en.Wedoany.com Reported - Archon Robotics, a Chinese general-purpose full-body embodied intelligence company, recently completed a seed funding round of hundreds of millions of yuan. Investors include ZhenFund, Gaorong Capital, IDG Capital, 5Y Capital, as well as Gobi Partners' joint fund with the University of Hong Kong, MiraclePlus, and Shanghai Innovation Institute. LightSource Capital served as the exclusive financial advisor for this round. The funds will be primarily used for the development of full-body humanoid foundation models, multimodal full-body motion data collection, team expansion, and the establishment of R&D centers and industrial collaboration ecosystems in multiple locations, accelerating the launch of an open-source humanoid foundation model within this year.
Founded in April 2026, Archon Robotics has its R&D headquarters in the Caohejing Hi-Tech Park, Xuhui District, Shanghai, China. The company focuses on general-purpose full-body humanoid foundation models, aiming to build full-body intelligence capabilities for humanoid robots. Unlike embodied intelligence approaches that only focus on task learning for robotic arms, grippers, or wheeled chassis, Archon Robotics emphasizes the movement, balance, torso leverage, bimanual coordination, and multi-joint linkage of the complete humanoid body. The goal is to equip humanoid robots with full-body mobile manipulation abilities closer to those of humans.
The key highlights of this funding round are not only the amount reaching hundreds of millions of yuan but also the strong composition of investors. Institutions like ZhenFund, Gaorong Capital, IDG Capital, and 5Y Capital have long focused on artificial intelligence, robotics, and foundation model entrepreneurship. The participation of Gobi Partners' joint fund with the University of Hong Kong and the Shanghai Innovation Institute also provides Archon Robotics with a stronger background in university research commercialization and regional innovation platforms. For a newly established embodied intelligence company, receiving support from multiple leading institutions in its seed round indicates that capital is shifting the competitive focus of embodied intelligence from robot hardware to "robot foundation models" and "embodied brains."
Archon Robotics' technical roadmap emphasizes "full-body intelligence." Currently, most embodied intelligence training data is concentrated on scenarios like tabletop operations, first-person perspective videos, single-arm grasping, and gripper actions. While this data helps robots perform fixed-point grasping, tabletop organization, and simple operations, it struggles to cover the more complex body coordination required in real homes or open environments. When humans perform actions like pulling a door, lifting heavy objects, squatting to pick something up, or sidestepping through a narrow space, the truly critical factors are not just hand trajectories but also center-of-mass transfer, torso angle, lower limb support, and whole-body force transmission.
This is why Archon Robotics chose the full-body humanoid foundation model approach. If a humanoid robot merely mimics end-effector trajectories, it may perform well on simple tasks but will struggle with variations in object weight, spatial layout changes, and continuous task sequences. A full-body humanoid foundation model needs to learn the logic of interaction between the entire human body posture and the environment, integrating limbs, torso, center of mass, and contact forces into the model training. Only then can the robot understand "how to stand, how to leverage force, how to balance, and how to operate" in real-world environments.
Multimodal full-body motion data collection will be a key area of investment for this round of funding. Full-body data differs from ordinary video data; it requires recording multidimensional signals such as human posture, hand movements, foot motion, center-of-mass changes, tactile feedback, environmental information, and task objectives. The collection cost is higher, and annotation is more difficult, but the data value is also greater. A single full-body motion data point containing center-of-mass movement and torso angle changes may be more valuable for helping a robot understand the physical world than a large amount of tabletop data that only records hand trajectories.
Archon Robotics also plans to build a "collection-training-feedback" closed loop. Embodied model training is not a one-time process. As model capabilities improve, they will in turn help the team determine which data is more valuable, which tasks still fail, and which motion patterns require additional collection. As the data covers more scenarios, the model's understanding of full-body motion and physical interaction will gradually strengthen. This type of data flywheel is key to building a long-term competitive moat for embodied intelligence companies.
The team background is also a significant reason this funding round attracted attention. Dr. Li Hongyang, founder of Archon Robotics, is currently an Assistant Professor at the University of Hong Kong, Assistant Dean of the Faculty of Computing and Data Science, and a mentor at the Shanghai Innovation Institute. He previously led the end-to-end autonomous driving project UniAD. Co-founder and CEO Dr. Li Tianyu is among the first graduates of the Shanghai Innovation Institute and holds a PhD from Fudan University. He participated in the development of the world engine solution for Huawei's mass-produced autonomous driving system ADS 4.0. Co-founder and AI Lead Dr. Chen Li is the first author of the UniAD paper. Team members come from universities and research teams including the University of Hong Kong, Tsinghua University, Shanghai Jiao Tong University, Fudan University, and Zhejiang University, possessing cross-disciplinary experience in autonomous driving, robotics, and large models.
Autonomous driving and embodied intelligence share commonalities in underlying methodologies. Autonomous driving must solve the closed loop of perception, prediction, planning, and control. Similarly, embodied robots need to understand the environment, predict action consequences, plan task paths, and control body execution. By leveraging experience from end-to-end autonomous driving models and world models, the Archon Robotics team can transfer complex systems engineering expertise to embodied intelligence. The difference lies in the fact that humanoid robots must handle more complex body movements and real-world contact, requiring the model to understand continuous interactions between hands, feet, torso, and external objects.
Following this funding round, Archon Robotics will accelerate the development of its full-body humanoid foundation model and plans to launch an open-source humanoid foundation model within this year. If successfully released, the open-source humanoid foundation model could provide a more general full-body intelligence base for robotics companies, research institutions, and developers, lowering the barrier to entry for humanoid robot algorithm development. The current embodied intelligence industry still lacks a unified technical paradigm, and an open-source foundation model could encourage more teams to develop applications and adapt hardware based on the same core capabilities.
Embodied intelligence is transitioning from a competition in hardware to a competition in brains. In the past, the industry focused more on whether a robot could stand steadily, walk, grasp objects, and complete demonstration tasks. In the next phase, the true determinant of deployment potential will be whether a robot possesses generalization capabilities—the ability to autonomously adapt across different hardware, different homes, and different tasks. Archon Robotics securing hundreds of millions in seed funding indicates that capital is betting on underlying model companies in embodied intelligence. Key points to watch going forward will include the release timeline of the first native humanoid foundation model, the scope of open-sourcing, the construction of the data collection system, cross-hardware transfer capabilities, and whether its model can demonstrate verifiable performance in real home and open-environment tasks.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
