Unitree Robotics (宇树科技) founder says company leads global humanoid shipments, teases speed and mass data push

Key claims and numbers

At the Yabuli Forum, Unitree Robotics (宇树科技) founder Wang Xingxing (王兴兴) said his firm’s second‑generation humanoid, released in 2024, has become the world’s most‑shipped humanoid over 2024–25 and that the company shipped roughly 5,000 units last year. The model is a compact 1.3‑metre platform aimed at lightness and agility; Wang noted the cheapest version retails for about ¥30,000. He contrasted that with the larger H2 humanoid — about 1.8 metres and 70 kg — which he warned can be “a bit scary” and advised keeping a 2–3 metre safety distance.

Performance, road map and cultural reach

Wang highlighted recent athletic benchmarks, saying Unitree swept several events at a Beijing humanoid competition — including 1,500m, 400m and a 4x100m obstacle relay — and that the 1,500m time was “about six minutes,” faster than most of his staff. He predicted that within months, humanoid robots, especially in China, could run 100 metres “under 10 seconds,” a claim he presented as an expected near‑term improvement. He also described a forthcoming 2025 AI upgrade dubbed “Kung Fu mode,” and pointed to the company’s Spring Festival Gala appearance, which blended classic Chinese martial arts and robot choreography and drew substantial overseas attention.

Data, generalization and the path to embodied AI

To address data scarcity, Wang outlined plans to deploy a full‑body teleoperation system and scale to “thousands or even 10,000” humanoids by year‑end to capture roughly 10 hours of data per robot per day, which he said would solve the data bottleneck within a couple of years. He warned that current systems still struggle to generalize: high success in trained scenes falls off sharply in new ones. Wang defined an “embodied‑ChatGPT moment” as a robot completing about 80% of tasks in 80% of unfamiliar scenes via language alone, and estimated that will take two to three years, rather than the more optimistic 18 months some foresee.

Tech partners and geopolitical note

Wang praised video‑generation advances — notably ByteDance (字节跳动)’s Seedance 2.0 — as key to aligning generated motion with robot control, which he argued could unlock robust world models for embodied AI. He also said many top labs and companies, including Nvidia (英伟达), use Unitree platforms. It has been reported that U.S. export controls on advanced AI chips and other geopolitical frictions could complicate hardware supply chains for Chinese robotics firms, a relevant risk as companies scale up production and compute needs. Wang closed by saying China currently enjoys a valuable first‑mover window that may not last if timing were different.