Bosch has joined hands with a Peking University professor to enter the field of embodied intelligent robots.

The world's top Tier 1 player is going to manufacture embodied intelligent robots in China.

Bosch recently officially announced the establishment of a joint venture focusing on embodied intelligent robots, specifically in the industrial manufacturing field where Bosch excels, with a focus on high - tech products such as dexterous hands and flexible robotic arms.

And Bosch's Chinese partner, Galaxy Universal, which has been established for only two years, is already one of the most - watched star unicorns in the field of embodied intelligence.

This time, the century - old industrial giant and the emerging force in embodied intelligence have joined hands, and the trend in the automotive circle is quite obvious.

Bosch Enters the Field of Embodied Intelligent Robots

Bosch's entry into embodied intelligence was announced at the "Open Bosch: Embodied AI Day" event on June 17.

On that day, Bosch China, Boyuan Capital, and Galaxy Universal jointly signed a strategic cooperation memorandum and established a joint - venture company, "Chengmai Boyinhechuang Technology Co., Ltd." (hereinafter referred to as "Boyinhechuang").

The new company will focus on the industrial field where Bosch is good at. It will develop flexible and high - precision solutions such as dexterous robotic hands and single - arm robots for manufacturing scenarios such as highly complex manufacturing, complex component handling, precision assembly, and flexible quality inspection.

Bosch is well - known to everyone. It is a global Tier 1 automotive supplier giant and a century - old player in industrial automation and automotive parts.

While Galaxy Universal may still be a new name to many people.

It is a leading player in embodied intelligence established in 2023, with its headquarters in Zhongguancun, Beijing. Its core team comes from well - known domestic and foreign universities and institutions, with backgrounds in fields such as computer vision, robot motion control, and reinforcement learning.

For example, the founder and CTO Wang He graduated from Tsinghua University and Stanford University for his undergraduate and master's degrees respectively. He is now an assistant professor and doctoral supervisor at the Peking University Institute for Frontier Computing Research, the director of the Peking University - Galaxy Universal Joint Laboratory for Embodied Intelligence, and the person - in - charge of the embodied large - model at the Beijing Academy of Artificial Intelligence.

The co - founder Yao Tengzhou graduated from the Robotics Research Institute of Beihang University, under the tutelage of Wang Tianmiao, known as the "Titan of the Robotics World" and one of the "South Xiang and North Miao". He also worked at the Shanghai Robotics R & D Center of the industrial giant ABB.

△

Just one year after its establishment, Galaxy Universal received support from many well - known investors such as Meituan, BAIC Capital, IDG, and Matrix Partners. It raised 1.2 billion yuan in two rounds of financing, with 700 million yuan raised in the angel round alone, setting the record for the largest angel - round financing that year.

The continuous pursuit in the capital market is not only because of the company's strong team but also closely related to its core technology.

Galaxy Universal's competitiveness comes from its technical architecture of "simulation data - driven + separation of the 'big brain' and'small brain' models".

The innovation lies in the decoupling design, which reduces the dependence between different parts of the system. The "big brain" focuses on high - level cognition such as understanding instructions and task planning, while the "small brain" specializes in motion control and dexterous operations, such as complex actions like grasping.

Two important models of Galaxy Universal support this technical architecture:

GraspVLA: The world's first end - to - end large grasping model, which can achieve seven generalization abilities (referring to the model's performance when dealing with unseen samples) through pre - training with synthetic data, including generalization of illumination, background, planar position, spatial height, action strategy, dynamic interference, and object category, and only requires fine - tuning with a small amount of real data.

TrackVLA: That is, the pure - vision navigation model, which allows robots to operate without mapping or additional sensors.

The technical foundation is built on a tens - of - millions - level simulation data factory. Galaxy Universal will generate a large amount of diverse scenario and action data, breaking through the dependence on real data, significantly reducing training costs, and accelerating iteration.

Bosch is attracted by such a technical route of "simulation training + synthetic data" and the R & D response speed in the Chinese local market.

