Unitree Technology Holds 2026 World Physical AI Conference, Wang Xingxing Outlines Three Major Strategic Directions
Unitree Technology held the 2026 World Physical AI Conference in Hangzhou, where Wang Xingxing outlined three major future strategic directions: enhancing the company's capacity for large-scale robot mass production, advancing small physical AI models through in-house development and partnerships, and building out ultra-large-scale computing clusters. Since its launch in May 2024, the Unitree G1 has cumulatively rolled off the production line at roughly 10,000 to 20,000 units, making it the world's best-selling humanoid robot. Unitree is currently conducting preliminary research on a V1.0 self-evolving large model for physical AI robots, and has announced a World Physical AI Competition that will award tens of millions of yuan in prize money each year.
Body · AI translation
On October 22, the 2026 World Physical AI Conference and Unitree Partner Ecosystem Conference was held in Hangzhou. Under the theme "Advancing Together with Yu, Growing Together in Symbiosis," the conference brought together more than 3,000 guests and featured 1 main forum, 7 parallel sub-forums, more than 20 open speeches, and more than 40 exhibition zones, with the main venue covering an area of more than 4,000 square meters. Wang Xingxing, Chairman, General Manager, and Chief Technology Officer of Unitree Robotics; Wang Jian, founder of Alibaba Cloud; Yao Ming, a leading figure in Chinese basketball; and other prominent guests appeared at the conference to jointly explore the paths for technological breakthroughs and ecosystem co-building in the physical AI industry.
The conference showcased on-site integrated deployment solutions for Unitree Robotics' robots across diverse scenarios such as urban governance, emergency firefighting, education and research, culture and entertainment, commercial services, and industrial applications. At the same time, Unitree's quadruped robots, humanoid robots, and the latest five-fingered dexterous hand, the Dex5-S, and other products were all on display.
In his keynote speech at the conference, Wang Xingxing introduced that Unitree Robotics' classic robot, the Unitree G1, has become the world's best-selling humanoid robot since its release in May 2024, with cumulative production of approximately 10,000 to 20,000 units. Previously, Unitree Robotics' fully open-source general-purpose humanoid robot foundation model, UnifoLM-WLA-1.0, set new records on multiple global technology leaderboards. Its open-source ecosystem has attracted universities and enterprises worldwide to jointly develop algorithms and contribute them back as open source, greatly accelerating the global diffusion and iteration of robotics technology.
Regarding industry technology trends, Wang Xingxing judged that physical AI has now shown signs of being broken through, very likely within the next 1–2 years or 2–3 years. Although a 5–10 year cycle is still needed for AI technology to fully penetrate from cutting-edge fields into the daily lives of the general public, breakthroughs in the physical AI field are accelerating: large models have already solved the core problems of robot planning and communication, leaving only the bottleneck of physical interaction success rates and contact stability. It is expected that the full breakthrough of capability boundaries can be achieved in as fast as 2 years or as long as 10 years. The core challenge for embodied intelligence to pass the ChatGPT moment is the alignment between the input and output of AI models and real robots.
"For top large-model companies, in a sense, they have basically already achieved self-evolution of models internally. The models write programs and collect data themselves, and humans only need to appropriately guide and participate," Wang Xingxing said frankly.
Based on this judgment, Wang Xingxing clarified three major strategic directions for Unitree Robotics' future layout: first, enhance the company's large-scale mass production capacity for robots to meet future market demand at the tens of millions or even hundreds of millions level; second, pursue both in-house development and partnerships in parallel to advance the deployment of small physical AI models adapted to its own hardware; third, build ultra-large-scale computing clusters to solidify the foundation for technological iteration. At the same time, the company is building a robot self-evolution system that replaces traditional manual debugging paths through a closed-loop model of "AI autonomous R&D—simulation training—real-world testing—feedback iteration," improving the efficiency of algorithm iteration.
At the conference, Wang Xingxing also officially launched the World Physical AI Competition. The competition invests tens of millions of RMB each year as prize money and is open to teams and individuals worldwide. It features eight tracks: arbitrary locomotion, track and field, combat, dexterous manipulation, production, sports events, multimodal interaction, and health and elderly care, aiming to unite global innovative forces and drive continuous breakthroughs in the boundaries of physical AI technology.
Wang Xingxing revealed at the conference that Unitree Robotics is currently conducting preliminary research on the Physical AI Robot Self-Evolution V1.0 large model. He expressed the expectation that in 2027–2028, Unitree Robotics' robots equipped with the physical AI large model will achieve significant technological breakthroughs, robots will be able to enter households, and in the future all of human civilization will be able to move to the next level.