The current status and future of the autonomous driving OS market - Zhihu (zhihu.com)
Author: Su Qingtao
From the perspective of car companies, QNX is a bit like Mobileye - although "easy to use", it limits the autonomy of car companies; while Linux is a bit like NVIDIA - more "difficult to use", but it gives car companies a lot of flexibility. autonomy and future-proof compatibility.
As car companies increase their investment in software technology and improve their capabilities, Linux's share of the autonomous driving market will gradually increase, while QNX's market share will gradually decline; at the same time, for cars that have failed to develop self-developed OSs based on Linux, For enterprises, there is still QNX as a "fallback route" that they can rely on, which in turn gives them less "worries" when developing their own OS.
It is worth noting that the car companies or autonomous driving companies that have developed their own self-driving OS (which have been successful), whether Tesla, Waymo, Mobileye, Huawei or Apple, have one thing in common: they have self-developed ASIC chips. So, is there a possibility that a car company that plans to develop its own autonomous driving OS will first recruit a chip expert from Nvidia or Mobileye to serve as the person in charge of autonomous driving or CTO, and then, one hand will focus on the chip and the other on the OS?
As the EE architecture of electric vehicles evolves from distributed to centralized, and the autonomous driving level evolves from L2 to L2+, NOA/NOP and even L4, the system's requirements for the autonomous driving OS (operating system) are getting higher and higher. Moreover, because it is related to life safety, the real-time and reliability requirements of autonomous driving OS are much higher than those of in-vehicle entertainment system OS.
The value of OS mainly lies in its ability to better allocate and schedule computing and storage resources. However, under the distributed EE architecture, each ECU basically only handles a single task, and there is no need for resource "allocation and scheduling"; even even Mobileye's EyeQ 3 and EyeQ 4 are only ASICs that run a single task and only need to process camera data. Therefore, a Tier 1 MCU is enough and does not require a high-performance OS.
MCU has few lines of code, simple processes and functions, and uncomplicated logic. Therefore, the risk factor of using MCU is very low.
In the era of domain controllers, many functions have been integrated into a powerful SOC chip. In particular, as the autonomous driving function moves from L2 to L2+ and NOA, the types and number of sensors have increased significantly, requiring The amount of data processed has increased exponentially. Such a complex task cannot be handled by a simple MCU. A powerful real-time operating system is required to handle it.
From L2 to L4, the operating system's requirements for real-time performance, functional safety, and information security are increasing.
In the past, chip manufacturers, L4 autonomous driving solution companies, and Tesla paid more attention to self-driving OS, while most car companies had little involvement in self-driving OS.
The head of the marketing department of an American ADAS chip manufacturer told "Nine Chapters Smart Driving": "We asked many car manufacturers and Tier 1s in 2019, 'What operating systems are you interested in?' The result was that everyone was confused and didn't think much about it. . Bosch is very strong and thinks that the solution of RTOS combined with classic AUTOSAR is very good, but no one knows whether RTOS is feasible when developing multiple cameras.
However, as advanced autonomous driving functions such as L2+ and NOA/NOP gradually become standard equipment in new cars, enterprising car companies must bite the bullet and "catch up" on autonomous driving OS courses regardless of whether they are interested or not; for individual cars, The company even plans to develop its own OS.
However, although the autonomous driving OS is important, it is not as "popular" as chips and lidar, which has also led to many people's misunderstandings about it. For example, once a company wants to develop its own self-driving OS, it will be interpreted by many people as "self-developed underlying operating system." Some media reports will even include such flaws as "Linux-based underlying operating system."
So, what are the currently available autonomous driving OSs? What do car companies mean by "self-developed OS"? What are the differences between self-developed OSs based on different kernels? In the future, who will have the say in autonomous driving OS?
With these doubts in mind, the author participated in the autonomous driving operating system forum organized by Yanzhi Automobile in mid-November last year. After the meeting, I talked with Dong Yuanwen, Chief Representative of BlackBerry QNX IOT Division Greater China, Li Xingyu, Vice President of Horizon Market Development and Strategic Planning, and Volkswagen CEO Zhang Renjie and others had in-depth exchanges.
1: Narrow OS and Broad OS
First, two concepts need to be clarified: narrow OS and broad OS.
