Crowdworks said on Wednesday it has built an in-house "Physical AI Data Lab" to strengthen its physical artificial intelligence (AI) data business.

The Physical AI Data Lab is an organisation that researches and validates multimodal data for physical AI that operates in physical environments such as robots, autonomous driving and manufacturing.

Crowdworks has introduced an actual humanoid robot and launched a system that collects, processes and verifies human behaviour and robot sensor data, including joint trajectories, force and visual information, using a teleoperation method in which the robot is remotely controlled. It will build data for training vision-language-action (VLA) models and domain-specific episode datasets.

It will supply the collected data to global robotics companies through its in-house platform, the "A1 Data Marketplace". It is currently distributing more than 100 terabytes of data, including datasets for gripper-based dual-arm robots, sensor-glove-based datasets and 10-finger robot datasets.

Lee Jun-ho (이준호), Crowdworks' chief operating officer, said, "The success or failure of physical AI depends on how accurately and precisely you can turn the countless variables of the real world into data." He added, "We will lead the physical AI market by supplying data solutions optimised for advanced industrial sites such as humanoid robots."