Looking back at 2023, this year will leave a strong mark on the autonomous driving industry. High-end intelligent driving products for passenger cars are ushering in a wave of cars, and the rise of large models is subverting the paradigm of autonomous driving AI technology. With the support of large models, autonomous driving technology will continue to be optimized and improved, especially in key areas such as general perception and cognitive decision-making. With the upgrading of technology and the explosion of the intelligent driving market, the mass production of autonomous driving has become crucial.
In 2023, Momo Zhixing and Tsinghua University's Intelligent Industry Research Institute (AIR) held the 2023 Autonomous Driving Boutique Open Class to deeply interpret the development trend of cutting-edge AI technologies and large models in autonomous driving products, as well as the application and mass production logic of intelligent driving products. On December 22, the 2023 Autonomous Driving Boutique Open Class came to the last session, and Zhen Longbao, Chief Delivery Officer of Momo Zhixing, went deep into the field of autonomous driving mass production, and worked with industry and technical experts to solve the problems and challenges encountered in the mass production and development process, and introduced in depth how to transform massive data and AI algorithms into engineering capabilities for mass production products.
In the sharing with the theme of "The Road to Mass Production of Autonomous Driving", Zhen Longbao shared the mass production experience of autonomous driving from the aspects of the difference between demo and mass production, the automotive-grade mass production of the automotive industry, and the mass production process of autonomous driving.
Zhen Longbao first introduced the meaning of "mass production". The so-called "mass production" refers to the large-scale and stable application landing, and autonomous driving requires a strong combination of software and hardware, involving chips, semiconductors, communications, satellites and other different fields, and its large-scale mass production is bound to bring a lot of challenges.
In Zhen Longbao's view, completing a demo is from 0 to 1 stage. In the demo phase, "everything to do is to prove the concept in its entirety". Specific to the field of autonomous driving, it is a variety of demo vehicles that can be seen in the past few years, they have different industrial computers, different sensor configurations, different vehicle forms, and basically do not have computing power, regardless of computing platform, just to make automatic driving capabilities first. This leads to a natural divide between the demo and the subsequent mass production phase. Due to the disconnect between the ability of the demo and the ability of mass production, the feasibility verification and modification of the scheme are improvised. Therefore, from the demo stage to the actual mass production stage, factors such as safety, performance, stability, cost control, higher-frequency scenarios, and wider coverage need to be considered.
Zhen Longbao believes that "highlighting one or two highlights in the demo stage can make people shine and make car companies more interested in going down." But when it comes to mass production, only one or two bright spots are not enough, and the remaining 99 missing items need to be filled. ”
When it comes to the mass production stage, it will also be divided into several levels, such as consumer grade, industrial grade and vehicle specification grade. Among them, the requirements of the vehicle specification level are the highest. Zhen Longbao said that the vehicle specification level refers to the production models produced by automobile manufacturers in accordance with the vehicle standards set by the industry, which requires a higher level of production management and technical level. Reliability, consistency, and robustness are indispensable components of automotive-grade products.
In contrast, for consumer-grade products, such as mobile phones and computers, consumers can accept its occasional black screen, and at most complain a few words, but if it is a vehicle-grade product with a black screen, it will have an impact on user safety. "As long as the vehicle is ignited, all systems are running 100% and there is no laziness or standby time. ”
In the face of such complex considerations and such high product requirements, Zhen Longbao said that the mass production of autonomous driving products will generally be divided into four dimensions.
The first stage is the platform project stage, that is, when there is no external customer project, the definition and development of the internal platform that we do ourselves. This belongs to the R&D project within the platform project, and when there is customer demand, mass production adaptation can be quickly carried out, and it will be divided into the entire creative production, including the collection of requirements, project planning, and product development and verification stages. After this phase, the entire project is planned, as well as product development and validation.
The second stage is the pilot project phase. In the pilot project stage, there will be further various regulatory requirements, including communication specifications, standards that the product needs to follow, functional safety and information security requirements corresponding to the mass production market and region, etc. "At this stage, we will cooperate with the hardware and really do system-level mass production around the project. ”
In addition, there are two phases: application project and validation project. Because some models will be remodeled every year, different facelifts have different contents, corresponding to different changes in content, reflecting different forms of verification for autonomous driving. In the verification stage, it reduces the variables compared to the entire pilot project, and the relevant content can meet the needs of rapid mass production as long as the relevant content undergoes certain targeted verification experiments.
At the scene, Zhen Longbao also took the development of the ASSPICE system as an example to introduce the development process of the autonomous driving system. "On the whole, we must follow the overall ASSPICE system development requirements, including requirements decomposition, system function decomposition, software function decomposition, and then related support systems, ASPICE, the entire life cycle management and after-sales management, etc., which is a complete process that needs to be composed internally to form self-circulation and self-growth. ”
In fact, the development requirements of the autonomous driving system can be divided into multiple dimensions such as hardware development requirements, non-functional requirements, ADAS functional requirements, and safety requirements, which include hardware layout requirements, DV requirements, basic software requirements, electronic and electrical architecture requirements, power-on and power-on network management, time synchronization, FOTA, high-speed NOH function specifications, urban NOH function specifications, parking function specifications, HMI display requirements, functional safety requirements, information security requirements, etc. In response to these questions, Zhen Longbao answered them one by one in the autonomous driving boutique open class.
Zhen Longbao pointed out that whether it is the development of product software and hardware, or the first test and version iteration, now, mass production is no longer the end of product evolution, but a starting point. "The content of each OTA upgrade includes the upgrade of firmware, functions, and vehicle requirements, as well as the further optimization of the overall system capabilities. For autonomous driving systems, the content of the corresponding OTA is particularly large and detailed, and its requirements for related systems will continue to upgrade with the OTA, so that users can continue to create a better experience. ”
The 2023 Autonomous Driving Open Class jointly organized by Momo Zhixing and Tsinghua University's Intelligent Industry Research Institute (AIR) has attracted the attention of many industry insiders and leaders throughout 2023. The course not only introduces the basic principles of autonomous driving products and AI technology, focuses on the development trend and technical principles of autonomous driving driven by large computing power and large models, but also goes deep into the deep water area of autonomous driving mass production, and jointly solves the industry-recognized mass production problems.
The cooperation between Momo Zhixing and Tsinghua Air has undoubtedly injected a strong vitality into the autonomous driving industry. The holding of this series of open classes not only strengthened the communication and cooperation between inside and outside the industry, but also promoted the popularization and development of autonomous driving technology. Through the autonomous driving open class, Momo Zhixing and Tsinghua University's Intelligent Industry Research Institute (AIR) joined hands with senior people in the industry to jointly start a journey of exploration of AI knowledge.
As a leading autonomous driving artificial intelligence technology company in China, Momo Zhixing has been committed to the exploration of cutting-edge autonomous driving technology, with the goal of realizing the vision of "making machines move intelligently and giving life more beautiful", and has always worked hand in hand with many like-minded partners to jointly help the prosperity and development of the autonomous driving industry.