The vehicle-mounted intelligent computing platform can be roughly divided from bottom to top into four parts: hardware platform, system software (hardware abstraction layer + OS kernel + middleware), functional software (library components + middleware) and application algorithm software. OS in the narrow sense specifically refers to the OS kernel that can be directly installed on the hardware; while OS in the broad sense includes from bottom to top all programs between hardware and upper-layer applications such as BSP, operating system kernel, middleware and library components.
The picture is taken from the public account "Somersault Cloud and Autonomous Driving"
System software mainly consists of three parts: hardware abstraction layer, operating system kernel and middleware. The hardware abstraction layer includes BSP and Hypervisor.
Hypervisor - an intermediate software layer that runs between the basic physical server and the operating system, allowing multiple operating systems and applications to share hardware, also called VMM (virtual machine monitor), that is, virtual machine monitor. Hypervisors are a "meta" operating system in a virtual environment. They have access to all physical devices on the server including disk and memory.
BSP (board support package) is a board-level support package. It is also a layer between the motherboard hardware and the operating system. Its main purpose is to support the operating system so that it can run better on the hardware motherboard.
The OS kernel, also known as the "underlying OS", provides the most basic functions of the operating system. It is responsible for managing the system's processes, memory, device drivers, files and network systems, and determines the performance and stability of the system.
Middleware, as a bridge between the underlying operating system and these applications, is a platform for managing, allocating and scheduling software and hardware resources, and plays a key role in decoupling software and hardware.
Specifically, middleware usually abstracts resources such as sensors and computing platforms, and adopts modular management of algorithms, subsystems, and functions. Through the unified interface provided, developers can focus on the development of their respective business levels without having to Learn about irrelevant details. This provides developers with great convenience to develop and integrate autonomous driving software quickly, efficiently and flexibly.
Functional software includes core functional modules of autonomous driving, such as programming frameworks for related algorithms (such as TensorFlow, Caffe, PaddlePaddle, etc.).
The pattern of autonomous driving OS kernels is relatively stable, with the main players being QNX (Blackberry), Linux (Open Source Foundation), and VxWorks (Wind River). Because building a new OS requires too much manpower and material resources, currently almost no company will develop a new OS kernel.
Currently, whether it is Waymo, Baidu, Tesla, Mobileye, or a number of self-driving startups and car companies, the so-called self-developed self-driving OS refers to self-developed middleware and applications based on the above-mentioned ready-made kernels. software.
Since developing middleware based on QNX is relatively simple and is subject to greater official restrictions from QNX, while developing middleware based on Linux has high technical barriers, developers can also enjoy greater autonomy. Therefore, usually, many companies provide The "self-developed OS" mentioned here specifically refers to tailoring the Linux kernel accordingly, and developing middleware and application software on this basis.
There are two opinions on whether Huawei's AOS is based on a completely self-developed OS kernel or a middleware based on the Linux kernel: many people think it is "completely self-developed", but there are also many practitioners in the operating system field who think it is "completely self-developed". , AOS is still based on the Linux kernel, but it has deeply tailored the Linux kernel, added some security firmware and unique drivers, and developed middleware on this basis.
It needs to be emphasized that the meaning of the word "middleware" is slightly vague - both system software and functional software include "middleware", so which piece of "middleware" is self-developed by the company? Self-developed operating systems can be divided into customized and ROM types. The customized type modifies the software in the OS kernel and the "middleware in the system software", while the scope of modification of the ROM type is limited to the "middleware in the functional software" .
The picture is taken from the Zhihu account of "Somersault Cloud and Autonomous Driving"
In addition, the autonomous driving OS that Volkswagen is developing in-house is also a broad OS. To be precise, it is a platform that integrates multiple underlying OSs such as Linux, QNX and VxWorks.
two. OS kernel: QNX, Linux, VxWorks
Based on whether the tasks that need to be executed are prioritized, the OS kernel can be divided into two types: time-sharing OS and real-time OS. The former is completely fair to each user/job and does not distinguish the urgency of the task; while the latter means that when a high-priority emergency task needs to be executed, it can be preemptively switched to the task within a strict time limit.
Since it is a matter of safety, autonomous driving requires a real-time OS. Both QNX and VxWorks are real-time OS, but Linux is a distributed OS.
