全球人形机器人产业正迎来"大脑"技术革命，2025年开年短短三个月内，美国机器人初创公司Figure AI 和谷歌DeepMind都先后公布了各自的通用具身智能大模型，同时，中西部首个人形机器人创新中心—— 成都人形机器人创新中心，也发布了国内首个基于3DSGs的人形机器人规划推理执行系统Raydiculous— 1。 谷歌DeepMind、Figure AI与成都创新中心 正以不同技术路径争夺产业标准话语权，人形机器人 的"脑"力角逐已经拉开帷幕。 ▍谷歌Deep Mind：具身大模型的"通用智能野心" Gemini Robotics 主要有三个方面的提升： 泛化性： Gemini Robotics 是一款 基于视觉-语言-动作（VLA）的端到端模型 ，能够处理全新的、训练 中从未遇到过的任务。例如，向机器人展示一个小型玩具篮球和篮网，并指示"灌篮"，尽管此前从未接 触过这些物体，但仍然理解了指令并完成了动作。Deep Mind称其泛化能力比现有模型提高了一倍。 而 Gemini Robotics-ER 是一款 视觉- 语言模型（VLM） ，专注于增强 空间推理 能力。例如，面对咖 啡杯时，它能识别适合抓取 ...

Core Viewpoint - The world's first humanoid robot half marathon took place in Beijing, showcasing advancements in robotics and generating significant public interest [1][14]. Group 1: Event Overview - The event featured 20 teams of humanoid robots completing a distance of 21.0975 kilometers [1]. - The winning team, Tian Gong, completed the race in a time of 2 hours, 40 minutes, and 42 seconds without changing robots [14]. - Various awards were given, including completion awards and special categories like best endurance and best gait [14]. Group 2: Robot Performance - The Tian Gong Ultra robot, standing 1.8 meters tall and weighing 55 kilograms, improved its speed from 6 km/h to a peak of 12 km/h [23]. - Robots displayed a wide range of performances, with some malfunctioning during the race, such as the only female robot "Huan Huan" which stopped shortly after starting [6][12]. - The event highlighted that larger robots do not necessarily have an advantage over smaller ones, as performance varied significantly among different robots [11]. Group 3: Audience Engagement - The audience showed great enthusiasm, cheering for the robots, with some human participants expressing that they were more interested in watching the robots than achieving personal bests [13][25]. - Despite some robots exhibiting awkward movements, the overall reception was positive, indicating a supportive environment for technological innovation [25]. Group 4: Technical Challenges - Robots faced challenges such as battery depletion and overheating joints, requiring human assistance for cooling [15]. - The Tian Gong team addressed hardware issues like stability and lightweight design, as well as optimizing motion control algorithms for better performance [23]. Group 5: Industry Implications - The event reflects the ongoing innovation in robotics and the importance of public acceptance and encouragement for technological advancements [25]. - The participation of various independent teams using the G1 robot indicates a growing interest and investment in humanoid robotics [21].