Urban NOA marks a milestone in the commercialization of autonomous driving, and also means the beginning of the second half of smart cars.
Since the Shanghai Auto Show in 2023, the battle over urban NOA routes has gradually become clearer. "Heavy perception + light maps", intelligent driving solutions based on pure perception and fusion perception routes, and BEV+Transformer models have become industry consensus.
Today, moving towards commercialization and landing competition, how to use efficient computing power support, perfect algorithm model, and a large amount of effective data to form a closed loop is the key to mass production of urban NOA.
In Bosch's view, the complete data closed loop that can support the intelligent driving full-stack solution consists of two parts: "software development data closed loop" and "AI-based data closed loop" .
On the one hand, autonomous driving technology companies have shown their muscles by realizing AI data closed-loop based on AI drive. For example, suppliers of intelligent driving solutions such as Momo Zhixing and Baidu have created their own data closed-loop solutions, making efforts in data collection, data screening, algorithm model iteration and other links, and are committed to achieving low-cost, large-scale, high-quality , Efficient data closed loop.
However, how to use automated means to efficiently complete data collection work including trigger and labeling, and ensure that the data is "correct" data that can be used for model optimization is one of the difficulties in AI data closed-loop.
On the other hand, the verification-driven (test-driven) software development data closed loop is good at traditional suppliers such as ETAS, and its challenges come from the acquisition, hierarchical storage and management of massive data.
In fact, in the verification-driven mode, the realization of software algorithm and model development, software production and release, and complete vehicle state data collection, etc., not only conform to the update of vehicle functions under the trend of "software-defined vehicles", but also help to realize the closed loop of vehicle data .
After all, software defines the car, and car development runs through the entire car life cycle. Compared with traditional car R&D coming to an end after "mass production", how to continue the development and delivery of automotive software after "mass production" of hardware is a major problem that new cars need to solve urgently.
Facing the many pain points of software-defined vehicles and data closed-loop, ETAS, the world's leading provider of embedded software development and automotive information security solutions and services, has launched a software-defined vehicle R&D support platform "Ladder Platform" .
"Starting from the code compilation, the cloud ladder platform can reach the car in real time to collect signals, collect data and store them in the cloud, and upload them to the cloud as data raw materials for analysis and code modification, so as to open up the data of all nodes on the car end. The cloud-ladder platform in the first stage has already achieved direct access to the vehicle end," said Du Xuan, Senior System Architect of ETAS .
Where is the software-defined car?
It is true that software defines a car, that is, under the support of a modular and generalized hardware platform, software is deeply involved in the definition, development and verification process of the car, and continues to create value through continuous optimization of customer experience.
From the perspective of users , standardized traditional cars cannot meet the individual needs of thousands of people. The software-defined car has brought about profound changes. More and more car functions are iteratively implemented through software updates. Smart cars that are always open and new have a wealth of optional applications, OTA software updates, no need to upgrade hardware, and more comfortable driving. Take advantage of the environment.
For OEMs , based on the user's driving experience, establishing a brand's differentiated competitive advantage at the automotive software level is one of the strategies to break through involution and win more market share; in addition, under the trend of software-defined vehicles, shorter Faster vehicle development cycle, rapid iterative update after SOP, reduced ECU software upgrade cycle, etc., also means cost reduction .
However, with the development of intelligent networked vehicles and the evolution of electronic and electrical architecture, the electronic and electrical architecture of the whole vehicle has changed from a distributed architecture to a centralized architecture, which poses more challenges to automotive software development.
First of all, under the domain centralization or cross-domain architecture, the function of the car will be moved up, and how to perform function iteration is the key. At present, the common practice in the industry is to decouple software and hardware, that is, to build a general software framework for the hardware black box, to abstract the interface devices, and to be compatible with different interfaces, or to decouple the internal modules of the software architecture. The software module interface is standardized.
It is reported that ETCH's cloud ladder platform solution, through middleware, opens up the classic distributed ECU system or the modern domain centralized system, establishes a heterogeneous distribution support mechanism, and flexibly completes the observation of vehicle signals and state control, and then realizes vehicle functions. fast iterations.
Secondly, the development of automotive software is more complex, and multi-party participation faces collaboration problems. How to realize the perception and touch of the vehicle entity is a problem that must be solved in the automotive R&D collaboration.
In the past, in the era of manual calibration, engineers mainly relied on laptops to record data and adjust parameters, which was not only time-consuming and laborious, but also difficult to adjust to optimal parameters.
After transitioning to the era of traditional automotive R&D tool chains, many tools are still bound to the window platform, using C# to develop human-computer interaction interfaces, and only support the original state of stand-alone or traditional C/S structure, which is seriously in line with the pace of the digital age. out of touch.
"An ideal software-defined vehicle research and development platform must be centered around the vehicle object; whether it is a vehicle test bench in the laboratory, a verification vehicle on the test site, or a calibration vehicle on the three-high test road, or an end-user manual. Regardless of the type, time, and location of the mass-produced vehicles in the car, the car research and development support platform has the ability to reach the vehicles." Du Xuan said.
In the car R&D collaboration process, it is necessary to observe a large amount of data in the car, and often face the problems of massive data, high real-time performance, and multi-party collaboration. Although it can rely on the cloud for information storage and transfer, it still needs a high-speed and stable network to support the communication channel from the collaborative crowd to the cloud and then to the car. Especially in vehicle verification status or fleet scenarios, the requirements for data processing capability and real-time performance are higher.
