2025.08. 25 本文字数：966，阅读时长大约1.5分钟 今天， 英伟达将公布最新的人形机器人"黑科技" 。 据上海证券报，8月24日，英伟达机器人官方账号在社交平台上发布了一张黑色礼盒的照片，附着一 张带着创始人黄仁勋签名、配文为"好好享受"的贺卡。 据英伟达的解释，物理AI依托于神经图形、合成数据生成、物理模型、强化学习和AI推理技术，是现 代机器人技术、自动驾驶汽车和空间智能技术背后的引擎。在生成式AI、代理AI浪潮之后，就是物理 AI浪潮。 英伟达今年已在寻找更多GPU的落地应用场景，在工业领域和机器人领域的动作越来越多，这两个 领域都与物理AI相关。在机器人领域，英伟达有Jetson Thor GPU机载计算机。有市场研究机构的 研究负责人告诉记者，即便很多芯片的算力都达到要求，但目前主流人形机器人的算力解决方案并不 多，其中不少机器人采用了英伟达的方案，因为其生态较完善，可以节省厂商开发时间。 英伟达与物理AI相关的产品也在加快面世。今年1月，英伟达在全球消费电子展CES上推出了基础世 界模型Cosmos，用于生成合成自动驾驶等场景，辅助物理AI系统开发。此次SIGGRAPH，英伟达 则推 ...

Core Viewpoint - The next wave of technology is Physical AI, which is being rapidly deployed by NVIDIA, leveraging neural graphics, synthetic data generation, physical models, reinforcement learning, and AI inference technologies [2] Group 1: Physical AI Development - NVIDIA's CEO Jensen Huang has emphasized the importance of Physical AI multiple times this year, indicating it as a significant future trend [2] - At the 2025 World Robot Conference, NVIDIA's Omniverse and simulation technology VP highlighted that Physical AI could unlock a trillion-dollar market [2] - NVIDIA presented 16 papers on Physical AI at the SIGGRAPH 2025 conference, marking it as the central theme of their participation [2] Group 2: Relationship Between Graphics Research and Physical AI - There is a close relationship between graphics research and Physical AI, with advancements in rendering leading to the creation of realistic virtual worlds for training advanced Physical AI systems [3] - NVIDIA has significantly improved real-time path tracing performance by 10,000 times over the past nine years, enhancing the training of AI systems [3] - Simulations play a crucial role in robot research, allowing for safe and efficient training in virtual environments, which can be several orders of magnitude more efficient than real-world training [3] Group 3: Expectations and Applications of Robotics - Different types of robots have varying expectations, from simple tasks for industrial arms to complex navigation for autonomous vehicles and versatile skills for humanoid robots [4] - The key to transforming generative AI into Physical AI lies in training data, with synthetic data being crucial to overcoming data challenges [4] - NVIDIA is actively seeking more applications for GPUs in industrial and robotics sectors, which are closely related to Physical AI [4] Group 4: Product Launches and Innovations - NVIDIA has accelerated the release of products related to Physical AI, including the Cosmos world model for generating synthetic driving scenarios and the open-source 70 billion parameter visual language model, Cosmos Reason [6] - New hardware solutions are being introduced, such as servers equipped with RTX PRO 6000 Blackwell GPUs for enterprise workloads related to AI and Physical AI [6] - Upcoming products include the RTX PRO 4000 Blackwell SFF and RTX PRO 2000 Blackwell GPUs, aimed at engineering, AI, and 3D visualization applications [6]