Excerpt from: Yan Zhi Zhi Car Vision
Author: Wen Li
Intelligent networked vehicles are moving towards hardware convergence, software-defined and data-driven, and the operating system is the soul of automotive software, is the current operating system subject to foreign countries? Two years ago, China proposed the top-level design of the autonomous driving operating system, how is it progressing? A range of questions need to be answered.
Dr. Shang Jin, General Manager and Chief Technology Officer of Guoqi Intelligent Control (Beijing) Technology Co., Ltd., Chief Technical Expert of the National Intelligent Connected Vehicle Innovation Center, Leader of the Basic Software Working Group of the China Intelligent Connected Vehicle Industry Innovation Alliance, and Visiting Researcher of the School of Vehicles and Transportation of Tsinghua University, it is most appropriate to ask him to interpret the above issues.
Why hardware convergence, software-defined, and data-driven?
"Today's hardware convergence, software-defined, and data-driven phase, how will intelligent connected cars evolve? What is the future direction? ”
Dr. Shang Jin pointed out that this view is a convergence of industry experts, but it does not originate from the automotive industry. The development of the ICT industry over the past 30 years is a testimony, and smart cars are being reinvented.
Chips, hardware, software or operating systems, data and cloud computing represent the focus and hotspot technologies of today. The 12 words "hardware convergence, software-defined, and data-driven" are a summative prediction of the industry's process from zero to maturity.
Chips are very important, there is no software without hardware, hardware convergence is the development characteristics of technology, but also the embodiment of industrial concentration. From the PC, server chips can be seen that the giants invest a lot, and in the end there is no way for small players to live.
Like the ICT industry, chip architecture, business models and industrial development have converged, and some small processors have become centralized domain control and more and more integrated; AI elements in large SoCs will occupy a lot of computing power, but cars are also inseparable from general-purpose computing and real-time controllable chips. The hardware in the chip architecture also includes different AI, ARM, and real-time heterogeneous cores, which are distributed on a centralized basis, forming multiple chips or even multiple boards.
"There have been many things about software-defined networking and software-defined things for a long time, how to understand software-defined cars or smart cars?" - Journalist
Dr. Shang Jin believes that software definition is a change or trend in the methodology of vehicle functional design and implementation under the trend of centralized intelligence vehicles. The essence of software definition is not to implement it in code, the reason why software is mentioned is because the system architecture is most closest to the application in the entire vehicle function. Software-defined means that a smart car that becomes a complex system requires a better architecture.
The first three things to make a good product are: architecture, architecture, architecture, but because of complex hardware, complex software, software definition is particularly reflected in three aspects: first, the entire system architecture is not a centralized architecture, including the new industry chain under the centralized architecture, hardware, operating system, application development, all need architecture first; second, the main function is achieved with software, automatic driving to use AI chips to achieve perception, but it is not the main function, only a small part of the entire automatic driving work, But it requires a lot of data, and even a certain degree of difficulty.
More importantly, we must have system products and architectures that support and reflect the value of software. The value or advantage of software has several aspects, the first is rapid function development, and the second is to ensure efficient development in the case of efficient, high-quality, high real-time, and high security, which can be updated and easy to iterate, such as OTA.
Traditionally, the automotive industry believes that software definition is the Internet, without taking into account the traditional process development system, in fact, the real software engineering or complex system software pays more attention to the process, since it is necessary to use software to implement it, it is necessary to ensure that more functions are developed on the basis of the best architecture, and the iteration in the life cycle is more efficient, the premise is high real-time, high reliability.
"As part of the development and testing of the car, using simulation to train autonomous driving models and verify more scenarios is also data-driven, how to understand the true meaning of data-driven?" - Journalist
Dr. Shang Jin replied that simulation is data-driven, but it is not yet comparable to data-driven in the true sense, and it is not the new core that smart cars need to consider. Data-driven is more outward-looking and places more emphasis on driving. Now the vehicle collects data and puts it in the cloud, and the data drive is mainly around the design of intelligent car products, how to use data to improve the quality of vehicle products, including real-time. Data-driven more emphasis on how to apply cloud computing to vehicle design, that is, the engineering design of vehicle products, which is based on data to improve the design, but also can be extended to data centers, and the cloud computing framework is used to support the functional development of intelligent vehicles.
