KITECH builds a 45 km/h treadmill dome to capture whole-body human motion for humanoid training data

The Korea Institute of Industrial Technology (KITECH) said in a press release dated Oct 14 that a team led by principal researcher Ahn Bum-mo in its AI and robotics division has built HHIP (Human-to-Humanoid Intelligence Platform), a facility that records a person's movement from head to toe in one synchronised data stream.

The set-up places a high-speed treadmill that reaches 45 km/h inside a dome ringed by optical motion-capture cameras, with floor force plates measuring how hard the feet strike, glove sensors tracking finger motion and EMG sensors on the skin logging muscle effort; the institute said its own software aligns all of these signals on a single time axis, which it described as the platform's central problem.

KITECH said the platform was first used to validate AirKnee, a knee-worn wearable robot it calls the lightest of its kind, by comparing walking and lifting with and without the device, and that it is now collecting joint-motion data from people doing work tasks to train humanoid control models.

Ahn said the team plans to accumulate a wider range of work motions for humanoids, wearable robots and rehabilitation. Most humanoid training data in MEKAWIRE's recent coverage, such as the HiPHI human-object interaction benchmark published on Oct 7, comes from vision-only capture; KITECH's set-up adds ground-reaction force and muscle signals, the variables a legged robot's controller has to reproduce.