Some time ago, some people questioned Tesla's "using (Linux) time-sharing system as the operating system, which is the most irresponsible for life." However, this question may not be true - Tesla's Linux system is a real-time operating system, and Whether it is developed in accordance with vehicle-level functional safety specifications remains to be confirmed.
(Picture taken from Gaogong Intelligent Automobile)
It is true that Linux was originally a distributed OS, but the work of improving Linux's real-time performance has never stopped. After optimization, the Linux kernel is also divided into a preemptive kernel and a non-preemptive kernel. Among them, the real-time performance of the preemptive kernel is sufficient to meet the needs of autonomous driving.
Sun Luyi, senior director of Shanghai software R&D at Ambarella Semiconductor, mentioned in the article "Thoughts on the Intelligent Driving Main Control Chip Operating System for Cars in the Chinese Market" published in August last year that in 2005, with the release of Linux 2.6, it began to support kernel preemption. Scheduling CONFIG_PREEMPT, this option allows kernel-mode programs to be preempted by higher-priority kernel threads as long as they are not protected by spinlock or are in an interrupt handler.
This greatly improves the task scheduling delay when executing many tasks. The default configuration usually achieves a scheduling delay of 1 to several milliseconds. The real-time patch CONFIG_PREEMPT_RT project is also relatively mature and can be manually patched to improve the real-time performance of the kernel. These optimizations have greatly improved the real-time performance of Linux, making Linux actually a hard real-time operating system.
The University of Bremen in Germany once conducted a test, which showed that when the system is running at full load for a long time, the Linux standard kernel without any patches is no worse than the real-time performance of various common real-time operating systems.
QNX is a semi-closed system, while Linux and VxWorks are both open source systems. The so-called "semi-closed" means that the QNX kernel cannot be changed by customers, but customers can write their own middleware and application software; the so-called open source means that all kernel source codes are open to customers, and customers can tailor them according to their actual needs. Very configurable.
The Linux Foundation launched the open source project Automotive Grade Linux (AGL) in 2012, and currently has more than 100 members. Currently, the application of Linux in the automotive market is mainly in infotainment systems. However, AGL's ultimate goal is to provide functional safety objectives that meet safety-critical systems to serve autonomous driving applications.
According to AGL's vision, in the future, member companies can share 70% of the code, and the other 30% will be differentiated development by different brand manufacturers to protect their respective commercial interests.
QNX passed the ISO26262 ASIL-D safety certification as early as 2011, which is the first real-time operating system to pass this certification. Unlike Linux, which is free, QNX requires commercial charges. However, with its security, stability and real-time performance, QNX still firmly occupies the number one market share of automotive embedded operating systems.
QNX is a microkernel, that is, only the most basic scheduling, memory management, drivers, file systems, etc. in the kernel are implemented by user-mode daemons. Its advantage is that it is super stable. Errors in drivers, etc. will only cause the corresponding process to die and will not cause the entire system to crash.
According to Dong Yuanwen, chief representative of BlackBerry's QNX IOT division in Greater China, currently, General Motors, Volvo, Audi, SAIC, Geely, NIO, Ideal, Xpeng, etc. all use QNX as their autonomous driving OS.
The QNX system provided by BlackBerry to all autonomous driving customers is an undifferentiated standard version, but the BSP driver is a customized version.
BSP includes Bootloader (basic support code to load the boot program of the operating system), HAL (Hardware Abstraction Layer) code, drivers, configuration documents, etc. It is the interface layer between the kernel and hardware, and its purpose is to provide virtualization for the operating system. Hardware platform, making it hardware-independent and portable on multiple platforms.
Compared with QNX, the greater advantage of Linux is that it is open source, can run on various CPU architectures, can be adapted to more application scenarios, and has a richer software library to choose from. Therefore, it has a strong Custom development flexibility.
Moreover, Linux is a macro kernel, that is, in addition to the most basic process, thread management, memory management, file systems, drivers, network protocols, etc. are all in the kernel. Its advantage is high efficiency and the ability to fully utilize the performance of the hardware.
In addition, the free nature of Linux is also very attractive to car companies.
Li Xingyu mentioned in his speech that the usage rate of Linux in the autonomous driving industry has been increasing in recent years. "Technology companies, Tier 1 and car companies will all invest based on Linux."
