Westlake Robotics Raises $100M for Embodied AI

Westlake Robotics, a Chinese startup focused on embodied AI and robotics, has successfully closed a significant Series A funding round, raising approximately $100 million USD (500 million RMB). This latest infusion of capital marks the company's fourth fundraising effort in just six months, underscoring strong investor confidence in its advanced robotics control technology. The round saw participation from prominent investors including SAIF Partners, Xiaomiao Langcheng, Henan Investment Group Huirong Fund, and Haiyuan Fund. Funds will be strategically allocated to further develop the company's full-body control AI models for humanoid robots and expand its engineering team.

Founded in January 2024 by esteemed Westlake University professors Wang Donglin and Zhang Yue, Westlake Robotics aims to commercialize cutting-edge research from the university's AI and robotics labs. The company differentiates itself by focusing on the integrated development of foundational models for embodied AI and the physical robot hardware. Its core team comprises seasoned professionals from tech giants like Alibaba and Tencent, as well as researchers from prestigious institutions such as Oxford and Cambridge Universities, bringing a wealth of expertise to the venture.

While the broader robotics industry is largely concentrating on enhancing visual recognition, language comprehension, and task planning, Westlake Robotics is pioneering advancements in motion control for full-body coordination in humanoid robots. The company identifies a critical gap: even as AI models improve robots' understanding of their environment, translating that understanding into stable, real-world physical action remains a significant hurdle. Tasks like opening doors or tidying spaces require intricate coordination of dozens of joints, dynamic balance adjustments, and precise manipulation, far exceeding the complexity of stationary robotic arms.

Addressing this challenge, Westlake Robotics has developed a proprietary general-purpose motion model called GAE (General Action Expert). Unlike traditional Vision-Language-Action (VLA) models, GAE, built on Transformer architecture, directly translates user instructions into comprehensive physical movements. It can also replicate human motion in real-time using motion capture technology, enabling dynamic motion generation without pre-programming. This adaptability allows GAE to be deployed across different robot platforms and supports remote operation capabilities, as demonstrated by successful control over 1300km distances.

The company's innovative approach is further embodied in its "dual pre-training architecture" of 'Cerebrum' and 'Cerebellum'. The 'Cerebrum' component, integrating VLA and world models, handles environmental perception, task comprehension, and decision-making. The 'Cerebellum' (GAE) translates these decisions into precise, millisecond-level full-body movements and balance corrections. This modular design allows for independent training and refinement of each component, enhancing adaptability to various robot hardware through standardized interfaces.

Westlake Robotics is actively addressing the industry-wide shortage of high-quality, real-world robot motion data by emphasizing "field training." They deploy robots in practical scenarios such as emergency response, power grid inspections, and domestic services to collect rich, real-time kinematic data. Although they unveiled their first humanoid robot, Westlake O1, in 2026, the company views its hardware development primarily as a means to bridge the simulation-to-reality gap and refine their AI models, rather than a core business focus. This iterative cycle of real-world testing, data collection, and model improvement is crucial for overcoming the limitations of simulation environments.

With a robust technological framework spanning algorithms, foundational models, and robot hardware, Westlake Robotics is well-positioned to pursue its long-term vision of creating versatile models applicable to a wide array of robotic tasks. The company has already secured significant commercial traction, with current orders nearing $24 million USD (100 million RMB) from partnerships with major players like Alibaba Group, JD.com, and Intel. Their applications span education, data collection, and infrastructure inspection, with a significant collaboration underway with the Zhejiang Provincial Longyou County government to establish a large-scale humanoid robot training base, accelerating commercial deployment.