Robots undergoing training in diverse range of fields

At the embodied intelligence training center of China Xiongan Group Digital City Technology Co Ltd in Xiong'an New Area, Hebei province, robots are being trained to perform tasks in a wide range of scenarios.

Using handheld controllers, data collection technicians remotely operate the robots to perform actions.

The data collected helps improve large AI models, which are then deployed to the robots, making them smarter.

Li Guoliang, assistant general manager of the company, said embodied intelligence is designed for a vast range of industry scenarios and applications across different sectors.

However, the challenge lies in the fact that the number of possible scenarios is enormous, he said.

"For example, if a robot is asked to move a stool from here, and the stool is replaced with a different one, our robots, powered by our models, can still recognize it as a stool and move it accordingly," he said, adding this demonstrates their generalization capability.

For robots, data serves as the foundation for understanding and perceiving the world. Robots need continuous exposure to data collected from diverse scenarios, along with ongoing data refinement, to train the models before they can perform effectively, according to Li.

The training center adopts a 1:1 real-scene approach and builds five practical application scenarios, namely guest reception and guidance, industrial cargo handling, household living, commercial retail, and logistics assembly.

It takes the lead in realizing a full set of capabilities covering unified equipment management, unified data collection, unified data labeling and model training for heterogeneous robots from multiple manufacturers.