Continuous optimization of chips, algorithms, operating systems, sensors, radar, etc.
Where is smart driving "on" (new technology and new progress ①)
Our reporter Liu Shiyao
"People's Daily" (19th edition on July 06, 2020)
Looking back at history, it is not difficult to find that every crisis will breed new opportunities, and any difficulty will not stop the pace of technological innovation. What opportunities does the new coronary pneumonia epidemic that is raging around the world bring to emerging technologies such as intelligent driving and smart medical? What new progress has been made in the frontier technologies of quantum computing and satellite Internet? What impact will the steady progress of the new infrastructure have on the development of these new technologies?
This edition will launch a series of "New Technology and New Progress" reports today, introducing the latest developments in new technologies such as intelligent driving, smart medical, quantum computing, and satellite Internet.
--editor
Click on the online car-hailing software, sign up, and wait for the review to pass. Users can call autonomous vehicles for free and have a test ride experience on the open test road... Not long ago, the large-scale manned demonstration application of the Shanghai Intelligent Connected Vehicle started. For the first time, Didixing opened its autonomous driving service to the public.
This news once again stimulated the public's eager anticipation for intelligent driving: When is there really no one, how to ensure complete safety, what difficulties are currently facing, and how far is it from large-scale commercialization?
The reporter interviewed many insiders in this regard.
The core principle has not changed, but the technology has been continuously optimized
Since this year, there have been many news about intelligent driving-
The unmanned logistics vehicle developed by Yushi Technology drove steadily at the SAIC-GM-Wuling Baojun base. This is the first unmanned logistics project in China. 80 unmanned logistics vehicles have been put into operation, no longer have safety personnel, and the overall logistics capacity and efficiency of the base have been comprehensively improved to help customers reduce costs and increase efficiency.
my country's first "smart expressway" that supports the application of autonomous driving technology-Hangzhou-Shaoyong Expressway, its related projects are also in progress.
In fact, intelligent driving is not a new thing. Experts divide it into five development stages—driving support, partial automation, conditional automation, highly automated, and fully automated. The last stage is unmanned driving. There are currently two development paths in the industry: one is a gradual route, which gradually adds some automatic driving functions to traditional cars, and finally transitions to fully autonomous driving; the other is a one-step route, which develops fully autonomous vehicles from the beginning This car is like a "four-wheel computer".
Intelligent driving is a master of modern science and technology, which brings together many artificial intelligence achievements such as vision, speech, language, and deep learning. According to reports, in intelligent driving technology, perception is like human eyes and ears to help vehicles observe the surrounding environment; decision-making is like a brain, real-time analysis of the driving space and other traffic participants' behavioral intentions; control relies on the system to control the vehicle, Complete driving behavior by hitting the direction, stepping on the accelerator, stepping on the brake, etc.
At present, intelligent driving technology has been applied to different vehicle types and different scenarios, carrying people or objects. Wang Feiyue, a researcher at the Institute of Automation of the Chinese Academy of Sciences and director of the State Key Laboratory of Complex System Management and Control, introduced that the application scenarios of intelligent driving are generally divided into open road scenarios, semi-closed and closed scenarios. The former includes intelligent driving taxis, etc., the latter includes high-speed logistics, BRT, mines, parks, logistics, ports and so on.
Today's smart driving technology, compared with a few years ago, the core principle has not undergone major qualitative changes, but chips, algorithms, operating systems, sensors, radar and other technologies have been continuously optimized. Recently, Ali's "Single Frame 3D Point Cloud Semantic Segmentation" algorithm has greatly improved the level of fine recognition of obstacles by vehicles. Even if the warning line temporarily pulled during driving is only 3 cm wide, the vehicle can easily recognize and bypass OK. As another example, driving on rainy days is a problem faced by global smart driving companies. According to Zhang Bo, CEO of Didi Mobility CTO and autonomous driving company, water splashes on rainy days can easily cause noise, and road slipping can affect tire grip. These environmental changes will place higher demands on the algorithms and control systems of automated driving systems. According to previous factual observations, even if Didi Autonomous Driving has experienced heavy rain, it can still run smoothly and take orders normally.
There is still a long way to go to commercialize large-scale open roads
All along, there are constant questions and debates around intelligent driving, and public expectations are high.
The field of intelligent driving does indeed face a series of pain points that restrict development.
The chip is like the digital engine of a smart car. It is responsible for converting data into knowledge, and its efficiency directly determines the quality of the decision.
"According to our statistics, every time autonomous driving goes up one level, the chip computing power will be turned by an order of magnitude. Moreover, there are extremely high requirements and standards in the R&D industry of car-level artificial intelligence chips." Yu Yu, founder of Horizon Kay believes that technology is the foundation of everything. Only by achieving higher computing power at a lower unit cost and making algorithms and chip architectures fit as closely as possible can vehicles be smarter. It is understood that Changan Automobile's new model UNI-T equipped with the Horizon 2 chip has been launched in the middle of this year. This is China's first mass-produced car-level AI chip.
