On May 22, IDC released "China Automotive Cloud Market Tracking Research, 22H2". The report shows that in the second half of 2022, China's automotive cloud solution market will total 1.762 billion yuan.
Among them, Baidu Smart Cloud ranks first in the autonomous driving R&D solution market with a market share of 35.9%, achieving an ultra-high-speed growth of 162% compared with the same period last year, and is in a leading position in the domestic automotive cloud market. This field is expected to become the main position of the solution market and a key variable in the future competition of the automotive cloud market.
In this market research report released by IDC, autonomous driving R&D solutions include data collection, model training, and simulation testing for OEMs and automotive technology companies (perception, decision-making, and control software and hardware service providers) before mass production of autonomous vehicles and hardware. According to IDC's forecast, the compound growth rate of the autonomous driving R&D solution market in the next five years is expected to reach 90.0%.
General process of autonomous driving research and development
In the process of autonomous driving research and development, data runs through all links from collection, labeling, training, and simulation. After being collected by the collection vehicle, the data will be transferred to the cloud. After compliance processing, the data will be used in two scenarios: one is the model training scenario, where the data is labeled manually or by a machine and used as a data sample for the training of the perception model. The second is the simulation scene. After data mining and scene modeling, a scene library that can simulate the real world is built on the simulation platform.
Subsequently, the virtual car equipped with intelligent capabilities such as trained perception models and planning control algorithms will be evaluated on the simulation platform. After passing the simulation test and the real road test, the entire intelligent system will be deployed in mass-produced vehicles to complete the closed-loop research and development from data collection to vehicle intelligent implementation.
In the entire data closed-loop research and development process, each link needs the support of supporting automatic driving tools. These tool sets are called automatic driving tool chains. A good tool chain can accelerate the development efficiency of each R&D link and accelerate the implementation of autonomous driving business.
At the same time, the autonomous driving tool chain needs resources provided by the cloud platform, and the capabilities of the cloud platform will directly affect the research and development efficiency of autonomous driving. According to the IDC report, autonomous driving starts from L2, and every time it evolves to advanced driving, the consumption of cloud infrastructure, platforms, applications, and services will increase by an order of magnitude, which poses greater challenges to the resource utilization of cloud platforms, model training efficiency, and infrastructure.
"Cloud-Intelligence Integration" Autonomous Driving R&D Solution
For the research and development process of autonomous driving, Baidu Smart Cloud provides a complete solution from the business side to the resource side , including an end-to-end data closed-loop, a tool chain that runs through the R&D process, a large model that improves the efficiency of the tool chain, and an AI base that provides powerful computing power for the entire process.
End-to-end data closed loop
Baidu Smart Cloud provides customers with a full-process data closed-loop service from the car end to the cloud, from data collection, data labeling, data management, data compliance, smart driving research and development, and cloud simulation. Baidu has compliant road collection experience covering millions of kilometers of road network length, the largest labeling base in the country and 100+ data mining algorithms, and uses high-quality compliant data to drive the rapid development of the autonomous driving business.
Autonomous driving tool chain throughout the R&D process
Baidu's autonomous driving tool chain provides developers and testers with an end-to-end tool platform covering L2 to L4 autonomous driving research and development. It fully integrates various long-tail requirements and scenarios encountered in Baidu's 10-year autonomous driving research and development process, and more than 50 million kilometers of real road test data, providing customers with a full-stack autonomous driving research and development tool. At present, it supports the efficient management of exabyte-level data, the simulation mileage of millions of kilometers per day and the data compliance of the whole process.
In addition, Baidu's self-driving tool chain can also output modular capabilities according to customer needs, helping car companies quickly integrate the self-driving tool chain with their own business, and efficiently realize the data-driven closed-loop research and development.
A large model for continuous improvement of the tool chain
The application of large models can further improve the efficiency of the autonomous driving tool chain. At present, the capabilities of Baidu Wenxin's large model have been applied to data mining scenarios. For example, by inputting keywords such as "pig that hit a tree" and "plastic bags flying in the sky", or by searching for pictures by pictures, it helps developers obtain the long-tail data they need from massive data for subsequent training processes and improve the effect of model training.
The large model will also be applied to scenarios such as data labeling and cloud simulation. For example, you can quickly label pictures by entering "label all traffic lights", and enter "build me a scene where the main car is going straight on the straight road" to quickly build a corresponding scene library to further improve the efficiency of autonomous driving research and development.
AI base that provides powerful computing power support for the whole process
The AI base provides powerful AI infrastructure support for autonomous driving tool chains and large models. Among them, the Baidu Baige·AI heterogeneous computing platform provides acceleration capabilities for data labeling, model training, simulation and other links, allowing car companies to quickly implement the autonomous driving business.
In the labeling scenario, Baidu Baige's cGPU sharing solution more than doubles the cost of automatic labeling. In the model training scenario, among the first batch of 17 perception models jointly optimized by Baidu and NVIDIA, the training throughput increased by an average of 138%, up to 400%, and the model training time was shortened by up to 80%. In the simulation scenario, Baidu Baige can support the simulation platform to achieve a simulation mileage of millions of kilometers per day.
Integration of cloud and intelligence, helping the intelligent upgrade of the automotive industry
IDC predicts that by 2027, the market size of autonomous driving solutions will reach 20.87 billion, and the proportion of the overall solution market will increase from 28.1% to nearly 70%. It will become a key variable in the future competition in the automotive cloud market.
Baidu Smart Cloud is based on the strategy of "integration of cloud and intelligence, deepening the industry", based on the AI base and Wenxin model, as well as the industry's leading full set of automatic driving tool chains, and has successively helped traditional car companies, new forces, commercial vehicles and solution providers and other industry users to implement the automatic driving business. Baidu Smart Cloud will continue to develop autonomous driving R&D solutions including "closed data loop - automatic driving tool chain - large model - AI large base" to help the automotive industry achieve intelligent upgrades.
Click "Read the original text" to cooperate and consult immediately