CATL's co-founder has invested in an embodied intelligence company. | Exclusive from 36Kr.

Author: Qiu Xiaofen

Editor: Su Jianxun

36Kr has learned from business registration information and insiders that the embodied intelligence company "Qianxun Intelligence" has completed its Angel + round of investment. The investor is Borui Capital (founded by Li Ping, the co-founder and vice chairman of CATL). It is worth noting that this is also the only embodied intelligence company that CATL has invested in so far.

"Qianxun Intelligence" was founded by Han Fengtao, the former CTO of Luoshi Robotics, in February 2024. Since its establishment, it has completed three rounds of financing in the past 9 months. It is understood that the company's new round of financing is still in progress.

The automotive industry scenario is another crucial application scenario for embodied intelligent robots after leaving the laboratory - the automotive and its upstream supply chain production lines are labor-intensive industries. Taking CATL as an example, according to the annual report, the number of production workers currently reaches tens of thousands, including a large number of overseas workers.

In the future, with the spread of the aging population trend, the willingness of young people to work in factories is decreasing, and the labor costs of factories will continue to soar.

The same situation is also happening overseas. As Chinese automotive companies continue to expand globally, they face higher labor costs overseas and the challenges of localizing. The introduction of humanoid robots can solve these problems.

Since 2024, most embodied intelligence manufacturers have placed their products in automotive factories for verification. For example, the humanoid robot company Figure AI has sent its latest Figure 02 to BMW's factory in Spartanburg. Musk has also sent Optimus to the Tesla production line to handle batteries and made the bold statement of "deploying 100 robots next year".

At the same time, the enthusiasm of automotive companies for embodied intelligence is also continuously increasing. 36Kr has learned that currently, automotive companies such as Tesla, Toyota, BAIC, SAIC, Xiaopeng, BYD, and Li Auto have entered the 赛道 (This word seems to be a mistake or not clear in the context. It might be intended to mean "track" or "sector". Without further context, it's difficult to provide an accurate translation. For now, I'll leave it as is.) by either developing robots in-house or investing in popular companies in the sector.

However, the combination of embodied intelligent humanoid robots and the automotive industry scenario is not that simple.

First, at the software level, for embodied intelligent robots to be implemented in automotive factories, a very high accuracy rate is required. In this regard, "Qianxun Intelligence" has launched a "Reinforcement Learning Based on Prior Knowledge of Large Models" framework, which can help the robot's brain continuously practice to improve the accuracy rate in the real world and thereby meet the high success rate requirements of the factory.

In addition, when implementing in the factory, how to complete the task with an appropriate cost of data collection is also a challenge.

It is understood that "Qianxun Intelligence" has also accumulated certain experience in high-sample-efficiency imitation learning. Through the "SGRv2 Framework Network Structure Design", the embodied intelligent robot products of "Qianxun Intelligence" can simultaneously combine the geometric information of the scene and the semantic information of the objects to complete high-data-efficiency behavioral learning training - compared with traditional algorithms, the data efficiency is improved by more than 20 times.

Secondly, the high cost has always been a key constraint that hinders the implementation of humanoid robots in automotive production lines. Currently, the price of a complete humanoid robot is basically as high as millions or even several millions.

And it is understood that cost control is a major competitive advantage of "Qianxun Intelligence". This team has previously managed the commercial implementation of tens of thousands of robots and has accumulated rich experience in the design and manufacturing of the entire robot. Therefore, it can produce products that meet industrial-level requirements while controlling the implementation cost.

36Kr has learned that the robot brain of "Qianxun Intelligence" can also simultaneously support more hardware adaptations, allowing a model to be migrated on multiple different embodied carriers. And the "one brain, multiple forms" implementation mode is also more conducive to reducing the cost of the automotive industry production line and meeting the different demands of actual factory applications.

end

end