Roboxing
All coverage
Analysis

How EngineAI says the T800's fight moves are made: martial artists' motion data, remapped to the robot, per one Chinese report

A Global Times report from the World Robot Conference quotes EngineAI's CEO on recording martial artists to train the T800 for URKL. One source; every claim is attributed.

EngineAI, the company behind the URKL humanoid fighting league, says the fight moves of its T800 robot start with human martial artists. The account comes from a Chinese report on EngineAI's stand at the World Robot Conference (WRC) 2026, published by Global Times Tech on 21 August 2026 and carried by Sina Finance. It is the only source we have read for these details, so everything below is attributed to it.

WHAT WAS ON THE STAND

The report says two T800 teams, named "Doujiang" and "Luzhu" in the report (roughly "soy milk" and "braised offal"), re-staged URKL's opening-night bout in an octagonal cage at the EngineAI stand, drawing crowds. It says this was EngineAI's third consecutive year at WRC. It does not say whether the robots in the cage were piloted or acting autonomously, so we do not state either.

HOW THE MOVES ARE MADE, PER ENGINEAI'S CEO

The report quotes EngineAI founder and CEO Zhao Tongyang saying that, to collect motion data that is both watchable and effective in a fight, the company invited many martial artists from China and also brought in overseas resources, and recorded their body movements to train large models.

Zhao is quoted saying that collecting the movement is only the start. A robot's skeleton differs fundamentally from a human one and its joints sit in different places, so human motion data has to be remapped. Engineers then transfer it to the robot body and tune it repeatedly, checking that each action is stable, expressive and has both defensive and offensive capability. The report adds that the company rents a large amount of cloud computing for motion-control algorithms and large-model training.

THE CONTROL STACK AS DESCRIBED

The report describes the T800's fighting ability as built on a loop of real-time visual feedback, decision and execution, and says the robot is moving towards a tactically capable autonomous agent. It also reports that EngineAI released "EngineAI Awaken" at the conference, described as a five-layer architecture modelled on the human brain: a reflex layer, an action-planning layer, a cognitive-reasoning layer and a self-growth layer. According to the report, heavy semantic reasoning runs at about 2 Hz and high-frequency motion control at about 100 Hz, decoupled so that slow model computation cannot cause loss of control of the body. These are EngineAI's own claims as reported; we have not seen technical documentation.

URKL AS A BUSINESS

The report quotes EngineAI co-founder and marketing head Yao Qiyuan describing URKL as a commercial fighting event IP, a stage for industry showcase and technology iteration. It also says a T800 began working at a Luxshare Precision factory in Suzhou in August 2026, loading and moving materials, which is outside fighting and which we have not verified.

WHY IT MATTERS

Fight footage tends to be read as either remote control or full autonomy. This report describes a third layer that applies to either: the repertoire of moves a robot can perform comes from human motion capture, retargeted to the machine, before any pilot or algorithm chooses when to use it. For the league-by-league picture of who controls what, see our explainer on remote-controlled and autonomous fighting. For the machine itself, see our T800 profile.