Embodied intelligence is now running into a "data wall."

In recent years, rapid advancements in large model capabilities have largely been built upon the vast amounts of text, images, and videos available on the internet. However, embodied intelligence requires data derived from interactions within the physical world.

Collecting this type of data is far from easy, and it is quickly becoming a critical bottleneck preventing the further progress of embodied intelligence.

As a result, the topic of data collection has attracted widespread attention across the industry.

At the 2026 World Robot Conference, data acquisition had already become a central showcase for numerous embodied intelligence companies.

Realman Intelligent Technology (Beijing) Co., Ltd. (hereafter referred to as Realman) is one such company. It has deployed its RealBOT robots into scenarios such as pharmacies, electrical distribution rooms, and food preparation sites: the robots handle restocking and medication retrieval in smart pharmacies, perform inspections in power distribution rooms, and collaborate with master bakers from Beijing Daoxiangcun to craft mooncakes via remote control.

At this year's WRC, a remote operator based in Beijing could control a robot at the Realman exhibition hall to make mooncakes, while simultaneously operating robots located at a factory in Changzhou in real time to complete equipment debugging and production line inspections.

From the perspective of model training, each instance of human takeover essentially represents a real robot trajectory generated by human demonstration. The operator decides the next move, the robot follows the instructions, and the cameras, joints, end-effectors, and other sensors synchronously record the entire process. Compared to completing pre-designed actions in a fixed training ground, this data comes directly from real work taking place in the moment.

According to sources familiar with the matter, Realman plans to deploy its RealBOT series robots to nearly a thousand real-world scenarios through its GLN remote operation network in 2026, continuously collecting real-machine data during actual operations.

At a time when real-machine data is scarce, this approach could open up significant room for growth for Realman.

On the capital front, Realman initiated its IPO guidance process on August 10 this year, with Haitong Securities serving as the guiding institution. The company aims to complete the guidance phase within the year.

Whether Realman can successfully enter the A-share market is now a key point of focus.