For Bosch, the technical complementarity between the two sides is maximized. Bosch has the genes for precision manufacturing of robots, while Galaxy Universal can provide the AI decision - making engine for robots.

It can be said that it's a perfect match.

Moreover, Bosch clearly had preparations in advance for entering the field of embodied intelligence. As soon as the joint - venture company was established, the next step was already planned.

On the day when Boyinhechuang was officially announced, the company signed a strategic cooperation agreement with United Automotive Electronics Co., Ltd. (UAES). The two sides plan to jointly build a robot laboratory, "RoboFab".

It should be noted that United Automotive Electronics is one of Bosch's important joint - venture companies in China. It was jointly established by Bosch and China United Automotive Electronics in 1995, and its business involves core components such as engine control units, electric drive systems, and body controllers.

As for China United Automotive Electronics, it was jointly invested and established by eight Chinese enterprises, with SAIC holding 54%. Its legal representative is Jia Jianxu, the current president of SAIC.

It is reported that the RoboFab laboratory will focus on the intelligent perception, operation control, and system deployment of robots in typical automotive manufacturing processes in the field of embodied artificial intelligence, creating a full - process chain from algorithm models to product production.

So Many Ways to Combine the Automotive Circle and Robots?

The model of collaboration between the automotive circle and robot players is not new in the field of embodied intelligent robots.

Simple business cooperation is quite common, so we won't list them here. Let's focus on the deeper - binding models.

According to the cooperation methods, these bindings can be divided into the "joint - venture faction" and the "investment faction". Here are some examples:

Let's first look at the "joint - venture faction" like Bosch. Joyson Group and ZhiYuan Robotics established a joint - venture company, "Puzhi Future", in Ningbo in April this year.

Joyson is also an automotive Tier 1 player. It has long - term cooperation with Volkswagen, Audi, BMW, General Motors, SAIC, Li Auto, NIO, etc., and has been deeply rooted in the automotive industry for more than 20 years.

And Guilin Fuda, which mainly deals with forged steel crankshafts for engines and automotive clutches, also said that after taking a stake in the robot company Changban Technology in March this year, it plans to jointly invest with the latter to establish a joint - venture company for Fuda Changban robot parts.

Actually, there are not many such joint - venture cases. Currently, the participants from the automotive circle are basically automotive suppliers with strong backgrounds.

However, there are many players in the automotive circle in the "investment faction", that is, taking stakes in robot companies.

Take ZhiYuan Robotics for example. BYD, SAIC, and BAIC have all made investments, with BYD holding 2.32% of the shares.

Standard Robots has also been invested in by Xiaomi and NIO. Xiaomi is its second - largest shareholder with a 9.29% stake.

Even Unitree Technology, which became very popular this year, had Geely participate in its just - completed Series C financing.

Of course, some automobile manufacturers have chosen the self - research route to enter the field of embodied intelligence, or adopted multiple routes simultaneously.

Whether a player is on the self - research route can be judged by whether it invests in building its own team.

Some suppliers are also involved in this, such as Digua Robotics spun off from Horizon Robotics.

However, currently, most of them are automobile manufacturers, including Tesla's self - developed Optimus, XPeng's self - developed Iron, GAC's self - developed GoMate, Xiaomi's self - developed CyberOne, and BYD's matrix of various robots through internal integration and external cooperation.

Whether it is self - research or cooperation, it is an obvious trend:

Participants in the automotive circle are scrambling for the window of embodied intelligence.

Entering the field of embodied intelligence essentially means transferring the AI closed - loop ability of "perception - decision - control" to the manufacturing of embodied intelligent robots. In fact, it is also a new attempt to extend, transfer, transform, and industrialize some knowledge in the automotive industry chain.

After all, embodied intelligence is the next trillion - dollar market and has connections with the automotive field. Most players in the automotive circle will invest in potential opportunities to grab a "ticket" for a good position.

This article is from the WeChat official account “Intelligent Vehicle Reference” (ID: AI4Auto), written by Jessica, and published by 36Kr with authorization.