A "The Wandering Earth 2" has triggered people's great discussions and deep thinking about life and technology. Autopilot technology is also subject to soul torture.
Even if the time set in the movie has reached 2030, the highly autonomous driving (L4) level car driven by Tu Hengyu unfortunately suffered a car accident, resulting in the death of his wife and daughter. This has raised concerns and suspicions about whether autonomous driving can be developed to the level of fully autonomous driving (L5) and whether it can truly withstand the test of reality.
(Image source: "The Wandering Earth 2" stills)
Can a perfect score in the autonomous driving test ensure that everything is foolproof?
People are looking forward to the day when they can realize driverless driving, but people's worries and fears about driverless driving have never been eliminated. Wu Gansha, the founder of Yushi Technology, who has changed from the president of Intel China Research Institute to autonomous driving for 6 years, made a metaphor for the current autonomous driving technology and autonomous driving maturity: "If automatic driving is an exam, the current technology can get 98 or 99 points." This is the first parable. The second metaphor is that if you think of the maturity and commercialization of self-driving technology as a marathon, it is now about a third of the way. ”
Wu Gansha explained: "We can understand autonomous driving technology as an exam in an infinite question bank. However, in the infinite question bank, you can get 99 points, does not mean that you are close to 100 points, in this industry we have a 90/10 principle. It seems that 90% of the road has been completed, but the remaining 10% of the road still requires 90% of the time and effort. That said, going from 99 to 100 is tough. ”
"This is a qualitative analysis, if measured from a quantitative point of view, take the data of Waymo, the industry's leading self-driving technology company, there is an important indicator called the MPI (miles per intervention, the average distance between autonomous driving and each time it is taken over by humans." Note that accidents can occur if the person does not take over. Data for 2018 should be every 1. 10,000 miles had a takeover, raised to 1 in 2019. 30,000 miles, 2020 was particularly good, reaching 2. 90,000 miles, which may be the cause of the epidemic, there are fewer cars on the road, but by 2021 the results have dropped to only 0. 80,000 miles to take over. ”
And what about human driving? "Data from U.S. drivers shows that insurance companies pay out insurance every 250,000 miles. 500,000 miles will be called once. It took 1.5 million miles to have one accident that resulted in injury. 94 million miles will have a fatal accident. “
"So if you want to make autonomous driving 20% better than humans, that's 110 million miles to have a fatal accident." To achieve such statistically significant features, it would take to accumulate 11 billion miles of driving data. Cars that require self-driving technology have driven 11 billion miles cumulatively, and if they hit and kill no more than 100 people, they can only indicate that they have reached the level of 110 million miles for a fatal accident. ”
But to get such proof, 11 billion miles is difficult for any self-driving technology company to achieve. Because if the company has 500 vehicles being tested at the same time, it will take 100 years to accumulate 10 billion miles of self-driving data. Even if the MPI is met, it will be difficult to prove that your skills are safer than driving. So my second analogy is that we reached 99, but if it was a marathon we only walked a third of the way.
"Pessimistic people are always right, optimistic people always succeed."
Even if the driverless driver of the infinite question bank can test 99 points, but can never get 100 points, unmanned driving will never be able to be commercialized? Wu Gansha doesn't think so. He pointed out: "There are three paths to go through the commercial use of autonomous driving technology. The first unlimited question bank, although you can't get 100 points no matter what, you can put a teacher next to him and correct mistakes at any time, which is what we call L2 assisted driving. Because the responsibility lies with the driver, the teacher corrects the error as soon as the automatic driving makes a mistake. “
"The second way is a fixed question bank. It is the autonomous driving done in the park. This allows for sufficient testing to ensure that each driverless test can score 100 points, which is also the scenario seen in The Wandering Earth 2 where driverless trucks transport ammunition and help build bases. ”
(Image source: "The Wandering Earth 2" stills)
"The third route is in an open space, or an infinite question bank, but his running speed is relatively low, and the size of the vehicle is relatively small, allowing him to patrol and distribute scenarios without 100 points may also be commercialized."
Therefore, from the perspective of practical and commercial use, the commercial use of the first unlimited question bank may not enter the mature stage until ten or twenty years to reach the L4 level or above. If it is the second or third case, construction machinery in the scene of ports and wharves has now entered the degree of commercialization.
Who will win between cycling intelligence and road coordination?
At present, many autonomous driving companies use the route of bicycle intelligence, and road coordination is another route that quickly matures the commercialization of autonomous driving. The advantage of vehicle-road coordination is to control the overall situation from the "perspective of God" and coordinate all vehicles on the road, rather than considering the problem of "game" with surrounding vehicles from the perspective of a single vehicle. The use of road-side data can greatly simplify the algorithm of autonomous driving of bicycles, and even reduce the computing power requirements and equipment requirements of the vehicle, and truly realize large-area unmanned driving to save economic costs.
