LatentVerse takes a third path beyond VLA and world models for embodied intelligence
LatentVerse, a new startup co-founded by Tsinghua University doctoral student Hu Yucheng, has secured a nine-figure RMB seed funding round from investors including GL Ventures and Crystal Stream. The company is developing an embodied-native foundation model, UTAM (unified tactile action model), which aims to overcome limitations of existing vision-language-action (VLA) and world model approaches in embodied intelligence. LatentVerse seeks to unify understanding, prediction, and action within a single architecture, addressing the data bottleneck and the need for greater generalization and dexterity in robotics. This approach targets complex real-world scenarios beyond controlled industrial settings, such as home environments and hotel room cleaning.
LatentVerse's emergence highlights a critical inflection point in Asia's embodied intelligence sector, moving beyond the established VLA and pure world model paradigms. The company's focus on unifying understanding, prediction, and action within a single architecture, exemplified by its UTAM model, addresses the persistent data bottleneck and the need for robots to exhibit stronger generalization, long-horizon execution, and dexterity for broader commercialization. This strategic pivot is particularly relevant for Asian markets, where demand for automation in manufacturing, logistics, and increasingly, consumer services and smart homes, is rapidly expanding. The ability to handle unstructured environments and complex tasks, such as grasping soft objects or navigating varied home settings, will unlock significant market opportunities.
The investment from prominent Chinese VCs like GL Ventures and Crystal Stream, alongside strategic players like Agibot and Robot Era, underscores the growing confidence in deep tech and AI startups in China. LatentVerse's ambition to reduce data collection costs significantly through innovative methods, including wearable devices for nonprofessionals, could democratize access to training data and accelerate development across the region. This approach not only positions LatentVerse as a potential leader in embodied AI but also signals a broader trend of Asian startups driving fundamental research and product innovation to solve real-world challenges in robotics, rather than merely adapting Western technologies.
分享这篇文章
相关文章
6 则
DeepSeek leads surge in low-cost Chinese open-weight models on US platform
LatentVerse's emergence reflects the broader surge in AI model development across Asia, as seen with DeepSeek.
美国还是中国？哈萨克斯坦在人工智能革命中拒绝选边站队
LUXEED 与 Aimoga 机器人：通过互联的人车家生态系统共同构建未来出行
“四方安全对话”外长在亚特兰大会晤，同意扩大联合演习及海上监测
稀有、宽敞又昂贵？深入了解新加坡小众“巨型组屋”市场