He explained that the price of new cars varies from year to year, especially the limited intelligence within the boundaries of low-end bicycles, and the problem of ownership; software needs to be upgraded during the life cycle, especially to ensure that the vehicle is used for 5 to 10 years. In a few years, the software will definitely surpass the current hardware, no matter the price, it is impossible to embed the hardware in 5 years, and it is impossible to know what hardware will be in 5 years.
These all point to the same thing: after the software-defined and hardware convergence in the car, go to the cloud to really break through the physical boundaries of bicycle computing. This requires the use of edge computing, cloud computing, which is China's chelu cloud. Therefore, the roadside and the cloud are computing resources, and the vehicle is not just a cockpit, etc., but needs to be guaranteed to be reliable in real time, so data drive cannot passively understand ICT progress only as rigid data and existing models, but is not limited to single development, using cloud computing or data centers to drive vehicle design. In fact, like software-defined and autonomous driving, it is all around the core functions of the vehicle, although the cockpit is also data-driven, but it is necessary to go one more level to consider what the end point of the smart car product should look like.
"No one knows what the hardware will look like in 5 years, and how will it support this software?" - Journalist
Dr. Shang Jin said that the bicycle may not know, but there will be edge computing, edge cloud or 5G, 6G high-speed communication, computing hardware on the edge cloud, because there is 5G or 5G optimization, the hardware in the car and outside the car is no difference, what is important is the extension of the software and operating system. So, data-driven is more compute-driven, but if it's just compute-driven, there's bigger hardware now. What Guoqi Intelligent Control hopes to emphasize is to start from the end point - the entire industrial structure and industrial ecology. Now the goal of smart cars is very simple, it takes 5 or 6 years to turn the mechanical product of the car into the same level as ICT. Like today's computers, notebooks and mobile phones, any ecology is integrated with cloud computing, and it is necessary to integrate it into the vehicle platform around the target function of the vehicle, not only the cockpit or the Internet of Vehicles, but the larger range of automatic driving or vehicle control power systems.
Cars have higher requirements, but they are the real biggest application market for 5G. 5G has high bandwidth, low latency and high stability, now high-speed rail with 4G to watch video has no problem, it is difficult to imagine what 5G has a relatively large application, in fact, the larger industry application is high-speed moving vehicles. The vehicle control system must be lower latency, not only automatic driving, but also the chassis and power system, so the vehicle is the largest application market for 5G.
Why should autonomous driving have a good operating system and architecture?
"From the computer to the current mobile phone, the core software is the operating system, so many years should be the monopoly of foreign operating systems, and the current car is also like this?" - Journalist
Dr. Shang Jin denied it, saying that as the definer and forerunner of the intelligent driving operating system, Guoqi Zhikong recently won the "2021 Top 10 Most Investment-worthy Enterprises in China's Automotive Travel Industry", which proves the significance of what it is doing.
The core product of Guoqi Intelligent Control is the operating system, to be precise, the intelligent driving/automatic driving/intelligent car operating system. The operating system is the core of the product, in the past other industries have proved that the definition of the operating system itself and industry-related, if the core of the smart car is only QNX, Android, Lunix, etc. used in the car now, it is wrong, because these are already mature things, everyone is using, but why do Volkswagen and other car companies still feel that the ability of the car is still far behind? Because these are not cores, not the defining intelligent driving operating system.
"Does that require autonomous control?" - Journalist
Dr. Shang Jin said that Guoqi Intelligent Control is not for the sake of autonomy and autonomy, the lack of self-research to do, the first thing to do must be the industry urgently needed, can play a central role; the second need to clarify that China's nuclear is defined for automatic driving, this formation of the national standard system of the core is not "the enemy has me", and no one has, so the start is industry leadership, industry innovation.