However, the components of the Linux kernel are much more complex than QNX, so the stability is much worse. Bugs in the development process often cause the entire system to hang up. Therefore, the threshold for development based on Linux is very high.
Zhang Renjie said: "The Linux platform has tens of millions of lines of code, and these codes were not designed by you. How do you ensure its security? You can only make some additions and subtractions on the periphery. Your ability to understand the code determines Can you solve these bugs? Currently, there are tens of thousands of bugs officially recognized by Lunix. Within a certain period of time, these bugs are not a big problem. But under certain conditions, they can cause a lot of trouble."
Linux has been slow to pass the ISO 26262 standard. In comparison, QNX's code and functional architecture are relatively simple, with a total of only 300,000 lines of code, and the possibility of failure is far less than that of Linux.
"High-tech Intelligent Automobile" also mentioned two major challenges in developing based on Linux in a recent article:
1. As an open source operating system, when Linux is used in safety-critical systems, there is a problem of invalid safety certification when the code base is constantly updated. Therefore, car manufacturers need to constantly re-do safety certification, but this may bring additional costs.
2. Subsequent patch management and version alignment are too complex. Engineers at one automaker that chose AGL early said that at least 4 Tier 1s offer different versions of Linux kernels for their specific products, meaning they need to manage 4 different AGL kernels for different patches. To get rid of these troubles, they are preparing for the next generation of products to switch from AGL to QNX.
In this case, why do Waymo, Tesla, Baidu and Mobileye still develop based on Linux? Cost considerations are on the one hand, and more importantly, to reduce dependence on software suppliers. "These codes are all open source, and I am capable enough to do more work on performance optimization."
Tesla has tasted the benefits of using the Linux 4.4 open source operating system for development. Since we fully master the core technology, we don’t have to be controlled by operating system manufacturers. Every time we find a problem, we can quickly correct and upgrade it through OTA to improve the user experience.
In short, from the perspective of car companies, QNX is a bit like Mobileye - although "easy to use", it limits the autonomy of car companies; while Linux is a bit like NVIDIA - more "difficult to use", but empowers This provides car companies with great autonomy and future-oriented compatibility.
As car companies increase their investment in software technology and improve their capabilities, Linux's market share in the autonomous driving market will gradually increase, while QNX's market share will gradually decrease.
The real-time capabilities of Wind River VxWorks have been proven on billions of devices in fields such as avionics, medical electronics and industrial security. Wind River senior product manager Wang Peng said in a previous live broadcast: The real-time performance of any "real-time" Linux cannot be compared with VxWorks.
VxWorks is also open source. The VxWorks real-time operating system consists of more than 400 relatively independent, short and concise target modules, but users can obtain all source codes and tailor and configure the OS kernel at the source code level as needed.
VxWorks' influence is mainly in avionics, and the automotive sector does not account for a high proportion of its revenue. Therefore, previously, Wind River's resource focus was not on the automotive market. But in recent years, Wind River has been increasing its investment in the automotive business.
Wind River not only sells commercial products directly, but also provides professional customized services. In the Audi A8 project, in addition to providing VxWorks, Wind River also provided development board support packages, middleware development and network stack integration for TTTech's automotive business branch.
The autonomous driving OS kernel of the BMW-ME-Intel alliance once adopted VxWorks. Vxworks is also used in modern self-driving production vehicles.
In October 2018, Wind River announced that it would provide support for the latest version of TTTech Auto’s safety software platform, MotionWise, for autonomous driving applications.
In February 2019, VxWorks passed ISO 26262 ASIL-D certification.
Although VxWorks and Linux are both open source systems, the difference between the two is that the latter is free, while the former requires a high licensing fee and development and customization costs are also high, which limits the growth of its market share.
Of course, the relationship between various autonomous driving OSs is not completely antagonistic.
For example, Baidu's Robotaxi system, a common saying is "runs on the Linux kernel", but Xilinx system architect and marketing director Mao Guanghui said: the sensing part runs on Linux, while the non-perceiving part runs on Linux. Run on QNX. "
At least two sets of autonomous driving OS are used on the Audi A8. Among them, Nvidia's Tegra K1 chip uses the Linux kernel, and Intel's FPGA Cyclone uses the VxWorks kernel.
