Author | Cheng Qian
Edit | Desert Shadow
The cleaning work of the streets and alleys is being fully taken over by robots, and unmanned sanitation robots can be seen everywhere in shopping malls, parks, and roads.
It is certain that unmanned sanitation robots have entered a critical period of commercialization. According to the relevant data of the National Bureau of Statistics and the Ministry of Housing and Urban-Rural Development, the market size of autonomous driving sanitation vehicles in mainland China will reach 270 billion yuan in 2025, and the penetration rate will reach 5%.
Behind the rapid expansion of the unmanned sanitation robot market, it has brought great challenges to the domain controller products, the key technology of the robot computing platform.
The increase in the intelligent demand of enterprises for the autonomous operation ability of unmanned sanitation robots has increased the attention to computing power, cost, stability, etc. Therefore, large computing power has become an indispensable key capability of the robot computing platform, but the computing power of traditional domain controllers mostly ranges from a few TOPS to hundreds of TOPS, and there are few platforms with thousands of TOPS or even larger computing power.
In this context, Yunji Zhixing, which has rich experience in the development of high-end autonomous driving solutions, is marching from a narrow "automotive platform" to a "generalized platform", opening the era of large computing power for the landing of robots in urban environments.
Based on the NVIDIA Jetson platform, Yunji Zhixing has built a robot computing platform, and its domain controller products can integrate four NVIDIA Jetson AGX Orin modules, with AI computing power of more than 1000TOPS.
The company has greatly shortened the development and deployment time of unmanned sanitation robots. Zhou Haifeng, hardware director of Yunji Zhixing, revealed that the product customization integrating four NVIDIA Jetson AGX Orin system modules took Yunji Zhixing from initial contact to final delivery in less than 3 months, which is far faster than more than double the industry average.
Zhidong had an in-depth exchange with Du Dianrong, vice president of business products of Yunji Zhixing, and Zhou Haifeng, hardware director of Yunji Zhixing, starting from the robot computing platform, in-depth dismantling of the difficulties and pain points of large-scale implementation of robots in the urban environment, as well as the innovative thinking and solutions of Yunji Zhixing.
1. Aiming at the multiple challenges of the urban environment, the domain controller with large computing power will be delivered in 3 months
Throughout the robot industry, large-scale landing applications are accelerating.
On the one hand, the core technology and key components of robots have been upgraded at an accelerated pace, greatly enhancing their application potential for road and operation in urban environments; on the other hand, the digital transformation of cities is accelerating, and robots are being fully integrated into it, from science fiction to reality, and robots such as smart mines and urban cleaning have shown a strong momentum of development.
There are many subtypes of robots in the urban environment, including garbage classification, garbage removal, cleaning, etc., taking the sanitation robot, which is the largest in the current market as an example, the scene it faces is complex, and it can be said that it exceeds the emergency situation that human drivers need to face in their daily driving.
Generally speaking, in addition to dealing with motor vehicles, pedestrians, traffic lights, road repairs, etc. on open roads, unmanned sanitation robots also need to complete cleaning tasks in scenarios such as paths without lane lines, lakes in parks, viaducts, and under big trees;
In addition, for the sanitation industry, the cost, maintainability, and stability of unmanned sanitation robots under harsh working conditions (ash, high temperature, low temperature, humidity, etc.) are also key factors, which determine whether the robot can be used to replace traditional manual cleaning.
These challenges and potential risks have also become the key to Yunji Zhixing's domain controller products for urban environments.
In the autonomous mobile robot industry, domain controllers are seen as centralized management and control units for robotic systems. The domain controller is composed of three parts: the domain master control processor, the operating system, the application software and the algorithm, and is equipped with the robot's automatic driving system and supporting algorithms, which can be used to perform tasks allocation, status monitoring, communication coordination and other functions, and quickly mobilize the core functions of the robot.
In the actual scenario, the robot needs to carry out a large number of refined perception of the urban environment, coupled with the intelligent and real-time requirements of human-computer interaction, so that the computing power carried by the domain controller is further improved, and the larger computing power can meet the actual operation needs of the robot.
Based on the above background, Yunji Zhixing has built a robot computing platform with large computing power, which can provide enterprises with customized domain controller product solutions according to different scenarios to meet diversified computing power needs.
Yunji Zhixing's domain controller product can integrate up to 4 NVIDIA Jetson AGX Orin system modules, which can provide AI computing power of more than 1000TOPS, and adopts a highly redundant system design and a highly scalable modular design, which can flexibly support various computing power configurations and adapt to the sensor solutions of high-end autonomous driving platforms.
Among them, Yunji Zhixing domain controller products integrate 3-4 NVIDIA Jetson AGX Orin system modules day by day, with a maximum computing power of more than 1000TOPS, and Benxiao integrates 1-2 NVIDIA Jetson AGX Orin modules, with a maximum computing power of more than 500TOPS. Users can choose Jetson AGX Orin 64GB and 32GB modules based on different computing power and power requirements, providing AI computing power of up to 275 TOPS per machine, and the power can be configured between 15W and 60W.