The current situation is that Tesla, which adopts a closed system, is riding the dust; the industry lacks ICT and vehicle integration, and some OEMs are also doing their own software companies, but the products have not been made, let alone mature. China and the United States, which have the deepest understanding of ICT, are more likely to propose and make such operating systems, while traditional automotive leading countries are not necessarily prescient. Even foreign AUTOSAR (Automotive Open System Architecture) cannot reach this height.
"So, how do you define an autonomous driving operating system?" Why do many OEMs feel so far away from smart cars? --Journalist
Quite simply, the industry lacks AI chips, traditional car chips have them, and smart chips are not these chips, and need to have large and complex hardware platforms that support operating systems. As for the operating system, it does not need any profound definition, in plain language, it is to support all hardware chips or hardware platforms, and can support the basic software of application development. The application determines the operating system, and this thing developed for the autonomous driving application is the automatic driving operating system. Since it supports such a core popular application as automatic driving, it is necessary to use customized applications and customized development to achieve the core role.
"Should this operating system be a universal platform?" - Journalist
The real successful marketization of the operating system must be used by everyone, in addition to the cockpit, chassis, power operating system, it mainly focuses on the development of autonomous driving applications, and finally should form a single operating system for the whole vehicle to support all the above application development.
△Product architecture of Guoqi intelligent control intelligent driving operating system
The most difficult and complex thing that Guoqi Intelligent Control is promoting is the automatic driving operating system. Intelligent car operating system must be in line with the core positioning of smart cars, the current kernel is far from reaching, it is obvious that in terms of intelligent performance, car companies have not caught up with Tesla because of the use of kernels or AUTOSAR.
The world is doing the integration of ICT and cars, and is doing basic software, but it is not yet mature, and it has not landed, so there is no need to have a sense of urgency that can be controlled autonomously, and the sense of urgency lies in the urgent need of the industry.
What is the progress of the top-level design of China's autonomous driving operating system?
Two years ago, the mainland proposed the top-level design of China's automatic driving operating system, and Guoqi Intelligent Control has only been established for a year and a half. Dr. Shang Jin said that the company's early gestation time is relatively long, including the definition of the entire structure, and the company is registered after having a positioned market and a product architecture definition suitable for this market. So far, the company has released three product versions of 1.0, 1.5 and 2.0, which basically realizes the industrialization of the intelligent driving operating system that was originally defined and now becomes the core of the national standard. From the perspective of industrial landing, it is now undergoing mass production and development with five or six OEMs, and will be loaded next year.
"The main engine factory is mainly domestic?" - Journalist
Basically, it is a domestic independent brand, because to some extent, the independent brand has higher requirements for independent controllability, but it is not limited to the independent brand.
"Can the operating system made by Guoqi Intelligent Control solve the safety problem?" - Journalist
The function of the operating system is to achieve or ensure the real-time nature of the vehicle system or automatic driving system, safety and application customization development of the main engine plant. Security is the most core thing through the operating system, covering functional safety, expected functional safety and information security, these aspects of consideration, design and implementation are indispensable parts of the operating system.
Another focus is on data security, which must be implemented inside the operating system. A mass-produced autonomous driving operating system must contain data security protection functions for intelligent driving, which is a must-do. The innovation of information security in intelligent driving also includes the integration of real-time, reliability and intelligent driving OS, without adding additional costs to ensure the reliability of the entire system.
△Data security protection demonstration based on Guoqi Intelligent Control Intelligent Driving Operating System
In addition, the amount of automatic driving or intelligent car data is very large, there are many sensors, several T data per day, the need for large-scale data processing, the requirements for data processing safety are higher; and the car is mobile, the mainland has a lot of data classification and grading access standards, large-scale massive data and mobile data, including the environment inside and outside the car, and even the data actively collected by the camera, are the characteristics of this data, which has not been encountered in the data security industry before.
The smart car is not only a storage device, but also a computing unit, in which the data is reprocessed, there is a lot of cooked data, there is a lot of new data, and there will also be data security issues. In addition to strong coupling with autonomous driving, the biggest challenge is to create new dynamic data security issues.
Due to too much people and geographic information involved around, the biggest demand for privacy protection may be smart cars. Other aspects of privacy protection have corresponding regulations, mainly for people, but the vehicle is not the same, it will enter a lot of sensitive areas, involving sensitive objects, unconsciously collected data, which is a new challenge for smart cars.