Simply put, when car companies choose OS kernels, they mainly consider factors such as openness, scalability, ease of use, security, reliability and cost, and then make trade-offs based on their own needs and capability systems.
A simple comparison of each OS kernel is as follows:
Part of the data in the table comes from the article "Thoughts on the Intelligent Driving Main Control Chip Operating System for Automobiles in the Chinese Market" written by Sun Luyi, Senior Director of Software R&D at Ambarella Semiconductor in Shanghai.
three. Middleware: provided by the manufacturer or self-developed
Among autonomous driving OS kernel manufacturers, Wind River will also provide customized middleware for car companies. The middleware running on the Audi A8's FPGA Cyclone is provided by Wind River. In addition, the middleware in the BMW-Mobileye-Intel tripartite cooperation project was once provided by Wind River.
But for car companies or autonomous driving companies, using middleware provided by OS manufacturers is only a stopgap measure. In the future, each company will need to make middleware that meets its own unique needs based on a secure core.
For example, Mobileye announced at the end of last year that it was making final preparations for mass production of its next-generation EyeQ®5 chip to switch to a Linux-based operating system. According to Mobileye, "BMW, as one of its mass production customers, has conducted a comprehensive safety assessment and rigorous testing of this system."
"High-Tech Intelligent Automotive" commented in a recent report: "For Mobileye, this is also a key step in its open strategy. The EyeQ5 operating system switches from the previous proprietary and customized operating system to Linux, which will increase the number of subsequent customers. The flexibility of customized development is also open to more than 100 AGL member companies."
Zhang Renjie said: "The difference and advancement of one car company's technology compared to another car company largely depends on your overall framework, which in turn depends on the middleware combination of the OS. Various middleware The pieces ultimately form your basic framework.”
Therefore, for car companies or autonomous driving companies, self-developed middleware is not something you are willing to do, but something you must do. Enterprising car companies are basically developing their own middleware for self-driving OS, but some car companies have spoken out about it, while others have not.
QNX can provide some middleware for customers, but most of the middleware is developed by customers themselves/entrusted to third parties. However, since developing middleware based on QNX is relatively simple and not a very cool thing, few companies will emphasize in PR that they have developed middleware based on QNX.
If you use the Linux kernel, you have to do all the middleware yourself. This is a more valuable but also destined to be more difficult road.
Zhang Renjie said: "When developing based on the QNX kernel, car companies complete the design of middleware through an addition process; while when developing based on Linux, they first do subtraction and then add, which is obviously much more difficult." ( During the subtraction process, if you are not careful, important content may be cut out.)
According to a former key member of Baidu's autonomous driving department, Baidu used the QNX kernel in response to the request of the car manufacturer for its autonomous parking project in cooperation with car companies; but for the more demanding Robotaxi, it developed middleware based on the Linux kernel.
The middleware developed by Baidu is not a brand-new system, but is based on the open source system ROS for cutting and patching. However, there are still high technical barriers here.
The predecessor of ROS was STAIR of the Stanford Artificial Intelligence Laboratory Robot Project, which was later organized and maintained by Willow garage. It is a free open source framework with a clear software architecture. The native ROS is a robot OS and cannot directly meet all the needs of driverless driving. However, Baidu has optimized the disadvantages of native ROS one by one:
1. Communication between native ROS nodes requires a unified ROS Master as a relay. If there is a problem with ROS Master, the communication between any two nodes will be affected.
In response to this problem, Baidu Apollo adopted a service discovery protocol called RTPS to make a "decentralized" transformation of ROS. That is, the current rosnode interface no longer relies on an independent master for management, but all of them are placed in a public domain demain. Each rosnode in the domain has information about other nodes in the domain, so that single nodes can be eliminated. Point failure.
The above two paragraphs of content and pictures are excerpted from the article "Difference Analysis of Apollo Open Source Software ROS Customized Version" written by CSDN blogger "Zhufeng Xiaohuang" in March 2019.
2. The messages delivered by ROS do not support forward compatibility, and interface compatibility issues will have a great impact on the use of historical data. In the field of autonomous driving, massive historical data is a treasure trove. If it cannot be used, it will cause huge waste.
To solve this problem, the solution adopted by Baidu Apollo framework is to use open source protobuf to encapsulate it into ROS msg to replace the native ROS msg. protobuf provides good message forward compatibility.
