元戎启行成立Superfluid Lab，押注物理AI基础能力
元戎启行组建Superfluid Lab，从智驾转向物理AI基础研究，核心是用闭环系统打通数据、模型与真实世界。
物理AI需要让机器理解环境、预测变化并行动，但数据、认知、行动与真实世界之间存在多重断裂。元戎启行成立Superfluid Lab，由前DeepSeek成员阮翀带队，以基座模型为核心，探索VLA与仿真闭环，试图建立一套能持续学习的物理世界智能底座。
正文摘录
- title: 模型路线趋同之后，Physical AI 的胜负手变了 - sourcecompany: 量子位 - bodymarkdown: Over the past year or so, hot terms like VLA, world models, foundation models, and data flywheels have been frequently mentioned by leading players in autonomous driving and embodied AI. More and more AI companies are starting to push models into the real world, aiming to give machines the ability to understand their environment, predict changes, and take actions. The core question is how AI can obtain data from the physical world, form an understanding of the environment, translate that understanding into action, and then continue learning from the outcomes of those actions. Early driving systems were highly modular software stacks. The perception module handled vehicle, road, and pedestrian detection; the predictio…