Unitree's Self-Evolving Robots: China-US Humanoid Race Enters New Phase 2026

Unitree Robotics is pushing the humanoid robot race into a new phase by developing self-evolving robots that can learn and improve themselves with minimal human intervention. The company's founder, Wang Xingxing, unveiled the concept at its physical AI conference in Hangzhou, positioning it as a top priority for the firm.

The move signals a shift from manual robot training to AI-driven automation, potentially accelerating how quickly humanoid robots acquire new skills. As the US and China compete for dominance in robotics, Unitree's approach could reshape the industry's development cycle.

Why Self-Evolving Robots Matter

Teaching a humanoid robot a new skill is notoriously slow. Engineers must collect data, train models, and fine-tune behavior—often finding that a skill perfected in the lab fails in a new environment and needs reworking.

Unitree aims to automate this loop so robots can gain new abilities faster and at greater scale. Wang told the South China Morning Post that as AI models improve, he expects them to handle more of the planning and iteration, reducing the need for human engineers.

How Unitree's Self-Evolving System Works

The system operates as a closed loop: AI agents search for research papers, write control code for the robots, test that code in simulation, and learn from the results. The best outcomes are then verified and presumably deployed to physical robots.

This approach mirrors how large language models iterate on tasks, but applies it to physical machines. By automating research and coding, Unitree hopes to compress months of engineering into days or hours.

• Automated research: AI agents scour academic papers for relevant robotics techniques.
• Code generation: Agents write control code tailored to specific robot hardware.
• Simulation testing: Code is tested in virtual environments to catch errors early.
• Iterative learning: Results feed back into the system to improve future iterations.
• Scalable skill acquisition: The goal is to let robots learn new tasks without human intervention.

The China-US Humanoid Robot Race

Unitree's announcement comes amid intensifying competition between China and the US in humanoid robotics. Both countries are investing heavily in AI and robotics, viewing them as strategic technologies.

China has produced several high-profile humanoid robot companies, while US firms like Tesla and Boston Dynamics have also made strides. Unitree's focus on self-evolution could give China an edge in development speed.

What This Means for the Robotics Industry

If successful, Unitree's self-evolving system could dramatically lower the cost and time required to deploy humanoid robots. That could accelerate adoption in manufacturing, logistics, and service industries.

However, the concept also raises questions about reliability and safety. Allowing AI to write and test control code without human oversight could introduce risks, especially in physical systems where errors have real-world consequences.

Unitree has not announced a timeline for deploying this system commercially, but Wang's comments suggest it is central to the company's long-term strategy. The background to this story is set out in China's Humanoid Robot Race: Progress, Pitfalls, and Promise. A related thread runs through China AI and Humanoid Robots: Beijing Warns Against Blind Investment. Earlier coverage in FCC Bans Humanoid Robots from China Over National Security Risks shows how the argument developed.

What to Watch Next

Observers should watch for demonstrations of robots learning new skills without human coding. Unitree's progress will also indicate how quickly China's humanoid robot industry is advancing relative to US competitors. A closer look at the same subject appears in Unitree Humanoid Robot Shares Surge 600% on Market Debut. Anyone following the wider picture can start with US Bans Chinese Humanoid Robots Citing National Security Risks. The same theme surfaces in FCC Bans Chinese Humanoid Robots Over National Security Risks, which takes a different angle.

As AI models grow more capable, the line between software and hardware development may blur, with robots themselves becoming active participants in their own evolution. Context on how this played out before sits in US Bans Humanoid Robots from China: National Security Risks. A companion piece, Humanoid Robots Compete in Chinese Sports, approaches the same subject from another angle.