Core Viewpoint - The article discusses the launch of the VLA model G0 by the domestic robotics startup Xinghaitu, which has raised over 1.5 billion yuan and achieved a valuation of 1 billion USD, highlighting its innovative dual-system architecture for intelligent robotics [2][8]. Financing and Valuation - Xinghaitu has completed five rounds of Series A financing in 2025, raising over 1.5 billion yuan, with notable investments from Meituan and Ant Group [8][9]. - The company is reportedly seeking a new round of financing at a valuation of 1 billion USD [9]. Product and Technology - The G0 model features a dual-system architecture that separates high-level thinking and low-level actions, enhancing the robot's ability to execute complex tasks [4][6]. - The G0 model utilizes 500 hours of high-consistency full-body motion data and a three-stage learning strategy to achieve long-range task execution from visual and language commands [2][4]. - The G0-VLM (visual language model) operates at a frequency of 2 Hz for task instruction output, while the G0-VLA (visual-language-action model) executes actions at 15 Hz and motion at 200 Hz [4]. Competitive Landscape - The dual-system architecture is becoming mainstream in the field of embodied intelligence, with other companies like Zhiyuan and Figure AI also developing similar models [7][8]. - Xinghaitu's G0 model has reportedly outperformed the π0 model from Physical Intelligence in various performance metrics [7]. Founding Team and Company Background - Xinghaitu was founded by a team with strong backgrounds in autonomous driving and academia, including professors from Tsinghua University [10][11]. - The company focuses on developing a comprehensive ecosystem for embodied intelligence, including both hardware and software solutions [10][22]. Market Position and Applications - The company's products are primarily targeted at research scenarios, serving clients such as laboratories and universities for technology research and data collection [20]. - Xinghaitu's robots have been delivered to over a hundred top developers globally, establishing it as a leading platform for embodied intelligence model training [19].

从制造到创造，从不可能到可能。《晚点LatePost》旗下汽车品牌。 具身智能没有现成答案，只能去找最有希望找到答案的人。 文丨张家豪 以下文章来源于晚点Auto ，作者晚点团队 晚点Auto . 赵行是 MIT 计算机视觉博士，后在 Waymo 担任研究员。2020 年回国加入清华交叉信息学院。许华哲在伯 克利获得博士学位，后在斯坦福博士后，是清华 "Embodied AI（具身智能）" 实验室负责人。 许华哲长于具身智能的操作，赵行则研究视觉感知和导航技术，二人的专长分别对应目前具身智能的两个 核心模块——机器人的智能操作能力，和机器人完成任务所需的环境感知能力乃至物理规律理解能力。 理想汽车目前的自动驾驶方案是端到端 +VLM（视觉语言大模型），其中 VLM 正是理想和赵行实验室的 合作成果。 编辑丨程曼祺 "太难选了，实在挑不出。" 一位密切关注具身智能的早期投资人近期告诉我们，他们在 5-6 个整机公司里 犯难，于是转而投资电机等机器人零部件企业。 难以决断，是前沿科技投资和创新的常态。越新的领域，往往越没有明确、清晰的技术与商业路线。 自 2023 年火热至今的通用具身智能和人形机器人正处于这个 ...