2,000-square-meter permanent embodied intelligence training space lands in Shanghai Zhangjiang

Download the app

+12,000 square meters of permanent embodied intelligence training space lands in Zhangjiang, Shanghai. The Paper reporter Zhang Jing 2026-10-23 11:31 Source: The Paper ∙ Sci-Tech 101 > Listen to full text Font size

To accumulate high-quality datasets and distill reusable skill packages, the 2026 Global Developer Pioneer Conference and International Embodied Intelligence Skills Competition will be held from October 23 to 25 at the Shanghai Zhangjiang Science Hall, where a 2,000-square-meter venue will be renovated to create a permanent embodied intelligence scenario training space.

On October 22, The Paper's sci-tech reporter saw at the Zhangjiang Science Hall that Bailian Group had moved supermarket shelves into the competition arena, with robots organizing shelves, mixing drinks, and offering samples in a supermarket scenario. "We set up the supermarket scenario here to see what level robots can reach in livelihood scenarios at the current stage, and also to verify the actual deployability of the embodied industry," said Guo Yongping, manager of the Commercial Internet Department at Bailian Group.

As of the end of June 2026, more than 70 embodied intelligence training grounds have been built and put into use nationwide, but training still lags far behind practical application; robots that have mastered tasks in the lab often have to learn from scratch when moved to a new production line or new shelves. Real-scenario training is becoming a common new direction for the global embodied intelligence industry. Scenarios propose requirements, robots conduct training, and relying on accumulated data for continuous iteration, mature solutions are replicated.

In supermarket scenarios, there are many types of goods, and when robots grasp them, different products give different force feedback; canned, boxed, bagged and other packaging forms must match different force levels, posing challenges to the flexibility and degrees of freedom of each robot joint. In daily operations, robots also need to restock, avoid obstacles, dynamically identify misplaced products and return them to their places, and check the consistency between price tags and products, which tests robots' precise grasp of product forms, names and specifications, and shelf information, as well as their ability to place items neatly. Xu Qi, deputy secretary-general of the Shanghai Artificial Intelligence Industry Association, told The Paper's sci-tech reporter that data collection is an unavoidable key link in humanoid robot training; high-quality datasets accumulated during real-scenario training can be further distilled into reusable skill packages. At present, the generalization ability of embodied intelligent robots is not yet mature, and the cost of technology transfer is high, so targeted training must be carried out based on specific scenarios to gradually form scenario-based skill packages and industry-level high-quality datasets, and use these to feed back into robot training, ultimately achieving a true commercial closed loop within specific industries.

"We let robots enter supermarkets essentially not to replace people someday, but to supplement employees' daily work. For example, after goods are delivered to stores, receiving and warehousing are done at night; store staff either work the night shift, or organize shelves one hour before opening, or stay an extra hour after closing to organize them. In the future, these tasks can be completed by robots. We hope robots will free people from a large amount of repetitive labor while adding some fun new things," Guo Yongping said. At present, participating teams are exploring new forms such as robot sampling services, but he admitted that whether carrying trays or controlling liquid stability and vibration, robots still lag behind humans.

In recent days, robot manufacturers and scenario providers have worked together to refine the scenario competition tasks, and engineers have continuously debugged and optimized on site. In just one week, the robots' performance has differed greatly from when they first entered on the first day. "At the current speed of technological iteration, these capabilities can be realized quickly. What robot manufacturers lack is scenario providers telling them the real requirements, which is also the original intention behind our work on this scenario," Guo Yongping said.

Currently, Shanghai is building a closed loop of training, competition, production, and finance, creating a replicable real-scenario training model, and striving to bring training, production line verification, and the permanent public embodied examination venue into the same main chain. As the host of the conference and one of the operators of the training platform, the Shanghai Artificial Intelligence Industry Association will evaluate the real task success rate, efficiency improvement rate, safety and reliability, and economic feasibility of complete machine solutions in the form of work skill packages. The conference builds an overall chain through simulated competitions, online simulation, and real-scenario live competitions, allowing robots to test their abilities in real tasks.

This competition features 9 key tracks and 23 events, building a real-scenario training space. Scenario providers such as Yanfeng Automotive, Yingzhong Information, Shanghai History Museum, Shanghai Institute of Quality Inspection and Technical Research, Bailian Group, STO Express, China Life Property & Casualty Insurance, Tianyi Digital Life Technology, Zhejiang Qiushi Institute, and Goertek will bring real needs from production lines, the service industry, and the public sector to the competition arena, genuinely testing robots' job task capabilities. Responsible editor: Huan Yanhong Proofreader: Zhang Yan The Paper news tip: 021-962866 The Paper, unauthorized reproduction prohibited +1 Favorite Report #EmbodiedIntelligencePermanentTrainingSpace View more