In recent years, the rapid development of AI big model technology has achieved results comparable to or even exceeding those of humans in some niche fields. It has also promoted the development of embodied intelligence technology and continued to enable traditional physical entities (such as industrial robotic arms, humanoid robots, etc.) to achieve complex interactions with the physical world.

However, there is a large industry gap between the general AI big model and the embodied intelligence industry brain. The core of this is that the general AI big model has weak domain adaptability to embodied intelligence scenarios. Therefore, it is necessary to develop its own dedicated brain in the field of robotics.

Domestic robot brain startups are booming. Ruoyu Technology , located in Shenzhen and established with the support of Harbin Institute of Technology (Shenzhen), is one of the few companies in China that focuses on the research and development of multi-modal large-model robot brains.

Recently, Ruoyu Technology launched the Ruoyu · Jiutian Robot Brain, which has realized group intelligence driven by a multi-modal large model, and verified the technical solution through an unmanned kitchen environment, driving multiple intelligent agents to cook in an unmanned kitchen. Through human natural language interaction with multiple robots, the entire process from ordering to serving is completed, and the long sequence of "cooking" tasks is planned autonomously. Different from the common embodied intelligence technology on the market, the Ruoyu · Jiutian Robot Brain can realize functions such as human intention understanding, high-frequency human-computer interaction, and complex task planning.

Long sequence mission planning (quickly generate complex mission plans)

When a human requests "a serving of stir-fried lettuce", the robot can independently understand human intentions and plan tasks, breaking them down into multiple steps such as "pick up vegetables in the food area", "weigh in the preparation area", "cook in the production area", and "serve in the consumption area" .

the Ruoyu Jiutian robot has a certain ability to generalize . For example, when training the model, it was only told to add garlic, but not to add chili peppers. But if the robot is asked to add chili peppers later, it can automatically understand that garlic and chili peppers are both ingredients, and know to replace garlic with chili peppers.

Ruoyu Jiutian Robot Brain is the group intelligence driven by a multimodal large model, which gives multiple intelligent agents the ability to autonomously divide labor and collaborate.

In simple terms, it is "one brain, multiple bodies". Multiple robots (including robotic arms and AGVs) are centrally controlled by Ruoyu Jiutian Brain. Multiple robots work together to efficiently perform complex tasks, and there are complex dependencies between these different tasks.

For example, two robots are cooking while another robot is picking up bowls and chopsticks. If the cooking time is extended due to a temporary change in the cooking process, the other robot can decide to pick up the next ingredient after picking up the bowls and chopsticks, instead of waiting, thereby improving efficiency. The robot brain will autonomously plan the task logic so that all robots will not stop as much as possible while ensuring the safety of collaboration. "This is difficult to achieve with traditional rule-based teaching methods," said Deng Yuping, Chief Product Officer of Ruoyu Technology.

Swarm intelligence driven by large models (giving multiple agents the ability to autonomously divide work and collaborate)

The Ruoyu Jiutian Robot Brain has also made technical optimizations in the cerebellum that controls the execution level of the robot in order to further promote its implementation. For example, in response to the challenge of difficult perception of diverse environments, Ruoyu Technology has developed a large-model 3D affordance perception technology that can perceive the functional affordance area of objects and can grasp a variety of objects without registration, complex light, or stacking.

3D intelligent recognition and capture (registration-free, stacking, and resistance to complex light)

In response to the challenges of difficult execution of complex actions such as "robots pouring garlic from ingredient boxes" and "using shovels to pour finished dishes from pots", Ruoyu Technology proposed diffusion model-driven imitation learning technology to achieve accurate execution of complex skills and actions. At the same time, combined with multi-modal large model-driven robot multi-arm collaboration technology, the collaborative ability and execution efficiency of the target task have been greatly improved.

Complex action execution driven by diffusion model (no need for on-site teaching, precise execution)

Large model-driven dual-arm collaboration (greatly improving collaboration capabilities and efficiency)

At present, international embodied intelligent brain achievements include PaLM-E of Google's artificial intelligence team, RT-1/RT-2/RT-X of Google DeepMind team, Physical Intelligence invested by Open AI, and so on.

Shenzhen Ruoyu Technology Co., Ltd. was established in 2023 and was incubated by the team of Harbin Institute of Technology (Shenzhen) Institute of Computing and Intelligence. The core team brings together more than dozens of national leading talents and young talents in the current fields of multimedia, NLP and embodied intelligence, covering all directions of large model research and development, including embodied intelligence, content understanding, content generation and large model lightweight and compression.

Ruoyu Technology completed an angel round of financing of over 50 million yuan in March this year. The investors in this round included Dongfang Precision and Kinzon Capital, and Yuanhe Capital served as the exclusive financing advisor. The company said that the funds will be mainly used for product research and development, business expansion and team building.

It is reported that Ruoyu Technology has reached a cooperation with Dongfang Precision Group. Dongfang Precision Group has now become the international leader in corrugated paper packaging intelligent equipment, the domestic leader in digital printing equipment, and the leading company in water power equipment. It continues to expand new business in the smart factory industry and realize the upgrade of intelligent equipment manufacturers to smart factory overall solution service providers. It can provide Ruoyu Technology with rich industrial scene data and downstream corporate customers, and Ruoyu Technology's robot brain can empower Dongfang Precision Group's core intelligent equipment and improve the intelligence level of Dongfang Precision's core products.

Dr. Sun Teng, co-founder and CEO of Ruoyu Technology, said: " The technological breakthroughs of Ruoyu Technology's Jiutian robot brain and robot cerebellum can be further applied to complex processes in the industrial field, such as welding, grinding, and dispensing, sorting and mixed palletizing in the logistics field, unmanned retail, unmanned pharmacies, and three-dimensional cleaning in the commercial field. In the future, it can be further extended to the consumer field. At present, Ruoyu Technology has also worked with industry partners in certain special fields to create more intelligent robots for specific scenarios, freeing people from dangerous, heavy, and repetitive labor, improving the quality and efficiency of the industry, and creating new quality productivity."