英伟达2025年技术图鉴，强的可怕......

Core Viewpoint - NVIDIA has emerged as a leading player in the AI infrastructure space, achieving a market valuation of $5 trillion, which is an 11-fold increase over three years. The company has transitioned from a graphics chip manufacturer to a key player in AI, particularly in autonomous driving and embodied intelligence [2]. Group 1: NVIDIA's Technological Developments - The Cosmos series, initiated in January, focuses on world foundation models, leading to the development of Cosmos-Transfer1, Cosmos-Reason1, and Cosmos-Predict2.5, which lay the groundwork for autonomous driving and embodied intelligence [5]. - The Nemotron series aims to create a "digital brain" for the agent-based AI era, providing open, efficient, and precise models and tools for enterprises to build specialized AI systems [5]. - The embodied intelligence initiatives include GR00T N1 and Isaac Lab, which focus on simulation platforms and embodied VLA (Vision-Language-Action) models [5]. Group 2: Key Papers and Contributions - The paper "Isaac Lab" presents a GPU-accelerated simulation framework for multi-modal robot learning, addressing challenges in data scarcity and the simulation-to-reality gap [6]. - "Nemotron Nano V2 VL" introduces a 12 billion parameter visual language model that achieves state-of-the-art performance in document understanding and long video reasoning tasks [12]. - "Alpamayo-R1" proposes a visual-language-action model that integrates causal reasoning and trajectory planning to enhance safety and decision-making in autonomous driving [13]. Group 3: Innovations in AI Models - "Cosmos-Predict2.5" introduces a next-generation physical AI video world foundation model that integrates text, image, and video generation capabilities, significantly improving video quality and consistency [17]. - "Cosmos-Reason1" aims to endow multi-modal language models with physical common sense and embodied reasoning capabilities, enhancing their interaction with the physical world [32]. - "GR00T N1" is an open foundation model for generalist humanoid robots, utilizing a dual-system architecture for efficient visual language understanding and real-time action generation [35].