Humanoid robots are set to enter a phase of accelerated development over the next two to three years, according to Xie Yipeng, head of industrial scenario solutions at Unitree Robotics. Speaking at the Gasgoo Symposium on Embodied Perception Fusion and Multimodal Large Models on September 15, 2026, Xie said that physical AI is most likely to emerge in factories, but noted that historical data formats and correlation methods often lag behind the needs of AI training, with older and more advanced industries carrying heavier burdens.
From first all-electric humanoid to commercial shows
Unitree launched the world's first fully electric humanoid robot in 2023 and pioneered this technical route. The iteration pace has since become clearer: in 2024, the company solved walking stability, enabling the robot to run, jump, and resist being knocked over, overcoming the once-mysterious three-axis inverted pendulum challenge. In 2025, Unitree began seeking commercial applications, and in the fourth quarter of that year, commercial dance and martial arts performances became a breakthrough, showcased during Spring Festival galas. The related technologies have subsequently been deployed in mass-produced products.
Xie emphasized that only when robots can learn dance and stage movements at a time and cost comparable to human dancers can the application be considered truly viable. The principle that follows is: if a robot can perform as well as a human at a similar cost, it is worth considering; robots are better suited for tasks that people are unwilling to do, or that are dangerous or toxic.
Open collaboration, full-stack self-development
Unitree maintains an open cooperation model, allowing tools, electric drives, perception sensor suites, radar, binocular vision, industrial wireless communication, and additional power systems to be integrated with its robots. However, Xie stressed that Unitree will not become an engineering company or handle project delivery. Instead, it will remain a cutting-edge humanoid and quadruped robot company, focusing on full-stack in-house development of core components such as reducers, gears, and motor stators and rotors, to ensure the fastest iteration, lowest cost, and most stable quality. The company also selects semiconductor chips tested to be most suitable for AI training, ensuring overall machine leadership and precise positioning.
According to Gasgoo, the remarks were made during a session focused on embodied perception and multimodal AI, reflecting Unitree's strategy to combine hardware innovation with software capabilities.
