When autonomous driving has gone through the technology competition period and comes to a new node of large-scale implementation, how to deal with the business challenges in the mass production process has become a new proposition.
On December 16, NVIDIA and IDC jointly released the white paper “Reality + Simulation, Ultra-large Computing Power Empowers Autonomous Driving” to discuss in depth the business needs and challenges of automakers in the development of autonomous driving, and how automakers and solution providers should respond. Cooperation to accelerate the development and implementation of autonomous driving.
At present, the development prospect of autonomous driving technology is good, and the overall market is in the stage of development from L2 to L3. Among them, the level of autonomous driving in the passenger car market continues to improve, and the penetration rate of new vehicles in the passenger car market of L2 autonomous driving will reach 10% in the first quarter of 2022 23.2% and will continue to rise for the foreseeable future, compared with 7.5% a year ago (Q1 2021). In addition, autonomous driving is gaining momentum in the commercial trial operation of taxis, and its application in closed industrial scenarios such as mining areas, ports, and airports has become more mature. Commercial vehicle autonomous driving technology is also developing steadily and gradually landing under the promotion of policies.
Autonomous driving technology drives the overall development of the automotive industry towards intelligence. At the same time, it also puts forward new requirements for stable computing power and controllable testing costs.
Autopilot systems are highly complex, and model training and simulation testing place extremely high demands on computing power
To use artificial intelligence to train an automatic driving system, the vehicle first needs to quickly and accurately identify key information in the driving environment such as lanes, pedestrians, and obstacles like a human driver. Through continuous repeated training and verification on the basis of massive data, the vehicle’s cognition level of the road environment gradually approaches the real situation, and the accuracy of judgment is continuously improved in this process. Autonomous driving requires machines to have a high degree of accuracy in judging the environment, so a large amount of scene data needs to be input in the early stage.
In addition, autonomous driving systems need to respond to environmental information like a human driver. This requires the machine to predict the movement trajectories of other traffic participants in the same road environment, so as to plan a reasonable travel route and adjust the vehicle’s travel status in time, which requires a lot of training to correct the predicted trajectory of the system.
In addition, in the early stage of the development of the automatic driving system, it is necessary to use virtual simulation technology to carry out simulation tests, to digitally restore the physical scenes in the real world through mathematical modeling, and to test the automatic driving system in the virtual environment constructed by the software program. This process also requires huge computing power support.
In general, the training of autonomous driving algorithms needs to complete a large number of calculations within a limited time, thus forming extremely high requirements for computing power. High-intensity computing power not only needs to be used for model training, updating, and iteration, but also needs to support the construction and rendering of scenes in simulation tests.
The high demand for large computing power has also been confirmed by industry research data. IDC’s quantitative survey results show that the investment in artificial intelligence computing centers in the autonomous driving industry will grow steadily in the future.
The construction and operation of professional computing power resources help car companies gain opportunities
The speed of development and launch of autonomous driving systems will directly affect whether car companies can gain market opportunities in this field. The white paper points out that computing power resources are a hard factor that directly affects the development speed. The optimization of the underlying architecture of the computing power cluster can shorten the system development cycle on a monthly basis, which directly determines whether the brand can occupy a high ground in emerging fields.
However, the construction of an artificial intelligence computing center has a high technical threshold, and the operation and maintenance process also requires a high level of experience accumulation, so it is necessary to cooperate with a solution provider with mature technology.
The advanced software and hardware technology of the data center solution provider determines the computing power level that the artificial intelligence computing center can provide, and the maturity of its solution determines the time period required for the initial construction and the computing power supply process. stability. For example, the selection of hardware and the scale of the network will have a direct impact on the computing power of the data center, involving professional knowledge in the IT field, and enterprises need to have relevant knowledge reserves and the ability to manage cross-industry partnerships. In addition, suppliers who build and operate artificial intelligence computing centers need to provide an integrated full-stack AI solution to ensure that the development project of the autonomous driving system can go online at the fastest speed and receive continuous and stable computing power support .
After the completion of the construction, the computing power of the artificial intelligence computing center can be put into the development process of the automatic driving solution. At this time, the most common problem in the industry is the maturity and reliability of the software, which means that developers need a team with industry experience to provide technical support to ensure the stable operation of the artificial intelligence computing center to the greatest extent. The investment of money and time is still the main problem that developers need to face at this stage. In addition, some teams urgently need to solve the problems of inefficient development tools and insufficient compatibility of underlying software and hardware.
At present, the most important considerations for automakers when selecting suppliers to build artificial intelligence computing centers, the top three are: the size of the manufacturer and the long-term continuous supply capacity (71%), and whether there are a large number of developers based on the solution of the manufacturer to develop (50%), and reliability (47%).
The construction and operation costs of data centers are huge, and funding issues run through the development process of autonomous driving solutions, which is an important factor affecting developers’ project decisions. Secondly, autonomous driving is still an emerging field for the automotive industry, and occupying the market first can bring huge advantages to brands. However, the time spent on building data centers in the industry varies, and there is a lot of room for potential optimization. In addition, the development of autonomous driving solutions and the operation and management of data centers belong to two different fields. Therefore, data center service providers with development experience can play a huge role in maintaining the stability of computing power.
NVIDIA’s full-stack AI solution empowers the computing power center and accelerates the implementation of autonomous driving products
The report pointed out that in order to enable computing resources to more effectively support the development of autonomous driving systems, the industry needs a full-stack AI solution and build an open platform. Automakers, traditional Tier 1 suppliers, autonomous driving technology companies, and data center solution providers should work together to advance the process of autonomous driving technology reshaping the automotive market.
NVIDIA provides the infrastructure for self-driving cars, including a complete set of hardware, software and workflow reference architectures for data centers required to develop self-driving technology, covering every link from raw data collection to verification, providing neural network development, training and Verification and simulation testing provide the required end-to-end building blocks. In this report, NVIDIA also shared its best practice experience with NIO and Continental, and elaborated on the various supports NVIDIA can provide in enabling autonomous driving development, providing a basis for vehicles that are considering deploying autonomous driving computing centers. Enterprises provide reference and ideas.
Leifeng.com Leifeng.com
This article is transferred from: https://www.leiphone.com/category/transportation/wWKHoByLJgLMTjaG.html
This site is only for collection, and the copyright belongs to the original author.