Unitree Unveils ‘First’ Fully Autonomous Humanoid Boxing Robot

HANGZHOU, China: Unitree Robotics has said one of its humanoid robots had fought fully autonomously, with no scripted movements and no human at a remote control, hailing it a world first. Video footage shared on Unitree's social pages shows two full-sized humanoids throwing punches and predicting each other's movements footage. 

Unitree adds that the video was recorded in real time and not sped up, and it ends with a six-panel display that the Japanese robotics publication Robostar read as the model's internal predictions of several possible next moves.  

Robostar, which reviewed the video, described the bout as a robot against a human. Unitree has published no benchmark results, no inference to latency figures, and no code or model weights. 

Unitree's earlier robot fights were run by people. G1 humanoids that boxed in a China Media Group tournament broadcast in May 2025 were handled by human controllers at ringside, using remote controls and voice commands. 

UnifoLM-X2-1.0 follows UnifoLM-WMA-0, which Unitree open-sourced in September 2025 with training code, inference code and weights. That system fine-tuned a video generation model on the Open-X dataset and then on five Unitree datasets, using its predictions of an interaction to feed an action head.  

The real-robot demonstrations included desktop scenarios; stacking boxes, cleaning pens, and loading cameras. Unitree has not said whether X2-1.0 will be released the same way. 

Wang Xingxing, Unitree's founder and chairman, described the wider constraint at the World Robot Conference in Beijing on Aug. 20.  

Task success rates can approach 100% in a fixed scenario given enough training data, he said, but drop sharply once the object or the environment changes.  

He also said the company had run robots on automakers' production lines and on simple tasks in its own plant without deploying them widely. 

Unitree listed on Shanghai's STAR Market on Aug. 19, raising 6.1 billion yuan ($905 million).