Das Team für Embodied Brain der Tsinghua-Gruppe hat in Summe Finanzierungen im Umfang von Hunderten von Millionen Yuan erhalten. Es setzt sich mit den führenden amerikanischen Unternehmen auseinander und hat bereits bei führenden Herstellern in der Branche Umsetzungen erzielt | Exklusivbericht von Yingke.

Autor | Huang Nan

Redakteur | Yuan Silai

Hard Kr has learned that the embodied brain company "Qianjue Technology" recently completed a new round of Pre - A+ financing. This round was invested by Junshan Investment, Vertex Ventures and Shixi Capital, with a cumulative financing amount of hundreds of millions of yuan. Maple Pledge served as the long - term advisor to provide private placement financing support. The funds from this round of financing will mainly be used for the evolution of core technologies, product standardization and the improvement of industrialization delivery capabilities.

"Qianjue Technology" is an embodied intelligence technology company incubated by a Tsinghua - affiliated team. Most of the core members of the team graduated from the Department of Automation of Tsinghua University and relevant artificial intelligence research institutions. They have long - term and profound scientific research accumulations and engineering transformation capabilities, and the talent density is at the top level in the industry.

As the only domestic enterprise that can be compared with the US company Physical Intelligence, "Qianjue Technology" has taken the lead in realizing the long - term task execution ability of general embodied intelligence in real - world scenarios and has product - level stability.

Currently, the technological bottlenecks of robots in fields such as complex environment perception, dynamic task planning and fine operation execution still need to be urgently broken through. Although the introduction of large language models has significantly improved the semantic understanding and logical reasoning abilities of robots, the integrity of their intelligent decision - making systems still depends on the in - depth integration of multi - modal perception models, behavior control algorithms and software - hardware collaborative systems.

As the core of the underlying technical architecture, the general "embodied brain" undertakes the central decision - making function. Its intelligence level not only directly determines the autonomous execution ability of robots, but also becomes a key factor in expanding the boundaries of scenario applications.

Among them, the "embodied brain" system independently developed by "Qianjue Technology" emphasizes real - time multi - modal perception, continuous task planning and autonomous execution without relying on preset strategies, which is highly consistent with the π0.5 model recently released by Physical Intelligence. The latter takes "function transfer" as the core, that is, the ability to complete complex tasks in untrained scenarios.

In the actual testing process, the embodied brain of "Qianjue Technology" has the characteristics of cross - environment, no preset and long - term operation. It has the ability of on - site demonstration and practical application. The scenario of the household service robot built by it is more complex than that of the π0.5 model, and it can conduct fully open on - site demonstrations. After the visitors issue instructions, the system's long - term autonomous decision - making and interaction ability can adapt to more than 20 types of embodied hardware forms. Taking the autonomous household nanny robot as an example, it can run continuously for several hours until the battery is exhausted.

Simulation diagram of the working demonstration of the robot equipped with Qianjue Technology's embodied brain (Source/Enterprise)

Based on the "brain - like computing" technical path adopted by "Qianjue Technology", while constructing a large decision - making model, it can connect the system closed - loop of perception - reasoning - behavior, form an intelligent central control for robots, and realize a qualitative change of robots from passive execution to active planning.

According to official news, "Qianjue Technology" will release a new round of business - side demonstrations, showing on - site the system's full - process autonomous decision - making ability in open and unstructured environments, covering multiple dimensions such as recognition, judgment, path generation and task execution, forming a complete "perception - decision - action" closed - loop.

Dr. Gao Haichuan, the founder and CEO, told Hard Kr, "The technology we present that can be commercialized is not simply training the VLA model, but reconstructing the input mechanism of robot decision - making with a brain - inspired neural reasoning mechanism, enabling robots to work for a long time relying only on their own sensors and truly having task autonomy. In the long - term open - scenario demonstrations, our system can operate stably and make continuous decisions, working directly in real unstructured environments without being triggered by human instructions."

Hard Kr has learned that the embodied brain of "Qianjue Technology" has been demonstrated on - site for more than a hundred times, with the longest demonstration lasting more than 2 hours. Since its establishment in June 2023, "Qianjue Technology" has completed the iteration from basic framework construction to product - level implementation. Its "embodied brain" has standardization capabilities, supports flexible adaptation and rapid deployment, is compatible with various humanoid and non - humanoid platforms, and has obvious advantages in terms of cost and engineering delivery efficiency.

In terms of commercialization progress, the embodied brain of "Qianjue Technology" has achieved stable operation in multiple scenarios such as household services, logistics and distribution, and commercial operations. It has reached cooperation with many leading embodied robot manufacturers, consumer electronics companies and Internet technology enterprises, accelerating the industrial implementation of general embodied intelligence.

Meanwhile, supported by the achievements of key Chinese brain science projects, "Qianjue Technology" has also joined hands with well - known 3C industry players to build the world's largest real - collected household scenario dataset, completed model training, and deployed it in actual products for operation, demonstrating strong versatility and cross - task adaptability. Compared with traditional solutions that need to rewrite scripts for each type of task, the "Qianjue System" can achieve task abstraction and reuse, and is a truly "scalable standard product of embodied intelligence".

With the completion of the new round of financing, "Qianjue Technology" will further accelerate technological evolution and product implementation to promote the large - scale popularization and application of embodied intelligence.