Du Dianrong said, to put it simply, the most direct difference between daily and Benxiao lies in the computing power configuration and the cooling method of the computing platform. This is also a more direct embodiment of its products for customers in different application scenarios.
In order to flexibly adapt to different application scenarios, the company's domain controller series products have a variety of system interfaces to meet different sensor solutions. At the same time, the autonomous machine autonomous driving system based on the NVIDIA Jetson platform improves the deployment efficiency, stability and reliability of Yunji Zhixing domain controller products.
In addition, Yunji Zhixing has also applied redundancy design in the main controller system, communication system and power system to support fail-safe operation. In terms of mass production and delivery, the domain controller product project development, verification and delivery are in accordance with the V-model process of mass production of vehicle specifications, providing system structure solutions such as water coolers, and the whole set of equipment can meet the needs of rapid on-board and mass production delivery.
A typical case is a sanitation robot enterprise with in-depth cooperation between Yunji Zhixing, which faces the challenges of algorithm complexity, sensor number and type, computing power demand, and working environment diversification in low-speed scenarios when laying out products for sanitation and cleaning. The daily and Leaping series products can not only solve the bandwidth and stability problems caused by the traditional multi-controller interconnection, provide basic computing power ranging from 500~1000TOPS, but also be compatible with all mainstream sensor interfaces to adapt to different deployment scenarios.
Facing the complex scenarios of the urban environment, Yunji Zhixing has integrated the previous technology accumulation and layout into the current challenges of industrial development, quickly laid out the corresponding products, and customized them for different application scenarios to accelerate the landing of unmanned sanitation robots in the urban environment.
2. Diversified computing power, scenarios, and ecological support to solve the problem of robot landing
From the robot computing platform of Yunji Zhixing, the core capabilities of the current robot computing platform for urban environment development can be abstracted, that is, it supports flexible deployment of computing power, supports complex scenarios, controllable costs, and high stability.
The key to building these core capabilities is the technology accumulation of Yunji Zhixing and the NVIDIA Jetson platform.
Ying Fei, head of the Jetson product line at NVIDIA's authorized distributor, mentioned that the robotics industry is undergoing an intelligent transformation led by generative AI, which has achieved unprecedented perception and adaptability. NVIDIA has a long-standing presence in the fast-growing robotics industry, providing state-of-the-art software and hardware AI solutions to empower robots. For example, at the hardware level, the NVIDIA Jetson series has multiple products that support different computing power, product design, external sensors, etc., and in terms of application platform, in addition to the software stack Jetpack, the robot simulation platform Isaac can provide a complete set of solutions from simulation to deployment. (Click "Read More" at the end of this article to download the NVIDIA Jetson eBook for more industry use cases.) )
Specifically, Du Dianrong mentioned that NVIDIA Jetson has high-speed interfaces, faster memory bandwidth, and multi-mode sensor support, which are the foundation of high-performance robot computing platforms. Based on NVIDIA's comprehensive software and ecosystem support, customers can easily develop applications and optimize algorithms, deploy complex models, and quickly integrate various peripherals, making it easier for developers to build high-performance robots on the NVIDIA Jetson platform, greatly accelerating the process of rapid product series construction and rapid integration and delivery of enterprises.
In addition, for various sanitation scenarios and different tasks faced by robots in urban environments, NVIDIA Jetson's complete product system provides support for computing chips of different sizes.
For the application of cutting-edge technologies, the NVIDIA Jetson platform is an ecological environment on both the cloud side and the device side, which can quickly migrate cloud training algorithms to the device side, and can save customers time on the device side as the algorithm is updated and iteratively and large models are deployed.
The full range of products of Yunji Zhixing are connected with the cloud-based operation management, data training, and scenario analysis platforms, forming an efficient and complete automation link.
She explained that the advantage of this for the landing of robots in urban environments is that when the robot encounters some extreme working conditions, it will automatically send the scene data technology to the cloud, and the cloud will quickly update and iterate the algorithm, and then transmit it to the device side to achieve application upgrades.
Therefore, the full range of products of Yunji Zhixing can provide tens to thousands of computing power for various applications, and the general platform can support more subdivided complex scenarios at the same time.
It can be seen that Yunji Zhixing's layout of robot products, the development trend of robots in urban environments, and the layout of the NVIDIA Jetson platform are in the same line.
According to Du Dianrong, Yunji Zhixing has completed the industry-leading unmanned road test, which requires its autonomous driving system to meet certain standards in stability, adaptability and safety, and the vehicle can achieve 1,000 kilometers of continuous human intervention on the open road, and can cope with various accidents and emergencies in order to obtain the corresponding test qualification.
This is due to the company's accumulation of high-end intelligent driving systems in urban scenarios, which can better serve various subdivided scenarios in the urban environment and land them in specific application scenarios of customers more quickly.