In fact, no matter relying on 5G or future mobile communication technology, for the massive data collection and real-time observation of smart cars, the data is sent from the car to the engineer through the cloud, the scene bandwidth is far from enough, and the mobile phone based on the car It is difficult to obtain a continuous high-speed and stable communication channel due to the nature and the interference of the mobile network. This is also one of the pain points of software development throughout the entire life cycle of the car.
In Du Xuan's view, the combination of calibration knowledge objects and large language models is a better solution for calibration domain knowledge processing. Through the accumulation and evolution of knowledge bit by bit, the calibration knowledge base built will help the cloud ladder platform to solve the pain points of automobile R&D collaboration.
Data-driven + vehicle-cloud integration, the cloud ladder platform realizes the closed loop of research and development
The cloud is a new productive force in the automotive industry, and the data-driven model of vehicle-cloud integration will become the key to competition in the automotive industry.
With the further development of automotive chips, it may become possible to have greater computing power and richer operators. Using the cloud-native platform to develop knowledge objects, relying on large models to train automotive software development tools, realizes the self-evolving closed-loop of vehicle-cloud integration. Mode is the pursuit goal of ETAS aerial ladder platform.
It is not difficult to find that ETAS is targeting the pain points of cloud-native knowledge management and data-driven needs in the OEM automobile development process.
After all, as the first company in the world to start deploying vehicle-cloud data collection, transmission, analysis and training on new cars, as well as discovering functional failures, Tesla has tasted the sweetness of the shadow mode.
Through the simulation test of vehicle handling and operation in real scenes, and then compare and analyze the operation results of the two, and clean and label the data set to train the algorithm model and update the deployment to the vehicle, Tesla's shadow mode can complete automatic Cyclic verification and iterative upgrade of the driving system.
But not all car companies can become Tesla. On the one hand, if most car companies build their own private clouds, or lack of overall consideration, it is easy to generate data islands and business faults, and the high cost is directly proportional to the difficulty in gaining. Therefore, OEMs who are unwilling to hand over their "souls" tend to "teach them to fish" in their demand for cloud services.
From the perspective of automobile research and development, ETAS is breaking the deadlock of OEMs.
It is understood that traditional automobile research and development has a mature theoretical model, namely the V model . This process model based on mass production as the end point of research and development provides a good end-oriented process constraint for traditional automobile manufacturing, but it is oriented towards new automobile functions. Changes in requirements can lead to very expensive research and development costs.
The DevOps R&D model popular in the Internet ecology emphasizes the use of agile methods to provide faster product delivery, and the computer-oriented environment can be called "the most ideal digital production environment" today. However, a car is a complex combination of software and hardware. The "hardware" attribute will cause its verification cost to increase exponentially, and the cycle rate of DevOps will be seriously slowed down.
For software-defined automotive new R&D scenarios, both the V model and DevOps have their own advantages. Grasping the "test-driven" core of software-defined vehicles, ETAS proposes a theoretical model of "two systems", a target system oriented to the final product function, and a support system oriented to product support operations , which can meet the needs of software-defined vehicles. space-time requirements.
"Automotive R&D support tools will become a part of automotive products. Two systems will run simultaneously inside automotive products, one is the production system (the part that performs driving functions), and the other is the R&D support system. The two rely on and transform each other." Du Xuan introduced to Gaogong Intelligent Automobile.
The ETAS Cloud Ladder Platform focuses on the support system for automotive R&D, supports extreme automotive R&D, supports knowledge intelligence as the core, and supports cloud-native biological collaboration. It can open up all vertical fields of automotive R&D and realize the closed loop of data-driven + car-cloud integrated R&D. To a certain extent, it also dispels the concerns of car companies about "losing their souls".
However, based on the aforementioned pain points of software development, it is not easy to build this cloud ladder platform even for ETAS, which has accumulated 29 years of software experience.
Du Xuan said that in order to completely connect the cloud, pipe, and terminal, it is necessary to complete the construction of information tools, object tools, knowledge tools, data channels, IOT channels, edge service adapters, and edge knowledge engines .
It is understood that on July 1, 2023, ETAS released the beta version of the cloud ladder platform and started the trial operation of the system. The first stage functions support 30% of the target functions of the vehicle and 70% of the development support.
It can be said that the rapid advancement of the cloud ladder platform is inseparable from the past business accumulation of ETAS.
Relying on complete software-defined automotive solutions, ETAS's business includes six major sectors: software development solutions (DEV), vehicle basic software (VOS), vehicle cloud services (VCS), network security (SEC), data acquisition and processing (DAP) and end-to-end solutions (E2E) .
It is worth mentioning that the "end-to-end solution (E2E)" integrates all smart car software of ETAS from the macro perspective of the vehicle operating system, including AUTOSAR AP/CP, information security components, etc., which can provide customers with A set of overall tool chain and one-click solution that integrates FOTA and remote diagnosis, interconnected drive test and remote calibration, and flexible data acquisition.
Secondly, as a wholly-owned subsidiary of Bosch, ETAS can efficiently create a better and easier-to-use tool chain with the help of its intelligent driving and other automotive business units on the cloud ladder platform. This is also the advantage of Bosch one.
Regarding the future planning of the cloud ladder platform, Du Xuan said that ETAS will continue to improve the cloud ladder platform to achieve a target system that is 100% oriented to the final product function and a support system that is 100% oriented to product support operations ; on the other hand, it will focus on knowledge management , by building an expert knowledge base, providing a database of basic algorithms in the field, building a system for continuous knowledge accumulation and improvement, and introducing large-scale artificial intelligence to deal with potential challenges brought about by continuous iteration of data.