In China, a growing number of robot trainers are training their “steel students” to better equip them for real-life applications.
“Every seemingly simple movement requires hundreds of repetitions of practice,” said Yan Jie, 26, a robot data collector, also known as a robot trainer. His robot “student” is learning to pick up a pillow from the sofa.
Driven by his passion for the robotics industry, Yan joined the Lingcifang Robot Data Collection Centre last October and has trained a variety of robots since then.
The collection centre is located in Hefei City, Anhui Province. Dozens of young people like Yan have their own robot “students” there. They guide the robots to perform basic training such as organising items, wiping desktops, and sorting parts. Their goal is to make these “steel students” understand instructions and do good work.
Typically, one robot is paired with two data collectors: one provides hands-on guidance, while the other observes and collects data in real time.
Train Robots in Real-life Scenarios
At the Embodied Intelligence Robot Training Ground in Yaohai District, Hefei, data collector Zhu Yangguang, 22, and his colleagues were focused on training robots for supermarket scenarios.
The training ground was put into use on Jan. 8th, with an area of 700 square meters. It is a 1:1 replica of real-life scenarios such as homes and supermarkets.
“Supermarket environments are far more complex than factories,” Zhu explained. “For instance, when instructing a robot to grab packaging bags from shelves, the size, weight and placement angle of the bags vary every time. The robot must not only perform rigid operations, but must have the generalisation ability to cope with changes.”
As he adjusted the robot’s grasping posture, he added, “There are no shortcuts to developing these capabilities. It all depends on accumulating massive amounts of repetitive data to help the robot learn gradually.”
More than 10 companies have already used this training base for robots. This training model of combination of virtual and real, using massive scene-specific data to drive skill development, is providing an efficient pathway for the rapid evolution of robotic capabilities.
“The joy of this job lies in constantly discovering and solving problems,” said Zhu. Through repeated debugging and training, their goal is to enable robots to replace humans in completing dangerous and heavy tasks.
At present, Hefei has gathered more than 200 upstream and downstream enterprises in the robot industry chain. The city has established a comprehensive R&D and manufacturing system covering core components for the robot’s “brain”, “cerebellum” and main body.
Written by Sha Liu, additional reporting by Ecns.cn.
If you liked this article, why not read: Unitree Robotics Opens Its First Physical Store in Beijing
