LatentVerse, an embodied intelligence company founded by 2001-born Tsinghua University PhD candidate Hu Yucheng, has completed a seed round financing of hundreds of millions of RMB, with investors including GL Ventures, Qingliu Capital, Zhiyuan, Star Era, InnoAngel and other funds. Hu, who resigned from ByteDance Seed in April this year, co-founded LatentVerse with team members from Tsinghua University, Nanyang Technological University, Peking University, Qwen, ByteDance, Xiaomi and other institutions. The company is positioned to develop native embodied foundational models, and Hu explicitly refuses the world model label, stating that many world model companies lack unique features. LatentVerse chooses a third technical route, neither building VLA nor the popular WAM, but unifying VLM and the world model in the same architecture. This contrasts with the mainstream VLA route, which can only utilize data with action annotations collected by teleoperating robotic arms at an average cost of more than 1,000 RMB per hour. Hu was one of the first researchers to prove that "world models can directly guide robot actions", proposing the multimodal diffusion VLA (PAD) in early 2024 and VPP, the first video-action model to realize closed-loop reasoning, in November the same year. VPP is known as the pioneering work of WAM, and later NVIDIA developed DreamZero based on it.