Another partner of Yunji Zhixing's sanitation scenario, when developing a small, fully intelligent autonomous driving sweeper based on the daily series of domain controller products, Yunji Zhixing completed the design and delivery of the system-level solution of the air-cooling system in only one and a half months, which is much faster than the average development process cycle in the industry, saving nearly half of the time for the development and deployment of its products.
At present, Yunji Zhixing's domain controller products have been delivered in batches for many fields such as urban sanitation, smart mines, and logistics robots, among which the two major scenarios of urban sanitation and smart mines have been commercialized, which not only directly affects the cost reduction and efficiency increase of customer enterprises, but also becomes an important part of the improvement of Yunji Zhixing's economic benefits.
3. The robot track takes off, from vertical application to embodied intelligence
Throughout the robot industry, embodied intelligence has become the next outlet for AI.
Since the beginning of this year, the frontier of embodied intelligence has emerged one after another. There are not only robot talents, such as the Stanford Robot Show Cooking "Manhan Full Seat", Tesla Optimus Prime incarnated as a nanny to fold T-shirts into "tofu blocks", Figure 01 can easily have a complete conversation with people after being connected to the large model, and is competent for complex tasks such as picking up garbage and cleaning up dishes and chopsticks, as well as the underlying development platform, NVIDIA has built a comprehensive development platform for embodied intelligent players, including the humanoid robot general basic model Project GR00T, based on NVIDIA Thor's SoC's new humanoid robot computer, Jetson Thor, etc......
The powerful ability of embodied intelligence lies in the ability to learn independently from environmental interaction, generate an understanding of the objective world, and use it for the iterative optimization of its own capabilities, matching entities for the landing of large models, and completing the last piece of the puzzle for general robots, which has become one of the hottest topics in the current science and technology circle.
Similarly, Yunji Zhixing is also exploring the application of cutting-edge technologies and conducting research and development for large models.
In the cloud, large models can help customers mark and classify scenes, or operate daily robot fleets, and at the same time, Yunji Zhixing is deeply studying the deployment of large models on the device side to help them build more subdivided and complex urban environment deployment. Du Dianrong revealed that at present, the scene analysis, attribution analysis, and data marking of large models in the cloud have been applied in the business of Yunji Zhixing.
However, compared with its impact on general artificial intelligence, Yunji Zhixing focuses more on the role of this cutting-edge technology in promoting the landing of robots in urban environments. Du Dianrong said that they believe that the application of large models in subdivided scenarios can be regarded as one of the means of intelligence in the sanitation industry to help reduce labor costs.
At the end of the day, Yunji Zhixing focuses more on providing customers with a more user-friendly and easy-to-use computing platform, so that customers can focus more on how to do better in robot applications. Du Dianrong said that in the short term, they hope to use the relevant experience of Yunji Zhixing in the automotive industry to serve customers in subdivided scenarios such as sanitation robots and mining robots, and support them to deepen the application of robots.
Supporting Yunji Zhixing to quickly become an industry leader in the explosive opportunity period of the robot industry is inseparable from the company's previous layout in R&D investment.
Zhou Haifeng said that Yunji Zhixing has a professional software and hardware team, including experts in special directions such as simulation, and most of the employees have more than 10 years of work experience, have deep technical accumulation in the field of automotive electronics research and development, and also have rich experience in large-scale mass production of passenger car projects.
The company has a first-class software and hardware development laboratory of more than 200 square meters, which can cover the whole process of testing and verification of hardware design. "It's also a big advantage that basically all customers value. He said.
In terms of software and hardware design, on the one hand, Yunji Zhixing has improved the importance of hardware architecture design and solved the complexity of system design, and on the other hand, it has decoupled hardware design to solve the problem of universal and reusability of modules on different platforms, and fundamentally ensured the high-quality delivery of designed products.
Yunji Zhixing has laid a solid technical foundation in the field of robotics, coupled with the strong support of the NVIDIA Jetson platform, it has taken the lead in the rapid development of robots.
From robots in vertical application scenarios to embodied intelligence, robots are leading the next revolution in the technology industry, and their huge potential for application in thousands of industries is exploding.
Conclusion: Acceleration of the application of robots in urban environments
At present, robots appear in more and more scenes in urban environments, from sanitation and mining to logistics and testing. Among them, the two major scenarios of urban sanitation and mining are at the forefront of the industry's cost reduction and efficiency improvement achievements and the process of large-scale landing, which has also become the second track of Yunji Zhixing's deep cultivation.
Behind this, Yunji Zhixing's technology accumulation of carrying robots has gradually deepened, which provides an opportunity for it to seize the large-scale implementation of robots that flexibly respond to the urban environment in the future. With the strong support of the NVIDIA Jetson platform and the entry of cutting-edge technologies such as large models, the landing of robots in urban environments may show a more amazing growth rate.