China's humanoid robotics sector is expanding its data infrastructure, with the Beijing Innovation Center of Humanoid Robotics collecting more than 3 million data entries within four months of operation.
According to CCTV News, over 300,000 of these entries are open-source. A staff member at the centre said the model could reduce data collection costs for small and medium-sized robotics companies by more than 50 per cent, improving access to training data.
The facility focuses on repetitive task simulation to train large-scale models. One example is fruit washing, where operators control robotic arms to replicate human actions. The process involves picking up fruit, turning on a faucet, cleaning the surface, and placing the fruit back into a basket.
Each robotic arm has six joints, aligned with the movements required for task execution. The sequence is repeated 500 times to generate 500 data entries, enabling the system to interpret the task of cleaning fruit.
Additional training scenarios include changing diapers, organising tools, and transferring products such as bottled water. These processes position the centre as a data production hub, generating training datasets across multiple applications.
Centralised data generation supports cost efficiency. The reduction in data collection costs allows smaller companies to participate in robotics development and scale operations more efficiently.
Similar facilities are operating in other regions. In Shanghai's Zhangjiang Robot Valley, a data centre supports simultaneous robot operation across simulated environments such as homes, supermarkets, and retail settings, allowing systems to adapt to different use cases.
Data services are also developing as part of the artificial intelligence value chain. In Chengdu, companies are focusing on data collection, labelling, and training, supporting broader deployment of AI systems.
The development of data production centres is contributing to the expansion of robotics applications, with structured datasets supporting model training and system performance across tasks such as fruit handling and other operational processes.
Source: Global Times
