Autonomous driving needs to evolve. Can artificial intelligence help it?

After the initial exploration, the entire industry has gradually become more pragmatic and rational. For autonomous driving to "evolve", it requires a comprehensive improvement in software and hardware performance, and it is also inseparable from the construction and improvement of industrial systems such as infrastructure, policies and regulations.

Data source: Beijing Autonomous Driving Vehicle Road Testing 2018 Annual Work Report

Autonomous driving, the "traffic star" in the technology industry, has many faces: its promising prospects give people endless expectations, while its highly complex challenges make people hesitant.

On the one hand, major companies brought their latest products to the stage to promote autonomous driving; on the other hand, experts were examining the application prospects of autonomous driving, but the prospects were still unclear. At the 2019 Big Data Expo held in Guiyang recently, autonomous driving presented such different faces.

From concept to product, can autonomous driving overcome the long technological barrier? In which fields will it be first applied? How far is it from you and me?

Form innovation

Cars are getting smarter

One of the important reasons why autonomous driving seemed out of reach was that the machines were too "stupid". Today, with the help of the latest artificial intelligence , radar, geographic information and other technologies, machines have become smarter: they can not only "see", but also have no blind spots; they have "intelligence" and know how to change lanes and turn, accelerate and brake. Whether the light at the intersection is red or green, whether to turn left or right, how far to keep from the car in front... In theory, all these can be judged by machines.

From manned to unmanned, it is not only the driving mode that has changed. In the view of Wu Gansha, an autonomous driving expert and co-founder of UISEE Technology, the innovation of travel mode will also create a new business model - passenger economy, that is, working, consuming and entertaining on the road.

The innovation of autonomous driving cannot be separated from the support of external environment such as infrastructure, policies and regulations. In recent years, China has promoted the integration of automobiles, transportation and communications, and built a relatively complete industrial system; in terms of technical environment, the linkage layout of 5G , artificial intelligence, etc. has also formed a relatively complete industrial chain.

my country was one of the first countries to explore the legislation and testing system for driverless cars. By the end of last year, Beijing, Shanghai, Shenzhen and other cities had successively issued local autonomous driving car testing management specifications or drafts for comments. Zhang Yonggang, director of the Intelligent Department of the Automotive Research Institute of Beijing Automotive Co., Ltd., suggested that a unified standard testing system should be formed at the national level to improve the evaluation of autonomous driving from multiple aspects such as scenarios, algorithms, and testing methods.

Technology Upgrade

The long march to tackle this problem is a difficult one

The prospect seems promising, but there is a long technical road ahead for autonomous driving, and everyone is "crossing the river by feeling the stones". Autonomous driving involves a wide range of technologies, covering many aspects of software and hardware. If any link is "lame", it will not work.

For example, unmanned vehicles need sophisticated sensing technology to see the surrounding environment clearly. Wang Feiyue, a researcher at the Institute of Automation of the Chinese Academy of Sciences, said that in recent years, the perception ability of cars has improved rapidly, but when encountering bad weather such as rain and fog, they become "blind", and there is still no perfect solution.

In order for unmanned vehicles to make choices based on the environment, they also need an efficient "brain". However, autonomous driving is a "good student" who abides by the rules. In reality, road conditions usually change rapidly, and it is difficult for machines to make reasonable decisions in some emergencies.

In addition, like almost all artificial intelligence applications, autonomous driving needs to be "fed" with large amounts of data in order to "evolve", which is very time-consuming.

Wang Feiyue introduced that autonomous driving is inseparable from communication technology. Only with faster and more stable transmission can a set of intelligent vehicle networking systems be formed. Experts at the Big Data Expo said that 5G communication will be the key to promoting the implementation of autonomous driving.

Wu Gansa believes that autonomous driving will become more and more difficult as time goes by, and it is a difficult "Long March": "The 99% we have completed so far may only be 1% of the journey, and the last 1% may still require 99% of effort." However, industry experts believe that autonomous driving is generally an engineering problem, just like building a house, not building a mirage. The difficulty lies in how to build it and build it well.

Similar to many AI applications, my country is relatively weak in algorithms and core hardware compared to leading countries, but has the "home advantage" of data and application scenarios. "The current autonomous driving technology is still immature, and one of the factors affecting it is that the computing efficiency is not high enough," said Yu Kai, founder and CEO of Horizon Robotics.

Ni Kai, an autonomous driving expert and founder of Hodo Technology, said that my country's high traffic density, rich traffic scenarios and large market will help local companies accumulate rich road test data and form a set of solutions that suit my country's conditions.

Safety is the original intention of developing self-driving cars. An organization calculated that to prove that unmanned driving is safer than human driving in most cases, it would take 11 billion miles, which means running 100 cars day and night for 500 years. The industry uses the number of miles traveled that require human intervention as one of the criteria for measuring the maturity of autonomous driving technology. At present, the data of the best performing unmanned vehicle is 11,000 miles. "It is important to establish global standards for autonomous driving safety," said experts.

Application landing

"Step by step" has become the consensus of the industry

According to the intelligence level of the vehicle, the industry divides autonomous driving into six levels from L0 to L5. Among them, L1, L2 and other assisted driving technologies are already widely used, but L3 and above have not yet been implemented on a large scale.

If you ask industry insiders about the application scenarios of autonomous driving, you may get multiple or even conflicting answers.

Ni Kai introduced that, generally speaking, there are two R&D paths in the industry. One is the "one-step" model, which is to develop L4-level autonomous driving cars; the other is the "step-by-step" model, which is to gradually add some autonomous driving functions to traditional cars and then transition to fully autonomous driving.

Wang Feiyue believes that autonomous driving will be popularized in specific closed scenarios such as factories, airports, and docks, followed by municipal buses and taxis, and finally open city roads.

After years of experimentation, the “step-by-step” approach has gradually gained recognition in the industry.

In fact, in the first few years of the rise of autonomous driving, the industry was overly optimistic about the technology, and with the stimulation of several international capital mergers and acquisitions, some bubbles appeared. Since 2018, some product deliveries have not been as expected, and several driverless car accidents have occurred, and autonomous driving has obviously cooled down.

"Autonomous driving is relatively easy to demonstrate, but difficult to implement." Ding Fei, executive director of IDG Capital, said that the cooling of the enthusiasm for autonomous driving is the tuition fee to be paid in the process of growth.

However, many industry experts believe that the current cooling of autonomous driving may be a good thing for the industry, as it will help truly innovative companies emerge and thus gain a place in the global autonomous driving industry chain.

Ni Kai introduced that artificial intelligence technology is promoting the development of autonomous driving in various fields, and the entire industry is also working hard to improve the performance of software and hardware. The industry predicts that in the next 3 to 5 years, some L3 autonomous driving vehicles will be mass-produced, and L4 autonomous driving will begin to be applied in limited scenarios such as valet parking and highways. The next 10 years may be the key period for implementation.