3. If a ROS node is hijacked, hackers can use this node to exhaust resources and bring down the system. If the messages between ROS nodes are intercepted or forged, the self-driving car may be controlled by hackers.
To address this problem, Baidu uses Linux container, a lightweight virtualization technology, to put each ROS Node into a sandbox and limit the permissions of each ROS node. Communication messages between nodes can be encrypted and decrypted to prevent man-in-the-middle attacks.
In addition to Baidu's Apollo, companies such as Bosch, BMW, nuTonomy, Autoware and Udacity are also developing autonomous driving technology based on ROS.
Apex, a third-party software company, develops autonomous driving software based on ROS 2.0. The company spends a lot of energy optimizing ROS 2.0. In February 2019, Apex received investments from Toyota, Volvo, Jaguar Land Rover, Hella and other car companies and Tier 1 companies, which means that these companies are also considering using ROS to develop autonomous driving technology.
However, developing based on the Linux kernel and ROS middleware will be particularly difficult.
At the end of last year, a star autonomous driving startup that had been developing based on ROS announced that it would switch to QNX. The reason given by the other party’s engineers was: “The QNX system is safer and has greater reliability and stability. ensure".
It seems that after the failure of self-developed OS based on Linux, car companies and autonomous driving companies still have QNX as a "fallback" to rely on, which in turn gives them less "worries" when developing self-developed OS.
However, switching from Linux to QNX in the middle caused a particularly large workload. First of all, the two kernels are very different, which means that the middleware previously developed based on Linux can no longer be used and has to be started from scratch; in addition, the workload of migrating and optimizing functional software and algorithms is also huge, and some even Some algorithms need to be significantly modified.
Therefore, for each autonomous driving company/car company, it is best to make a full assessment of its own needs, capability system, and risk tolerance from the beginning, and then choose the most suitable core for middleware development. A plan that suits you.
Four. Who will have the say in autonomous driving OS?
So, can car companies have the final say on what core they choose to use for development? There are many different versions of the answer to this question.
According to Dong Yuanwen, QNX has established in-depth cooperation with NVIDIA, Qualcomm and other autonomous driving chip manufacturers, but it is weakly bound to NVIDIA and strongly bound to Qualcomm. That is to say, Nvidia can pre-integrate QNX in Xavier and Orin, but it will also give car companies the right to choose independently; while Qualcomm Snapdragon Ride customers can only use QNX as the self-driving OS and "have no independent choice."
Engineers at an autonomous driving startup said that the QNX they use is a version optimized for specific chip integration by BlackBerry and chip manufacturers to ensure optimal system performance on the chip.
Zhang Renjie believes that QNX’s bundling with Nvidia and Qualcomm is easy to understand. "It is not enough to build an OS. To be successful, the core point is to build a strong ecosystem (software + hardware). Without powerful chips, the advantages of advanced OS cannot be exerted. In the early stage of industry competition, OS manufacturers must form a very strong strategic coupling with chip manufacturers."
This kind of bundling is a win-win situation. Not only can OS manufacturers increase the number of software installations by bundling with chip manufacturers, but chip manufacturers can also make chips more popular by bundling with safe and reliable OSs.
However, the CMO of an autonomous driving chip manufacturer believes that this is just an overly optimistic assumption. "For QNX, only by fully embracing the chip company can the chip company push your OS to the OEM. But does the OEM have to choose you?"
The CMO said: "At this stage, QNX is the most cost-effective solution, but if you are not good enough, no matter how hard the chip manufacturers push it, the OEM will not choose you, and then the chip manufacturers will turn to support another OS. Manufacturers. After Windows, the possibility of another OS as powerful as Windows appearing in any industry is very low."
A manager in Bosch's AI department said: "The chip will not be tied to a certain OS unless the chip is the OS manufacturer's own, like Apple's M1 to MacOS."
Ambar Sun Luyi once mentioned in an article: "The choice of OS in the automotive market mainly comes from the experience accumulation of the development team, or from the default support of the SDK of the main control chip factory." The so-called default support means that the chip manufacturer's SDK is based on which The OS developed determines what OS car companies can only use.
But many people don't agree with this statement.
For example, the person in charge of autonomous driving at a certain OEM said: “The OS can be software decoupled from the chip through middleware.”
