Original title: “Nine major questions take you to understand: It’s all 2020, why haven’t driverless cars come yet? !! 》, The title picture comes from: Oriental IC
2020 will be an important time point for driverless cars.
The British Guardian predicted in 2015: By 2020, you will become a “permanent back seat driver”. A 2016 BusinessInsider headline titled “1000 Autonomous Vehicles Will Be On The Road By 2020”. General Motors, Google’s Waymo, Toyota and Honda have announced that they will produce driverless cars in 2020. Musk has also said that Tesla will reach this goal in 2018, and if it fails in 2018, it will eventually be achieved in 2020.
Now that 2020 is here, driverless cars have not been able to appear as promised.
In fact, despite the extraordinary efforts of many top brands in technology and automotive manufacturing, driverless cars are still out of reach. Now, you can buy a car that will brake you in time when a collision occurs, or you can buy a car that helps you cruise automatically and run smoothly on a highway, but you ca n’t buy a real car without Man driving a car.
Why is this? Perhaps we can get answers from the following nine questions.
1. Why are self-driving cars taking longer on the road than expected?
Self-driving cars rely on artificial intelligence to work. In the first 10 years of the 21st century, the progress made by AI can be called “great”. We have seen great progress in speech generation, computer vision and object recognition, and games. For example, in the past, it was difficult for AI to identify in a picture. Some kind of animal, but now it’s worthless to AI.
But in the field of driverless cars, the limitations of AI are still very obvious. Even if you invest a lot of time, money, and energy, no team really finds a solution that allows AI to play a vital role in practical problems-navigating with high reliability and precision. Training self-driving cars requires a lot of training data. The ideal way is to show them billions of hours of real driving pictures and teach the computer good driving behavior.
But the main problem is insufficient data.And there is a “expensive price” to obtain some data, such as: witnessing a car accident ahead, or encountering debris on the road … Only in these cases, AI can obtain relevant data and come to a solution.
Due to insufficient data, there may be many gaps in the training of driverless computers. Automakers are trying to solve this problem in a number of ways: they drive the car for more miles, train the car for more simulations, and even design specific situations to get more training data.
These actions seem to have had some effect. For example, Waymo’s car can indeed roam the streets of Arizona without a driver, but in fact, all this progress is not so smooth. (more on this below) .
2. What will the world look like when driverless cars are on the market?
Despite setbacks, many companies continue to invest in self-driving cars. Once self-driving cars are successfully produced, the world will change dramatically and companies will make a lot of money.
And consumers also want some change. For example, you can read or rest on the way to work; self-driving cars will also provide a lot of convenience for the disabled. After all, many of them cannot get a driver’s license, and it is too difficult to go to a business superintendent, hospital, or company alone.
The driverless car of the General Motors Cruise Department is waiting to pass at a crossroad in San Francisco
However, some researchers believe that unless we make major street changes to make it easier to communicate information to driverless cars, driverless cars will not be widely available.in. Large-scale street renovations are expensive, requiring not only money and manpower, but also national coordination across the country.
3. What are the “leaders” in the field of driverless cars doing?
Evaluating the advanced nature of self-driving car research, there are two core statistics that can be referenced.
One is how many miles it travels, which means how much training data a company has and how much money it has invested in it.
Another is “self-driving off” data. Self-driving off refers to a situation where a car has to be temporarily taken over by a human driver because the computer cannot handle special situations during the driving process.
In both respects, Waymo is undoubtedly the “leader.” Waymo recently announced that the company’s self-driving cars traveled a total of 20 million miles: In 2018, Waymo traveled 1.2 million miles in California, with 0.09 miles “self-driving off” every 1,000 miles. In second place is GM’s Cruise: it has driven about 1 million miles and has a “self-driving off” figure of 0.19 per 1,000 miles.
4. Is driverless really safer than human driving?
On March 18, 2018, the first serious accident occurred in a driverless car: In Tempe, Arizona, Uber, a driverless car with a safety driver killed a 49-year-old woman. She was riding a bicycle across the street.
This incident makes people think, is driverless driving really safer than human driving?
Waymo has traveled 20 million miles without a fatal accident, but given that its cars have far less mileage than humans, it is too early to prove that they are safer than human drivers.
On March 28, 2017, Uber driverless car drove across Fifth Street in San Francisco
A fatal accident occurred in Uber’s self-driving car with a small mileage. Although the company did not release relevant data, according to last year’s IPO (IPO) document, it traveled “millions of miles”, this The numbers are far from human mileage. In addition, investigations into the first fatal case of driverless cars have shown that Uber driverless has made many preventable mistakes.
The National Transportation Safety Board (NTSB) The accident report released in December 2019 showed that when the accident occurred, the “short-range camera” And “ultrasonic sensors” are not used.
The NTSB report also shows that there is a problem with false alarms. For example, a hazard is detected in a safe situation and reprogrammed based on the detected hazard-the driverless system suppresses braking in one second and calculates an alternative path. Therefore, even if the reprogrammed car suddenly reacts and the next movement will be dangerous, it will not brake immediately, but will continue to drive according to the alternative path. The collision that should have been avoided, but the computer “accurately” allowed the car to continue driving for a full second, causing a disaster.
The system design of Uber’s self-driving cars is based on the assumption that pedestrians will only cross the road on the sidewalk, so when a pedestrian does not use the sidewalk, it cannot identify itself. To make matters worse, when the system cannot recognize whether the object in front of it is a person or a bicycle, it cannot retain any information about the movement of the object.
So, in Uber’s lethal case, the system felt the presence of a pedestrian six seconds before the impact, but did nothing before colliding with the pedestrian at a lethal speed.
5. Is driverless car environmentally friendly?
Proponents believe that driverlessCars are good for the environment.
They claim that with self-driving cars, many people may no longer want to buy a car and use shared cars instead, thereby reducing unnecessary trips. In addition, some people believe that human drivers’ “force acceleration, emergency braking” and other actions will cause the engine to idle and accelerate fuel consumption, which can be avoided by driverless cars.
But as driverless cars get closer, these so-called benefits are starting to become less realistic. There is not much evidence that computer driving is more fuel efficient than human driving.
A small study shows that adaptive cruise control can improve fuel efficiency by 5% to 7%, but it doesn’t make much sense beyond that. In addition, the researchers studied the impact of cars with higher fuel-saving capabilities on mileage and found that in most cases, driverless cars with higher fuel efficiency did not mean lower emissions.
A study aimed at “estimating the impact of driverless cars on car use behaviors” suggests that when you own a driverless car, people may use it more. Researchers provided a family with the cost of using a week’s self-driving car service. They tried to use this method to simulate a family life with a driverless car. The result was disappointing: more trips.
Of course, just studying a week’s driving habits is not enough to prove anything. Currently, researchers are preparing more data for future research, and the results after expanding the sample and time axis are expected.
6. If self-driving cars are not necessarily safer and not more environmentally friendly, why research and development?
Reading here, you may have some pessimism, but there is more exciting news in the driverless field: The development of driverless cars may make life easier for the elderly and disabled And provide people with better, safer, and more economical options. Future research and development will make driverless cars safer—maybe they will be safer than human drivers.
In a sense, we are at an embarrassing transition moment. We long for the arrival of driverless cars, but at the same time know that the road is difficult.
Each new technology is accompanied by dangers in the early days of its invention, but when it matures, it may bring great convenience to human beings. Such examples are actually not uncommon. For example, when airplanes were first invented, they are increasingly becoming an indispensable means of transportation in daily life.