Unitree Robots Perform Kung Fu in Beijing

The Unitree H1 robot, featured in the demonstration, stands 180cm tall and weighs 47kg. It set a world record for bipedal running speed at 3.3 m/s and features high joint torque (360 N·m at the knee) powered by a replaceable 864 Wh battery. The robot perceives its environment using a combination of 3D LiDAR and an Intel RealSense depth camera. Hangzhou-based Unitree Robotics recently achieved a valuation of $1.7 billion after a Series C funding round that attracted tech giants like Tencent, Alibaba, Ant Group, and Meituan. Founded in 2016 by Wang Xingxing, the company now employs over 1,000 people and has reported annual revenues exceeding 1 billion yuan. Unitree is reportedly preparing for an IPO on Shanghai's STAR Market, targeting a valuation of up to $7 billion. The demonstration is part of a broader push for "embodied intelligence" across China, with cities like Shanghai and Beijing investing heavily in the sector. China's Ministry of Industry and Information Technology reported that the country produced nearly 100 embodied AI robotic products since 2024, accounting for 70% of the global market. This growth is supported by a national strategy to become a world leader in AI by 2030, with a goal of monetizing AI into a trillion-yuan industry. For orchestrating such multi-robot systems, open-source frameworks like Microsoft's AutoGen and CrewAI are gaining traction. AutoGen utilizes a chat-centric model for asynchronous agent communication, while CrewAI focuses on a higher-level abstraction of role-playing agents to simplify collaboration. These frameworks are critical for solving handoff challenges, as research shows multi-step workflows with 95% reliability per step can still result in only a 36% overall success rate over 20 steps. Reliability in these multi-agent systems remains a significant hurdle; evaluations show top AI agents achieve less than a 55% goal completion rate in complex enterprise environments. Addressing this involves establishing clear automation boundaries and continuous learning loops from production data. Recent research papers highlight the importance of agent architectures that unify long-term and short-term memory management and enable self-evolution based on experience. From a user experience perspective, designing for autonomous agents requires new principles. Key challenges include visualizing the agent's reasoning process, providing users with clear controls to override actions, and managing what the agent remembers. Microsoft's "Agent UX Design Principles" emphasize making the AI's status and capabilities transparent to build user trust and ensure human control. The regulatory environment for AI in China is rapidly evolving. While a comprehensive national AI law is still in development, Beijing has implemented targeted regulations for algorithms, deep synthesis, and generative AI services. In September 2025, the government announced it had issued 30 national standards for AI, with 84 more in development, signaling a move towards a more structured governance framework.