More than 100 humanoid robot models go to school
China's newly inaugurated heterogeneous humanoid robot training base will be fully operational in July this year, according to Interesting Engineering. Named the National and Local Co-Manufacturing Humanoid Robot Innovation Center, the facility spans more than 5,000 square meters in Shanghai's Zhangjiang Hi-Tech Park. More than 100 robot models from many companies are undergoing training here to establish a database for industrial, service, medical and agricultural applications.
According to Xu Bin, general manager of the center, the facility seeks to address key challenges in humanoid robot research and development such as general technology, industrial ecosystem and solutions for critical application scenarios. Because humanoid robots from different manufacturers differ in design and functional modules, they create heterogeneous and incompatible databases. Therefore, the center was born to enable large-scale data sharing and use.
In the first phase, the center will train robots with about 45 basic skills such as grasping, collecting, arranging and transporting. According to Yang Zhengye, market systems manager at the center, in these jobs, robots are expected to learn from repetitive motion data collected by their sensors. The facility compiles data on how robots complete step-by-step processes through simulations of daily and professional work. This data is then fed into version development to hone the robot's ability to automatically perform more complex actions.
To collect a database of movements, a team of trainers continuously perform repeated actions, sometimes hundreds of times per day, so that the robot can imitate the movements and improve. Small differences in the shape and angle of objects create unique data that helps the robot adjust to environmental changes.
The training center focuses on 10 key application scenarios, including industrial, indoor and travel service tasks. Those tasks are divided into many small jobs such as folding clothes, assembling or arranging objects, and cleaning machines in toxic environments.
Currently, the center is in the testing phase, generating 20,000 - 30,000 data entries per day. Once fully operational, daily data entry could increase to 50,000, with the goal of exceeding 10 million this year. One of the center's long-term goals is to create a data exchange platform so robot developers can share data for specific situations such as home applications or medical care.
The new data center is proof that China is urgently building an ecosystem for high-end robotics and AI. Through systematic collection and distillation of motion data from multiple robotics platforms, engineers can refine algorithms and share breakthroughs faster.
An Khang (According to Interesting Engineering)