An engineer from an autonomous driving start-up company said: "The SDK of autonomous driving chip manufacturers usually does not adapt to only one OS. Customers mainly choose the most suitable OS based on their own needs. We use NVIDIA chips, but we use QNX and Linux OSs. They have all been used, and there is no problem of being kidnapped by the chip supplier."
The CMO of the above-mentioned chip manufacturer said: The SDK we launch will be compatible with multiple different OSs. If customers have some special requirements, we will add additional ones.
In an interview with Jiazi Guangnian, Shang Jin, CEO of China Automobile Intelligent Control, a domestic operating system startup, said that one of the major attractions of China Automobile Intelligent Control to OEMs is that it is “not tied to chip suppliers” and “Huawei, Horizon , Black Sesame and NVIDIA are all our partners."
Generally speaking, car companies and chip companies have a higher say in what kind of OS to use than OS manufacturers. One interviewee said: “The right to speak comes from who is closer to the user. Of course the OS has certain decision-making power, but the OS only provides basic functions. The application layer is done by the car company because it is closest to the user. Know what users want.”
Dong Yuanwen mentioned in his speech: "QNX does not have its own Roadmap. The will and money of the OEM determine QNX's Roadmap." This also confirms the saying that "car companies have a greater say".
However, Horizon CEO Yu Kai believes that if a "smart car OS" appears in the future, integrating autonomous driving OS and smart cockpit OS, and establishing a strong developer ecosystem and application ecosystem, this OS will be in front of car companies. Will have a greater say.
Yang Ke, Vice President of Guoqi Intelligent Control, and Li Xingyu, Vice President of Horizon Market Development and Strategic Planning, both mentioned a point of view: Judging from the 30-year history of the IT era, usually, an OS will only be deeply polished based on one hardware architecture. This is true regardless of win-tel in the PC era or Android-ARM in the mobile Internet era. It is difficult to optimize Windows very well on ARM, and it is also difficult to optimize Android very well on x86 architecture.
Li Xingyu said: "Currently, the self-driving OS is still in the stage of contention among hundreds of schools of thought. As competition intensifies, a player will come out to deeply polish the underlying OS and chips to achieve the best experience. By then, the self-driving OS market will have grown from a hundred schools of thought. The rivalry has turned into a standout."
In the era of smartphones, Apple is the representative of self-developed OS and chips and in-depth polishing. In the autonomous driving industry, as of now, there are five such players: Waymo, Tesla, Apple, Mobileye, and Huawei. Among them, Apple is the only player that develops its own OS kernel.
If we only make chips, the application rhythm will be controlled by the progress of OS manufacturers; if we only make OS, performance optimization
Will be subject to the pace of chip manufacturers. But if you do both yourself, you won’t have such worries. This was largely the motivation behind Microsoft's decision to develop its own PC chips at the end of last year.
Although the OSs and chips of Waymo, Tesla, and Apple are powerful, they all serve their own businesses and are not supplied to external parties. However, Mobileye and Huawei are positioned as Tier 1. Mobileye’s chips have lower computing power and are in L4. There is insufficient competitiveness in the market. Therefore, based on the current industrial structure, Huawei is the only supplier in the L4 market that has both autonomous driving chips and OS.
Huawei's MDC810 has the highest computing power among mass-produced autonomous driving computing platforms, and the CPU used on the MDC platform is also self-developed; its autonomous driving OS kernel has obtained the industry's highest level of functional safety certification (ISO 26262) in the safety field. ASIL-D). It is understood that Huawei’s autonomous driving OS will be the first to be commercialized on BAIC Jihu’s Alpha S.
Recently, some people joked that Huawei, which provides operating systems, is "not just Tier 1, but also tier 0" - building a complete ecosystem centered on itself, and eventually becoming the "Microsoft" of the smart car era.
A veteran in the OS industry said that Huawei's next goal should be to build many middleware, application frameworks, and applications on top of the kernel, and ensure that these components can also meet functional safety requirements. In addition, in order to attract more developers, Huawei also needs to create a stable and efficient tool chain.
When talking about Huawei's "chip + OS" pattern, an engineer who switched jobs from Huawei to Bosch said: "No chips and no OS are the biggest differences between Bosch and Huawei; among them, in moving from a distributed architecture to a centralized architecture In the process, the OS is more important than the chip, which is the real brain. If Bosch cannot handle these things, it will only become more and more passive in the future."
