top of page
Robots 91

Newsletter N°91 - Summer Edition 2026 

🤖 Robots/AI: Beijing clears dual AI models for humanoid robots

Beijing Humanoid Robot Innovation Center has secured regulatory approval for two foundational models developed for its Huisikaiwu embodied intelligence platform, marking three national firsts for China’s humanoid robotics industry. The Pelican-VL and WoW models have simultaneously completed Beijing’s latest filing process for generative AI services, establishing the company as the first in the sector to obtain approval for both a general-purpose “brain” foundation model and an embodied world model.

ROBOTS 91.jpg

Beijing Humanoid Robot Innovation Center secures regulatory approval for its two AI foundation models. (Photo: Beijing Humanoid Robot Innovation Center)

The achievement is significant because Chinese regulations require domestically developed generative AI services to undergo security assessment and regulatory filing before being offered externally. To meet these requirements, Beijing Humanoid Robot Innovation Center developed its own testing framework covering data compliance, model capabilities, output safety, scenario adaptation and risk management. More than 100,000 test cases were reportedly conducted across refusal behaviour, non-refusal scenarios and keyword detection, including adversarial testing

ROBOTS 91 2.jpg

Xu Zhiyuan, algorithm expert at Beijing Humanoid Robot Innovation Center, training humanoid robots to acquire new skills. (Photo: Shanghai Securities News)

The two models address complementary layers of robotic intelligence. Pelican-VL is positioned as the robot's cognitive foundation, providing general-purpose visual-language understanding to support perception, reasoning, task planning and interaction. Rather than focusing solely on conversational capabilities, the model is designed to serve as the underlying intelligence layer for embodied AI applications.

​

WoW, meanwhile, is described as China's first registered embodied world model. It connects digital environments with physical robots, enabling machines to model their surroundings, anticipate potential outcomes, simulate movements and adjust actions dynamically. This capability is particularly important for humanoid robots operating in unstructured real-world environments, where fixed rules and pre-programmed movements are insufficient.

​

Together, the two models create a complementary “brain + world” architecture. Pelican-VL handles global cognition, reasoning, task planning and multi-robot coordination, while WoW focuses on physical environment modelling, action simulation, risk prediction and motion rehearsal. The combination is intended to create a complete loop from understanding the environment and planning a response to simulating and executing the action on a physical robot.

​

The regulatory milestone also opens the door to commercialisation. Beijing Humanoid Robot Innovation Center plans to launch Token-based access to the Huisikaiwu model family, progressively making API access available to industrial customers, research organisations and developers. The models will also be integrated into the company's Tiangong 3.0 humanoid robot, enabling the company to offer combined software-and-hardware solutions for industrial manufacturing, commercial services and specialised operations.

​

The development illustrates an important evolution in China's physical AI ecosystem. While many previously registered robotics-related AI models have focused primarily on human-robot interaction, the latest filing concerns the deeper intelligence layers required for autonomous physical operations. Regulatory approval therefore becomes not only a compliance milestone but also an enabler for deploying foundation models across real-world robotic applications.

​

More broadly, Beijing's move highlights how AI model regulation and humanoid robotics commercialisation are increasingly converging in China. By establishing a regulatory pathway for both cognitive foundation models and embodied world models, the initiative could help accelerate the transition from experimental humanoid robots to scalable, API-enabled robotic intelligence services.

bottom of page