'Robot school' in Hefei, E China's Anhui starts to train first batch of 'students'

A trainer teaches a robot to fold clothes at an embodied intelligent robot data acquisition pre-training field, dubbed "robot school", in Hefei, capital of east China's Anhui Province. (People's Daily Online/Ding Ji)

A robotic arm slowly reached for a wrench, carefully adjusting its grip and angle before tightening and turning a screw at a workstation.

Standing beside it, trainer Wu Meiling wore a virtual reality (VR) headset and held motion-sensing controllers, working in sync with the robot. As she mimicked the screwing motion in the virtual environment, the robotic "student" followed her lead precisely.

This was a scene in an embodied intelligent robot data acquisition pre-training field, dubbed "robot school", in Hefei, capital of east China's Anhui Province.

Located in the Yangtze River Delta digital technology demonstration park, the school officially began operations recently and is training its first batch of "students," over 10 robots from multiple domestic robot development and manufacturing enterprises.

They're being trained for roles in specific working scenarios including logistics box handling, parts picking, home services, and customer guidance in tours and retail.

According to Wu, she and her colleagues input about 200 action sequences per day into each robot.

A robot developed by Chinese tech company VSTC receives training at an embodied intelligent robot data acquisition pre-training field, dubbed "robot school", in Hefei, capital of east China's Anhui Province. (People's Daily Online/Ding Ji)

Based on this data, the robots build their own working models and gradually move beyond hands-on instruction from their "teachers" to achieve autonomous operation.

The school's approach—using real motion data to build machine learning models and finally cultivating skills—is designed to give robots both precision and flexibility.

A screw-fastening robot must not only learn how to tighten screws, but also recognize which screw to tighten and handle various screw types, said Wu, explaining the robots' learning goals.

The key to the training service's effectiveness lies in its physical environment: a near-realistic factory floor. For companies developing intelligent robots, such settings are often hard to access during the designing process, according to Wang Shuai, an executive of the Hefei-based subsidiary of humanoid robotics company Leju Robotics.

A robot moves a storage box at an embodied intelligent robot data acquisition pre-training field, dubbed "robot school", in Hefei, capital of east China's Anhui Province. (People's Daily Online/Ding Ji)

Training in a real-world environment helps robots adapt to unpredictable variables, which can reduce the risks when robots are later deployed in actual production settings, Wang said.

Glitches and mishaps are part of the training process. Robots occasionally break down or fall over, but to developers like Ji Chao, founder of a robotics company based in Hefei, that's a necessary step in the development of truly autonomous robots.

"Our aim is to train robots to generalize from one example to others, so that they can perform reliably no matter the setting—and that only comes through real-world exposure," Ji said.

The "robot school" is also a testbed for data sharing in China's fast-growing robotics sector.

Run independently by the International Advanced Technology Application Promotion Center, it is currently the only public training platform of its kind in the country, according to Sun Dandan, an executive of the robotics division of the center's Hefei branch.

The "school" provides standardized data interfaces that allow various types of robots—industrial mechanical arms, service bots, and more—to access and receive training. It also offers data acquisition services to robot body manufacturers based outside Hefei.

To serve businesses at different stages of development, the center has innovatively introduced three models for the "school": joint operation, independent operation, and service-based support.

"The service-based support model allows robot developers to purchase professional services regarding computing power, training scenarios, and data acquisition for algorithm iteration at reasonable costs, thus creating a virtuous circle," Sun said.