies announcing new funding rounds, indicating a growing interest and potential in the robotics sector [6][7].

Core Viewpoint - The article highlights the advancements in general-purpose robotics showcased at the World Robot Conference (WRC), particularly focusing on the capabilities of the humanoid robot "Aibao" developed by Zhifang. The robot demonstrates versatility in performing various tasks, indicating a significant leap towards achieving general intelligence in robotics [4][12][26]. Group 1: General Robotics - Aibao is characterized as a general-purpose robot capable of performing multiple tasks and adapting to different environments, moving beyond the limitations of specialized robots [12][13]. - The robot can autonomously switch between tasks, such as drumming and making ice cream, showcasing its flexibility and adaptability [14][15]. - Aibao's design includes over 34 degrees of freedom, allowing it to perform complex movements and tasks efficiently [21]. Group 2: Technological Innovations - Aibao is powered by the GOVLA model, which integrates global perception, omni-body coordination, long-horizon flexible tasks, and rapid learning capabilities [27][28]. - The GOVLA model enables Aibao to perceive its environment in 360 degrees, allowing it to execute tasks that require understanding of spatial relationships [28]. - The robot's ability to coordinate its entire body for complex tasks marks a significant advancement in robotic capabilities [30][32]. Group 3: Commercial Applications - Aibao has begun commercial deployment in various industries, including automotive manufacturing, semiconductor production, biotechnology, and public services [43][45]. - In automotive manufacturing, Aibao performs tasks such as material handling and quality inspection, enhancing efficiency in production lines [45]. - The robot is set to be deployed in airports for providing information and assistance to travelers, demonstrating its utility in public service environments [45].

瑞财经 刘治颖 近日，桥介数物（深圳）科技有限公司（以下简称"桥介数物"）宣布完成天使+轮及PreA轮融资，天使+轮投资方包括复星创富、潜能集团 等，PreA轮由隐山资本领投，明荟致远、沂景资本等机构跟投。 此次融资募集资金将主要用于公司通用机器人感控一体动作开发平台的研发升级，扩充并加速更加前沿技术的开发进程，提升产品化能力，进一步推动这一 全球领先的的通用机器人动作开发平台实现商业化落地。至此，桥介数物在半年时间内连续完成3轮融资，累计金额近亿元。易凯资本在该系列交易中担任 了桥介数物的独家财务顾问。 桥介数物成立于2023年5月，作为国内规模最大的通用机器人运动控制系统研发商，公司致力于通用机器人"小脑"（运动控制系统）的研发，在强化学习与 模仿学习、仿真运动控制、运控数据分析以及sim2real等领域建立了领先的核心技术体系，这些技术共同构成桥介数物完整的技术图谱。 桥介数物正在开发全球第一款通用机器人动作开发平台，该平台将针对人形、四足以及轮足等各类机器人构型的运动控制需求，为全球通用机器人产业提供 世界一流的运动控制数字基座能力，有望成为未来机器人运动控制领域的核心驱动力。 桥介数物创始人尚阳星， ...