Meiyang builds a closed-loop physical AI path: turning the human body into a computable real world

Wanmeijing AIMIRA builds a closed product loop of "Perception-Understanding-Decision-Making-Execution-Feedback" to answer the core proposition of Physical AI

When AI steps out of the screen, the rules of the game are completely changed: the real human body has spatial structure, tissue state and constant temporal changes; software can regenerate if it outputs a wrong sentence, but once a machine makes a wrong movement, the result may be irreversible. The difficulty of Physical AI never lies in building a larger model, but in the long chain of "Perception-Modeling-Decision-Making-Execution-Verification-Relearning". Guangzhou Meiyang Innovation Technology (Calilia International Group / Wanmeijing AIMIRA) has turned this chain into a product system: FAITH acts as the perception entry, the "Nuwa" real human body world model is responsible for modeling and understanding, medical-grade intelligent hardware and embodied devices perform execution, and effect and long-term changes flow back to the system.

Founder of Meiyang, Dr. Jiang Yuzhen, said: "Nuwa is not a 'human foundation model' that can answer everything, but a set of multimodal real-world models centered on the real human body, with both spatial structure and temporal dimension. Digital humans solve the problem of 'how to present a person', while we solve the problem of 'how to enable machines to continuously understand what changes are happening to this real person'."

Figure 1

Starting from 3DGS: Not only realistic rendering, but also accurate measurement

Nuwa is not a digital human: starting from 3D Gaussian Splatting, it builds a human world model that can "measure accurately". 3D Gaussian Splatting (3DGS) takes into account both appearance and rendering efficiency, but "realistic rendering" does not mean "accurate measurement". Meiyang adopts a hybrid route that combines Gaussian representation, point cloud and 3D mesh: the Gaussian layer retains a realistic appearance, calibrated geometry provides a measurement interface, and time series tracks changes of the same person. The team's 3D data base has accumulated about 26.59 million multi-view high-definition human body images, and some modeling processes only take 5 to 20 seconds, which demonstrates the sustainably deliverable 3D asset production capacity.

A fine 3D surface alone is still not equal to understanding the human body: a depression may correspond to soft tissue volume, and the risk of an injection path depends on subcutaneous blood vessels. In addition to 2D and 3D vision and multispectral skin information, Meiyang further targets non-invasive perception such as ultrasound soft tissue and blood vessels, photothermal and diffusion waves, and establishes a "personal long-term dynamic digital file" — images, status and treatment information at different time points are jointly aligned and accumulate value over time. The boundary of medical scenarios is clear: new modalities must be verified, clinical capabilities are expressed in accordance with the registration scope, and will not replace existing standard diagnostic methods.

Perception - Decision - Execution: FAITH and Medical-grade Hardware: Connect Nuwa to real services

The commercial problem of Physical AI is: models require real data, data comes from deployment, and deployment requires paying users. Meiyang's solution is to promote delivery and model construction in the same scenario: centering on head, face and body rejuvenation services, the closed loop of "Assessment - Plan - Execution - Result - Reassessment" supports both commercial delivery and technology accumulation. On the execution side, the ASP-1M hardware body obtained the Class III medical device registration certificate from the National Medical Products Administration in July 2026, providing a hardware foundation for subsequent AI assistance and automated execution.

Figure 2

The revenue path starts from hardware, consumables and services, extends to model and agent subscriptions, and expands reuse through third-party device access and technology licensing. FAITH has completed productization, and the intelligent upgrade of existing devices has obtained partial paid verification. The focus of the next stage is to form recurring revenue.

From Understanding the Human Body to Understanding Devices and the World

Guangzhou Meiyang Innovation Technology Co., Ltd. is promoting the construction of Physical AI infrastructure with the human body as the entry point. The company has established a joint Physical AI laboratory with South China University of Technology, carrying out research on machine forgetting, edge intelligence and privacy protection, gathering interdisciplinary scientific research forces of academicians including Chen Junlong, Wang Yingjun and Andreas Mandelis. Meiyang uses FAITH multimodal perception and the "Nuwa" real human body world model to connect intelligent decision-making, embodied execution and result feedback.

Figure 3

This also points to Meiyang's long-term ambition: to build a real-world model base as a core national technological asset. From understanding a real person to understanding a continuously operating device, Meiyang has always accumulated the same capability: to make machine judgments based on real-world evidence, and make physical actions verifiable for results.