"Qianjue Technology" Completes Tens of Millions of Yuan Angel Round Financing for the Development of an Embodied Brain Benchmarked against Physical Intelligence | 36Kr Exclusive

Author: Qiu Xiaofen

Editor: Su Jianxun

36Kr has learned that the innovative enterprise "Qianjue Technology", which focuses on the embodied brain, has completed an angel round of financing of tens of millions of yuan, led by InnoAngel Fund, and jointly invested by Tsinghua Alumni Seed Fund, TusStar Ventures, Jiushang Capital, Gong Hongjia's family office Jiadao Gongcheng, etc.

This round of financing will be used to increase R & D investment, promote product iteration and market expansion, and provide customers with more advanced and reliable embodied brain solutions.

The embodied intelligence can be divided into two parts - the brain part responsible for decision-making and the cerebellum responsible for motor execution. However, in the more than one year since the upsurge of embodied intelligence, the vast majority of domestic companies have focused on the research of the cerebellum, and as a result, motion control and complex operation models such as RT, RDT, and WALL-A have been born.

In contrast, The development of the domestic embodied intelligent brain is limited, basically remaining at the stage of directly calling the multimodal base model, and there are few truly "embodied brains" for the robot industry and at the product level.

As a key element to promote robots to enter general application scenarios, the embodied intelligent brain has recently received much attention in the industry. A representative case is - at the beginning of November, the US-based embodied brain startup "Physical Intelligence" completed a $400 million financing led by Amazon founder Jeff Bezos and OpenAI, and its valuation quickly soared to $2 billion, increasing sixfold in three months.

"Gao Haichuan, the founder of 'Qianjue Technology', told 36Kr that the academic exploration in this field has a long history. In the past seven years, the research of the Tsinghua Brain-like Computing Research Center has covered 341 important papers of the founding team of Physical Intelligence, and integrated other global cutting-edge research routes in this field. Eventually, the selected technical route coincides with the technical research route of Sergey Levine, a core member of 'Physical Intelligence', and the technical route of 'Qianjue Technology' is directly benchmarked against it.

The core technology research and development of 'Qianjue Technology' is supported by the Tsinghua Brain-like Computing Research Center and the VIPLAB of the Department of Automation, drawing on the "brain-like" technical route and applying it to the perception, decision-making, and control of robots.

The current generative large models cannot solve the problem of the implementation of embodied intelligence - although the existing large models have achieved general intelligence in the network environment, when it comes to robot application scenarios that involve interaction with the real physical world, they still face huge challenges such as the lack of fine-grained perception information of objects, the inability to perform multi-step dynamic reasoning and decision-making, hallucinations, difficulty in obtaining operational data, and high local computing power consumption.

In response to the above problems, 'Qianjue Technology' has created the only product-level robot perception and decision-making large model in China, which can work fully autonomously and dynamically respond to environmental changes, allowing general robots to cross forms, environments, tasks, and objects, and achieve true generalization.

For example - when a mechanical dog is integrated with Qianjue's brain, just open the box and press the power button, and the mechanical dog will actively jump onto the sofa, shake hands with people, and get close, making autonomous cognitive and behavioral decisions in a completely unfamiliar environment, without the need for human commands or the need for "professional machine training" as usual.

In terms of hardware, 'Qianjue Technology' builds a "Brain Dock", a robot-specific end-side software and hardware integrated computing solution, around the Tianjic series of brain-like chips, and takes the domesticization route to get rid of the dependence on cloud computing power and NVIDIA computing power. It is understood that the current "Brain Dock" supports 13B embodied large model reasoning, and can be compatible with both Qianjue's brain and the cerebellum models of partners, with performance comparable to "NVIDIA 4090", and the power consumption is also reduced by 1-2 orders of magnitude.

"Qianjue Technology" Product

In terms of the team, Gao Haichuan, the founder and CEO of "Qianjue Technology", has been the team leader since 2018, leading the brain-like dual-arm robot team to design multiple dual-arm autonomous decision-making robots from scratch; the chief technical advisor, Professor Chen Feng, is one of the first batch of brain-like research experts in China and undertakes many major projects of the China Brain Project. By October 2024, Qianjue Technology has accumulated an embodied perception and decision-making pre-training data set of hundreds of millions of scales, covering a wide range of application scenarios.

Regarding this investment, Wang Sheng, a partner of InnoAngel Fund, said that InnoAngel is very optimistic about Qianjue Technology, not only because it is the first and only enterprise in China that focuses on the embodied intelligent brain, but also because Qianjue Technology's exploration in brain decision-making intelligence is of great value to the improvement and coordination of the robot industry chain. The entrepreneurship of Dr. Gao Haichuan and the team is not only the commercial practice of a capable team, but also will promote the formation of a consensus on the real robot "brain". We are glad to see Qianjue Technology filling this gap.

end

end