In addition to chips, operating systems, sensors, high-precision maps and other software and hardware work together to achieve maximum benefits.
To be smarter, machines also need to learn large amounts of data quickly. Wu Gansha, CEO of Yushi Technology, said that from a statistical point of view, 11 billion kilometers of road test data are needed to prove that an autonomous driving system is 20% better than human driving safety. The industry is actively making efforts to explore the combination of virtual and real to reduce the cost and risk of testing. It is understood that Ali builds the world's first autonomous driving "hybrid simulation test platform", the system's daily virtual test mileage can exceed 8 million kilometers.
“Didi’s self-developed in-vehicle equipment Jushi covers more than 50% of the orders on the Didi platform. Through this simple installation, Didi can obtain nearly 100 billion kilometers of driving scene data every year, thereby realizing the automatic driving algorithm. Iteration." Zhang Bo said that only by exhausting all the possibilities can the uncertainty of real road conditions be restored, and the system can accurately respond to the emergencies of real road conditions.
At the same time, scientists are also using methods such as reinforcement learning, imitation, biology, etc. to make people's social experience knowledgeable, giving vehicles some ability to "know what they know and why", but they are still in very basic exploration. stage.
Safety is an important value of intelligent driving and the most basic requirement. In subsequent technological iterations, on the premise of ensuring safety, continuously reducing R&D costs and maintaining the balance between cost and efficiency are still serious challenges facing practitioners.
To achieve large-scale commercial applications, technology alone is not enough. Experts believe that to achieve this goal, at least five conditions must be met at the same time-mature technology, perfect social foundation, synchronized laws and regulations, cost reduction, and good social acceptance. Obviously, each condition still has a long way to go.
At present, many closed scenarios of domestic intelligent driving are gradually being implemented, and a certain degree of commercialization has been achieved. But experts say that the road to be achieved is still a long way from commercializing large-scale open roads.
The introduction and implementation of new infrastructure policies have brought many benefits to intelligent driving
The sudden new epidemic of pneumonia has greatly increased the demand for "intelligence" and "unmannedness" in the whole society, bringing new economic growth points to the entire intelligent driving industry.
On April 16, the Ministry of Industry and Information Technology released the "Key Points for the Standardization of Intelligent Connected Vehicles in 2020", pointing out that this year will form an intelligent connected vehicle standard system that supports driving assistance and low-level autonomous driving, and establish an intelligent connected vehicle standard formulation and implementation evaluation. mechanism.
From January to May this year, there have been many large-scale investment and financing events in the field of smart driving in China. China has the world's largest automobile consumer group, and some organizations predict that the market size of China's smart driving industry will exceed 170 billion yuan in 2020.
It can be said that intelligent driving is ushering in good development opportunities.
Compared with foreign countries, my country's intelligent driving has its own characteristics. The huge amount of complicated traffic conditions provide rich data and scenarios for intelligent driving. The demand for resumption of production and economic development and the momentum of economic development have provided a broad market for intelligent transportation. From the government to enterprises and the public, the new thing of intelligent driving has a high degree of recognition. The introduction and implementation of a series of new infrastructure policies will create better software and hardware support for intelligent driving.
"The new infrastructure layout of 5G technology and artificial intelligence technology can better meet the new requirements for coordination of vehicles, roads and people, and better interconnection on the new generation of communication technology infrastructure." Wu Gansha said.
Historically, the birth of new technologies and new applications has often gone through the stages of swarming in, bursting bubbles, and regrouping. In the opinion of the interviewees, intelligent driving is also going through the process of tide-shaking. If you want to become a leader in the intelligent driving industry, in addition to following the trend and seizing opportunities, you must sharpen yourself and concentrate on research.
"Enterprises must have their own original products in the field of general technology and have exclusive and in-depth technical capabilities. In the vertical field, they should combine application scenarios to quickly land, truly solve industry problems, in operating systems and chips with independent intellectual property rights. The core software and hardware fields continue to accumulate. At the same time, we must unswervingly promote the coordinated development of the entire industrial chain to fully prepare for the systematic fast-forward of intelligent driving in China." Wang Feiyue said.
At present, a number of domestic mainstream car companies and Internet companies have merged across borders, combining artificial intelligence and hardware facilities to jointly develop intelligent driving cars. More and more smart driving startup companies are independently researching core technologies to find the most suitable profit model and application scenarios. Facing the future, embracing change, seeking refinement and being pragmatic, I believe that the domestic smart driving industry will usher in better growth and transformation.