"Can these issues be solved at the operating system level?" - Journalist
The landing point of innovative technologies to meet these challenges is in the operating system, and the entire system must ensure real-time, reliability and low cost, which is the challenge of integration and integration, so people who do data security must understand the automatic driving operating system and know how to integrate into a product. This is a feature of information security products, and its real landing is inseparable from the deep understanding and integration of the protection object, in order to ensure that the whole is a product, especially the real-time and reliability of the protection object when it is a 2B product.
Where is the positioning of Guoqi Intelligent Control in the supply chain?
Dr. Shang Jin told reporters that Guoqi Intelligent Control is a technology start-up company and a national platform enterprise, the main task is "wisdom integration, China solution", the main products are intelligent cars lack of core and less core. The core product of its definition, the autonomous driving operating system, is also the top-level design of the entire industry and will serve all OEMs. Its operating system supports chips of different sizes, first of all, open, decoupled, standard. Open products are soft and hard, but not full-stack development, but suitable for oem custom development on it. Open is the interface or interface between the operating system and the hardware is open, can integrate different chip hardware forms, the above application development is also the same. Decoupling refers to the operating system inside the kernel, AUTOSAR, software algorithms, etc., the boundary is also open, decoupled, other manufacturers can also do a part, by the National Automobile Intelligent Control to do the integration of the core part, a total of a main engine plant required operating system, which even has the contribution of the main engine plant. Openness is not only at the border, but also includes parts inside the operating system. Such an operating system is based on the contribution of industry chain manufacturers and can be suitable for all models.
"What is the relationship between these products and national standards that you are making now?" - Journalist
Because the lack of core and less core is an urgent need in the industry, Guoqi Intelligent Control has done some work in the top-level design, the operating system national standard system is directly promoted by the National Intelligent Networked Vehicle Innovation Center, and the National Automobile Intelligent Control has played a leading and contributing role, and has done specific implementation work, including the top-level design of the national standard of the interface between the layers. Of course, each national standard also has the participation of all OEMs and industrial chains. In this regard, a lot of things that Guoqi Intelligent Control has done are aimed at the entire industry, especially the synergy of industrial resources, with innovative technology and system integration to continue to lead and support the development of the industry, and enhance the position of mainland intelligent networked vehicles and related industries in the global value chain.
"What are the features of several of your products?" What needs can be met? — Journalist
Now there are two types of products, four units, one type of product is in the car, hardware and software operating system integration, the main engine factory can develop and load on the car; the other type of product is outside the car, such as edge cloud. The so-called four major units are actually composed of four units of two types of products - intelligent car operating system (ICVOS), intelligent automobile domain controller (ICVHW), vehicle-cloud collaboration basic software (ICVEC) and information security data security (ICVSEC). In-car products include OS and hardware, as well as third-party hardware, as well as self-developed market segment hardware; there are three units such as data security or information security inside the OS; the off-car OS is almost the same, including data, car cloud basic platform, equivalent to cloud hardware.
"Is the outside product often referred to as roadside equipment?" - Journalist
It is true that it can be deployed on the roadside and edge clouds, but if the roadside network connection can be upgraded to another level, it needs the standard open system of the roadside, as well as various open architectures and high real-time and high-security implementations. The roadside has perceptual hardware, operating systems and algorithms, which cannot be done in a closed house, and it should also be a standard open system, so that many things can come in. In addition, if it is only provided to the car to do intelligent driving assistance reference, the meaning is not big, it should play the same main role as the sensor in the car, and its realization requirements are high real-time, high reliability, and the OS in the car meets these requirements, which can be partially extended to the roadside. It can not be called an OS, but it is a standard and open real-time, reliable, secure architecture or the core product that implements this architecture.
How should Tesla's success be interpreted?
"At present, some domestic test areas and pilot areas are testing autonomous vehicles on the road, what do you think about this?"