Another thought: Apple once claimed that it did not build cars, but produced and sold "autonomous driving systems." If one day Apple increases its To B business and becomes Tier 1, then the market will become even more lively.
Of course, if Huawei and Waymo both build cars, their Tier 1 status may not be preserved. In this case, Huawei and Waymo will become automakers with self-developed chips and OS like Tesla and Apple; while Mobileye As the only supplier of self-developed chips + OS, Mobileye’s next challenge is how to launch high-computing power autonomous driving chips as soon as possible.
Unfinished words:
The number of players in the self-driving OS market is increasing rapidly.
For a long time, when Zebra Networks' AliOS was mentioned, the first reaction in the industry was "smart cockpit operating system". However, at the end of 2020, Zebra Networks obtained the ISO 26262 Automotive Functional Safety Management System certification issued by TÜV Rheinland, Germany. The grade of its certificate is "ASIL-D". At that time, the press release stated that "this indicates that Zebra's overall research and development system meets the highest level of automotive functional safety requirements, providing a key guarantee for creating a safer intelligent driving operating system."
It seems that AliOS is not willing to just be a smart cockpit operating system. It has greater ambitions.
Last month, Red Hat, the world's largest Linux system manufacturer, announced plans to create a new Linux-based enterprise operating system for the automotive industry and provide a series of ISO 26262 certifications from "infotainment to assisted driving systems."
Also recently, Tier 1 ZF and TTTech Auto were also revealed to be developing operating systems and software ecosystems. ZF is already in the game, can Bosch and Continental be far behind? Maybe, maybe Bosch is already developing an autonomous driving OS?
In addition, what is somewhat "unexpected" is that Daimler, which handed over its passenger car autonomous driving and truck autonomous driving technologies to Nvidia and Waymo respectively last year, also announced in mid-April that it would build a software center to independently develop operating systems. .
However, is there really a need for so many “self-developed autonomous driving OS” (specifically self-developed based on Linux) on the market?
From the perspective of building a developer ecosystem and optimizing product performance, the OS market needs a certain degree of concentration to ensure that buyers of OS manufacturers can "get" enough developers and at the same time achieve a larger installed capacity.
Since the automobile market is large enough, several leading OS manufacturers have the opportunity to establish a huge developer ecosystem and achieve a large installed base. However, car companies that develop their own OS face much greater challenges. .
The OS self-developed by car companies is more "self-produced and self-sold" and will not be open to the outside world. Even if car companies are willing to open it, friends and businessmen may not be willing to use it. A senior director of a car company said in an interview with Jiazi Guangnian: "This means that when sales are not enough to share the development costs of the software system, independent software will not be able to achieve iterative updates or even maintenance."
In addition, we found that car companies or autonomous driving companies that develop self-driving OS (and achieve success), whether Tesla, Waymo, Mobileye, Huawei or Apple, have one thing in common: they have self-developed ASIC chips.
Dong Yuanwen once explained the mystery in a speech at the forum: "Because Tesla uses self-developed ASIC chips, many algorithms do not need to be scheduled by Linux. Therefore, it is relatively easy to use Linux for autonomous driving development, but for Other manufacturers don’t.”
So, perhaps, the prerequisite for car companies to succeed in self-developed autonomous driving OS is to have self-developed ASIC chips?
Volkswagen, which is developing its own self-driving OS, has recently been revealed to be planning to develop its own self-driving chip - which will most likely be an ASIC. Mercedes-Benz does not have chips, but they are in-depth cooperation with NVIDIA. So, will they also cooperate with NVIDIA in the subsequent development of self-driving OS?
Several Chinese car companies also plan to develop their own self-driving OS. For them, it is also imperative to develop their own ASIC chips for self-driving?
My colleague Mr. Sun Li has this judgment: Next, the status of chip talents in the autonomous driving circle will become higher and higher, and many autonomous driving leaders will emerge from the chip industry. So, is there a possibility that a car company that plans to develop its own autonomous driving OS will first recruit a chip expert from Nvidia or Mobileye to serve as the person in charge of autonomous driving or CTO, and then, one hand will focus on the chip and the other on the OS?