Physical AI Startup 厘清智能 Raises Multi-Million Seed Round

A new player in the Physical AI arena, 厘清智能 (Liqing Intelligence), has successfully closed a substantial seed funding round, amassing several hundred million yuan (approximately $30 million USD). This significant capital injection underscores investor confidence in the company's unique approach to developing AI systems that interact with the physical world. The funding was led by prominent venture capital firms including Sequoia Capital China, Hillhouse Ventures, Matrix Partners China, and IDG Capital, alongside strategic investments from industry players like 智元机器人 (Zhiyuan Robotics), 灵心巧手 (Lingxin Qiaoshou), and 世纪金源 (Centennial Group).

Founded by a team with deep ties to Tsinghua University and led by former NVIDIA researcher Li Yiming, 厘清智能 is deliberately distancing itself from the prevailing hype surrounding "world models." Instead, the company emphasizes the development of a comprehensive Physical AI Infrastructure. This infrastructure is designed to bridge the gap between data acquisition, model training, and real-world deployment, aiming to solve practical problems rather than chasing abstract theoretical constructs. Li Yiming articulated that a world model is merely a component, akin to a horse carrying a vital shipment, but valueless without the entire logistical system.

The company's proprietary infrastructure centers on two core self-developed components: a data pipeline engineered to scale data collection from hundreds of thousands to tens of millions of hours, and a physics engine that facilitates a Real-to-Sim-to-Real learning loop. This closed-loop system leverages real-world data to build simulations for robotic reinforcement learning, which are then applied back to physical robots. This approach enables fine-grained manipulation skills such as cutting, screwing, and grasping, with the capability to deploy across diverse robotic embodiments and various operational environments, including manufacturing, retail, hospitality, and healthcare.

The broader AI sector has seen a surge of companies labeling themselves as "world model" providers, often encompassing video models, 3D models, and vision-language-action (VLA) systems. However, 厘清智能's strategy focuses on building a robust, end-to-end system. Li Yiming highlighted the scarcity of talent capable of integrating both software and hardware expertise, a critical factor for success in Physical AI. The Tsinghua University affiliation has provided a strong talent pool, with the team's average age around 23, fostering rapid development and innovation.

This focus on a full-stack, integrated approach is seen as a key differentiator. By controlling every stage from data collection to model training and hardware integration, 厘清智能 aims to ensure seamless information flow and optimized performance across its systems. This "heavy" approach, involving significant upfront investment and cross-disciplinary technical challenges, has deterred many competitors but is viewed by 厘清智能 as essential for achieving true generalization and problem-solving capabilities in complex physical environments.

Looking ahead, 厘清智能 plans to release cross-scenario world models by the end of the year and achieve scaled solutions by 2028. The company's ultimate vision is to deliver integrated hardware and software solutions that are adaptable to various robotic platforms and operational contexts, effectively offering "World Model as a Service." This strategic positioning aims to address the growing demand for intelligent automation across a multitude of industries, moving beyond theoretical advancements to practical, deployable AI.