Beijing's Shijingshan district has launched the third phase of its humanoid robot training center, expanding its capabilities beyond data collection to include tactile sensing and virtual-real training.

The new phase, unveiled on Friday, introduces high-precision tactile data acquisition and simulation-based data generation. It aims to help robots handle delicate, contact-heavy tasks more efficiently while reducing testing costs.

The center addresses longstanding industry challenges, including the difficulty of transferring skills across different robot designs, a shortage of high-quality multi-modal datasets, and the lack of standardised data annotation.

Dong Jianqiang, senior engineer at Beijing Tashan Technology Co, a project partner, said the initiative seeks to provide more standardised solutions.

"Different robot designs are often trained separately, making skill transfer difficult," he said.

In practice, engineers recreate real-world scenarios — from factory floors to homes — and select appropriate hardware. Tasks are tested in simulation before human operators collect real-machine data, which is then used to train autonomous robots. Applications range from industrial production and agricultural processing to hotel services, household cleaning, smart assistance, and medical operations.

The center is currently training more than 100 heterogeneous robots across industrial, commercial, and domestic scenarios, with a focus on what engineers call the "last centimeter" of physical interaction, he said.

Covering over 10,000 square meters, the Shijingshan facility is now China's largest humanoid robot data collection and training center, according to local officials.

Yang Cheng contributed to this story.