However, Wu Gansha believes that "the future will be based on bicycle intelligence, supplemented by road coordination." ”
"Because in road coordination, bicycle intelligence represents the lower limit of ability, and without bicycle intelligence, it is unrealistic to rely only on road coordination." Vehicle-road coordination cannot completely cover every road in China, and cars will not be like our 4G, 5G, everyone's mobile phone can be used. Secondly, it is also very difficult to get all cars to take advantage of road coordination. Therefore, it is impossible to achieve 100% coverage of road coordination in the country, and it is expected that only 1% of vehicles will be able to use road coordination in the next 2-3 years. “
"Although road coordination cannot be completely covered, it can still be achieved in certain areas, such as airports, terminals or specific highways. However, for a technology to be applied on a large scale, it must achieve a closed loop: who pays, who will operate, customers must be perceived, and customers must be willing to pay. If you only do some pilots, there is no way to promote and apply them on a large scale. ”
"In addition, road coordination is more difficult to determine responsibility in the event of an accident than cycling intelligence. If it is a bicycle intelligence, as long as the problem of the car is solved, it basically solves all the problems. However, the number of links in the middle of road coordination will increase significantly, and the determination of the cause of errors will become more complicated. ”
The real battleground for autonomous driving technology is not in ToC, but in ToB
Many people are now looking at the ToC market, but the real value of autonomous driving is in the ToB market. Wu Gansha said: "The self-driving technology that everyone can perceive now is used in passenger cars, taxis, driverless buses to help people get from one point to another. However, a lot of autonomous driving technology is used in logistics. ”
"From the perspective of a city and a company, the main thing is to transport goods. In the city, trunk logistics support the operation of the city. In enterprises, from raw materials and products need logistics support, as important as capital flow and information flow, most enterprises ultimately have to deliver their own manufacturing/production of physical products to customers. ”
"The essence of the digital transformation of enterprises is to solve the problem of human uncertainty and inefficiency. Human uncertainty includes that labor is becoming more expensive and scarce, and fewer people are willing to do repetitive tasks. Even when driving, the technical level of the driver is uneven, and different people have different learning abilities, and the time to master skills is also long. People's emotions can also be affected by various factors. And self-driving technology is a good way to avoid these uncertainties for people. Therefore, after digitalization, we can have more comprehensive wisdom and more efficient planning, and ensure that the entire process of enterprise production achieves a perfect state of Just in Time. ”
The ToB autonomous driving application market is vast, but it faces three major difficulties
Autonomous driving has broad market prospects in the ToB market. According to a Carnegie Mellon University report, cars with intelligent driving functions have a 10% increase in fuel economy, higher levels of automation, higher energy efficiency. More cost-effective: Intelligent driving is of great significance to scenarios with high labor costs, such as long-distance truck transportation, which can save 6~150,000 yuan in labor costs per vehicle per year.
However, lack of standards, misallocation of funds, and lack of scenario-based solutions are the three major dilemmas faced by the market. Wu Gansha pointed out: "The first is the lack of standards. Without a unified standard, good technology and bad technology compete together, blindly competing at low prices, resulting in the phenomenon of bad money driving out good money in the market. The lack of standards is also detrimental to customer selection, hindering the adoption of new technological innovation drivers. ”
"Second, the mismatch of subsidy funds. At present, local governments encourage enterprises to innovate and use financial subsidies to enterprises, which is a good policy, but the government is also worried that some autonomous driving companies will cheat subsidies. It is better to use a post-imposed subsidy and distribute the subsidy to users who adopt the new autonomous driving technology than to directly distribute the subsidy to the provider of the autonomous driving technology. This can encourage more companies to try new technologies and use autonomous driving technology to promote their transformation and upgrading, and at the same time play a positive role in promoting the promotion of new technologies. ”
Finally, there is the lack of scenario-based solutions. Wu Gansha said: "Driverless technology is not a product, but part of the solution, part of digital logistics. This requires the joint efforts of industry associations, leading enterprises, and driverless technology providers to provide solutions suitable for the real application scenarios of enterprises and provide practical and effective landing solutions for their logistics problems. ”
Even if Tu YY only has 2 minutes of autonomy, Tu Hengyu will do everything possible so that his daughter can have a complete life in the digital world. Even if the autonomous driving technology that can score 99 is an endless marathon, it will also allow countless rigorous and determined technology and engineers to continue to explore and improve, although they are not perfect, but they are infinitely approaching perfection.