Sou Hu Cai Jing·2025-09-19 14:14Core Insights - Both Nvidia and ByteDance are focusing on the development of advanced robotics, despite their different core competencies in chip technology and algorithm models respectively [1][9]. Group 1: Nvidia's Developments - At the GTC conference in March, Nvidia showcased the NEO Gamma humanoid robot, which utilizes the latest GR00T N1 model for training and can autonomously perform tasks like room organization [3]. - Nvidia launched the Jetson AGX Thor chip, boasting a computing power of 2070 FP4 TFLOPS, which is 7.5 times more powerful than its predecessor, Orin, marking it as a significant advancement in robotics [3][5]. - The Jetson platform provides a comprehensive ecosystem for developers, allowing them to leverage existing models like GR00T N1 for easier training [5]. Group 2: ByteDance's Innovations - ByteDance's Seed team released the self-developed VLA model GR-3, which can perform complex tasks such as hanging shirts and understanding abstract language commands in a household environment [5][7]. - The GR-3 model employs a mixed training strategy, achieving a parameter scale of 4 billion and outperforming Nvidia's GR00T model in various task tests [7]. - ByteDance has produced over 1,000 logistics robots for e-commerce warehousing, showcasing its commitment to robotics beyond content creation [11]. Group 3: Industry Trends - The focus of artificial intelligence has shifted from language and generative models to enabling robots with cognitive abilities, marking a crucial step for technology to integrate into real-world applications [11]. - Both leaders, Zhang Yiming and Jensen Huang, are driving the development of robotic intelligence, indicating a significant evolution in the tech landscape [12].