LatentVerse develops embodied foundation models that combine vision, language, action and touch for robot learning. Founded by a Tsinghua University team, it is researching pretraining, world models, dexterous manipulation, data infrastructure and edge deployment.
About
LatentVerse was founded in May 2026 by researchers from Tsinghua University’s Institute for Interdisciplinary Information Sciences. Founder and chief executive Hu Yucheng previously worked on embodied foundation models at ByteDance Seed. The company’s research programme spans large-scale pretraining, world models, dexterous manipulation, data infrastructure and deployment on robot hardware.
Its UTAM architecture uses touch as a native model input throughout the architecture, supporting closed-loop perception, decision, action and correction. The company is at an early research and product-development stage; no commercial deployment scale was disclosed with the seed financing.