Chinese Scientists Achieve Breakthrough in Satellite-Driven AI Control for Humanoid Robotics

A collaborative effort between GuoXing Aerospace Technology and Shanghai Jiao Tong University has marked a significant milestone in the convergence of space-based computing, artificial intelligence, and robotics. This pioneering experiment successfully demonstrated the remote operation of humanoid robots on Earth via a sophisticated satellite network. By deploying Large Language Models (LLMs) directly into orbit, the team enabled logical reasoning to occur on the satellites themselves, effectively bypassing traditional ground-based servers and significantly reducing operational latency.

This technological leap is strategically vital as it establishes a new paradigm for accessing AI token services from space, potentially offering a workaround for current international restrictions on high-end AI chip exports. Wang Yabo, the Executive Vice President of GuoXing Aerospace Technology, confirmed that this event represents the world's first deployment of a universal, large-scale AI model controlled from a ground station to an active satellite constellation. During the tests, an operator on the ground issued verbal commands that were transmitted to the orbital fleet, where the Alibaba Qwen3 model—housed within radiation-resistant hardware—processed the requests and generated precise digital instructions.

Once these instructions were calculated in orbit, they were beamed back to Earth, where an open-source software agent known as OpenClaw translated the data into physical movements for the robot. This closed-loop system proves that artificial intelligence can serve as a functional, active computing core for orbital control systems. Such a capability is essential for maintaining autonomous operations in environments where ground infrastructure is compromised or non-existent, such as major disaster zones or isolated geographical regions.

The feasibility of this "orbital brain" concept was validated by the Qwen3 model’s performance, which processed ground queries and returned actionable results in less than two minutes. This success paves the way for managing a diverse array of autonomous systems, including unmanned aerial vehicles and self-driving cars, from space. Chengdu-based GuoXing Aerospace Technology has already begun the aggressive rollout of its infrastructure to support this vision, starting with the launch of the first "Xingshidai" cluster of twelve specialized computing satellites in May 2025.

Launched from the Jiuquan Satellite Launch Center via a Long March-2D rocket, this initial cluster is just the beginning of a massive orbital network. The company’s roadmap includes expanding the fleet to 1,000 active units by 2030, with a final goal of 2,800 specialized satellites by 2035. This architecture will utilize high-speed laser inter-satellite links to provide a staggering 100,000 petaflops of inference capacity and 1 million petaflops for AI training on a global scale.

With the second and third groups of satellites scheduled for deployment in 2026, China is positioning itself at the forefront of the space computing revolution. This move mirrors a broader international trend toward orbital processing, highlighted by the November launch of SpaceX’s Starcloud-1 satellite, which features Nvidia graphics processing units. As space becomes the new frontier for high-performance computing, the integration of orbital AI and terrestrial robotics promises to redefine global communication and autonomous technology.