This is a good proof that China's connected vehicle program and architecture have landed in the demonstration area, and all localities are paying attention to and recognizing this development direction, which is a good thing. If we can really catch up with Tesla, we must use China's infrastructure to promote the development of vehicle products.
"Do you mean that Tesla is focusing on bicycle intelligence, while China is doing better in terms of vehicle-to-road collaboration?" - Journalist
There is nothing wrong with Tesla being a bicycle intelligence, but its premise is also that the Infrastructure of the United States is not enough, so it feels that there is not much need to do anything else. If there is enough infrastructure, it will also think more. China is indeed a networked V2X, but everyone's interpretation of V2X is not complete enough, too much emphasis on its results: as long as there is a roadside, as long as there is communication; pay more attention to the demo, very little attention to how to improve the vehicle product design change or absorption of the driving effect, that is, data-driven. It is not enough to understand V2X as a roadside device, but it should be added that an important part of data-driven or networked is to feed back the core functions of the vehicle through the off-vehicle ecology, not just the mature applications of the Internet of Vehicles or the cockpit. The industry does not understand the role of intelligent driving domain, chassis domain, and body domain, and it is important to break through the boundaries of bicycle computing and sensing, while still maintaining and supporting the real-time, reliability and safety of vehicle products. In addition, roadside cameras or other sensors should ensure that the safety level supports the intelligence of bicycles, and truly use L1 and L2 hardware to achieve L3 without more hardware costs.
"Robotaxi and passenger cars are two different technical routes, which is better or worse?" - Journalist
Whether this is a technical dispute from L4 or L5, or from L1 and L2, the hottest discussion in previous years is Tesla and Waymo who can succeed? Today, it can only be said that Tesla succeeded, because Waymo has not been pushed into operation, but Tesla has not only succeeded in building cars, but also autopilot is also going to L4.
The problem is that at present, there are still many theoretical bottlenecks in the real realization of L4 and L5, such as strong artificial intelligence and how to ensure AI safety. In terms of industry promotion, there may still be some problems with the size of Google Waymo. On the other hand, from the perspective of the market, everyone has gradually realized that assisted driving or L2 and L3 are also just needed, and there is also a large market, which will also drive the design of intelligent cars. Intelligent network connection is the second half, not necessarily driverless to promote, L3 centralized domain control or large domain control can also drive the transformation of intelligent vehicles.
Tesla's high market value today is not only the L2 or L2+ function with the largest deployment, but the entire vehicle product design has reduced the blow of several dimensions to the peers, and the autopilot function behind it is only a small part, but it has driven the overall architecture, chip, and operating system of the vehicle, supporting more, better and more efficient design and wider use of OTAs.
Research on autonomous driving will continue, but the automotive industry is not the end of the revolution, let alone its main route. People need L2, L3, automatic parking, which is a bit like laying eggs along the way, you can find wider roads, and you will lay a lot of eggs. If it is really compared with Waymo, Tesla's some scene cars are not as good, but it seems that this is not the case, what everyone really cares about is how the vehicle is redesigned, or how the core functions are designed, as well as the ecological changes brought by the ecological changes and the ecological added value brought by the data.
"The above mentioned the networking, intelligence, and electrification in the new four modernizations of automobiles are very clear, and there is sharing?"
Yes, sharing is not necessarily unmanned, sharing is not necessarily without a steering wheel, what is needed is actually the last 1 km, L3, L4 can achieve sharing. Now everyone is learning tesla, including driverless companies, which is not to say that Tesla's driverless and self-driving routes have won. It only drives the product with the market, or guides the market with the product, and the market landing in turn improves the technology, which is a better model innovation.
Tesla has its advantages, but this also proves a problem from another level, there is no doubt that the deployment volume is large, the use is frequent, on the one hand, the product is really good, there is more data, it can definitely train the promotion system, including some extreme cases; in turn, such a large data has not achieved a complete L4, indicating that there is indeed a theoretical bottleneck, or the relationship with the data is not so large. From the perspective of the industry, Tesla is still in accordance with the laws of the market, not stuck to the goal of automatic driving, and is building automotive products.
Tesla is indeed ahead, but why can't it achieve the L4? This question is worth thinking about!
END