When it comes to self-driving cars, the future should be now.
in In 2020, you will be a "permanent rear seat driver", The Guardian Expected In 2015. "By 2020, there will be 10 million autonomous vehicles on the road." 2016 year. These statements are accompanied by General Motors, Google's Waymo, Toyota carswith Honda They will build self-driving cars by 2020. Elon Musk Forecast Tesla will achieve in 2018 -Then, if that fails, by 2020.
But this year has come, but autonomous cars have not.
Despite the tremendous efforts of many leading technology and car manufacturers, fully autonomous vehicles remain out of reach beyond special trial programs. You can buy a car that will automatically brake you in the event of a collision, or a car that can help you stay in the driveway, even a Tesla Model S (both my partner and I own) Its autopilot mainly deals with highway driving.
However, as the engineering teams at these companies worked hard to make autonomous cars work, almost every of these predictions had been withdrawn.
what happened? this is You may have nine questions about this prestigious technology and why it still hasn't achieved the future we promised.
1) How does a self-driving car work?
Engineers have been experimenting with prototypes of self-driving cars for Decades. The idea behind it is simple: equip the car with a camera that can track everything around it and own the car React if you want to turn Unite into one. Teach on-board computers road rules and relax them to navigate to their own destinations.
This simple description can be avoided very complicated. Driving is one of the more complex activities that humans do every day. Adhering to the rules of the road does not drive like people because we do things such as eye contact with others Confirm who has the right of way; react to weather conditions; otherwise, make judgment calls that are difficult to code with strict rules.
And, even simple driving operations (for example, tracking objects around a car on the road) are actually more complicated than they sound. Take Waymo, a sister company of Google's self-driving car leader. Waymo's carThis is quite typical in other self-driving cars. They use high-resolution cameras and lidar (light detection and ranging), a method of estimating the distance to an object by reflecting light and rays. sound Drop something.
The car's computer combines all of them to draw a picture that shows the location of other cars, cyclists, pedestrians, and obstacles, and where they drive. For this part, a lot of training data is needed-that is, This car is going to draw The millions of miles of driving data collected by Waymo formed expectations for how other objects might move. It is difficult to get enough training data on the road, so cars also need to be trained based on simulation data, but engineers must ensure that their AI systems can correctly promote simulation data to the real world.
This is a far cry from a complete description of the systems that autonomous vehicles are using while on the road. However, it illustrates an important principle to keep in mind when thinking about where our self-driving cars are: even the "simple" things hide surprising complexity.
2) Why take Getting self-driving cars on the road longer than expected?
Self-driving cars rely on artificial intelligence to work. 2010 was a great decade for AI. We see Translation; speech generation; computer vision and object recognition; and games. Artificial intelligence used to be difficult to identify dogs in pictures; now this is a trivial task.
It is advances in artificial intelligence that have driven optimistic forecasts for autonomous vehicles in the mid-2010s. Researchers expect that we can move on from what they have seen (and still see) in other areas.
But when it comes to self-driving cars, the limits of these benefits become very obvious. Even if you invest a lot of time, money, and energy, no team can figure out how to get AI to solve a real problem: walking on the road with the required degree of reliability.
Many problems require large amounts of training data. The ideal way to train a self-driving car is to show it billions of hours of real driving pictures and use it to teach the computer good driving behavior. When modern machine learning systems have large amounts of data, they do perform well, and when there is little data, they perform very poorly. But collecting data for self-driving cars is expensive. And because certain events rarely occur (for example, witnessing a car accident ahead or encountering debris on the road), the car may exceed its depth, as it rarely happens in training data.
Car manufacturers try to solve this problem in a number of ways. They traveled more miles. They have Train the car in simulation. They sometimes Engineer details This way they can get more training data on these situations.
They are getting closer. Waymo's car does roam the streets of Arizona without a driver (a handful of specially screened people can call them like Uber). If all goes well, they may expand to more cities later this year (more on this below). But this is a difficult problem and progress is slow.
3) What will the world be like with autonomous cars?
Despite setbacks, companies continue to invest, as driverless cars, once they emerge, will make a huge difference to the world and make a lot of money for their creators.
Many consumers will want to upgrade. Imagine you could read or sleep in the morning while driving to work or on a long journey. Taxi and ride-sharing companies also seem to offer self-driving cars instead of paying drivers (actually, Companies like Uber are betting). Self-driving cars should also A huge difference for Americans with disabilitiesMany of them have no driver's license, have trouble going to work, going to the store and seeing a doctor.
Experts disagree on the fundamentals of whether autonomous vehicles will change American car ownership. Some people believe that if people can order a car on the phone and get timely robotic driving anywhere, there is no need to own a car.
Others point out that people generally still own cars even in areas with high ride-sharing coverage, and self-driving cars may be no different. Polls show Most Americans don't want to drive self-driving cars to work -But this situation may change quickly once such cars really exist. Gallup polls It was found that a small percentage (9%) of Americans would get the car immediately, a larger contingent (38%) said they would wait for a while, while the other half firmly said they would not use it.
Over time, our infrastructure may change Self-driving cars are easier to navigate, in fact Some researchers think Unless we make major changes to the streets to make it easier to convey information to these cars, we will not make widespread use of self-driving cars. That would be expensive and requires nationwide coordination, so it seems likely Focus on the broad introduction of self-driving cars, not before it.
4) What are the leading driverless car programs and what are they doing?
Nearly every major automaker has tested waters at least through autonomous vehicle research. But some people take this more seriously than others.
There are two core statistics that can be used to assess how advanced self-driving car programs are. One is how many miles it has traveled. This represents how much training data the company has and how much investment is being made to get the car on the road.
The other is every mile (the time when the driver has to take over because the computer cannot handle the situation). Most companies do not share these statistics, but California requires them to be reported, so California statistics are the best way to understand how various companies operate.
In both respects, Google's sister company Waymo is undoubtedly the leader. Waymo just announced 20 million miles traveled, Most of which are not in California. In 2018, Waymo drive California has 1.2 million miles, with 0.09 miles per 1,000 miles of separation. In second place is GM's Cruise, which traveled about 1 million miles with a release of 0.19 per 1,000 miles. (Cruise debate (The numbers were even more impressive than they looked when they tested their cars on difficult streets in San Francisco.)
Both companies are far ahead of all others in terms of mileage and disengagement in California. Although this is only a limited snapshot of what they do, most experts consider them to be leading programs overall.
5) Will self-driving cars not kill women? How did that happen? What safety issues are involved with self-driving cars?
On March 18, 2018, this was the first time a self-driving car had hit a pedestrian. An Uber car with a safety driver in the rear crashed Elaine Herzberg, a 49-year-old woman riding a bicycle across the road on Tempe Street, Arizona.
The incident reminds people that autonomous vehicle technology still has a long way to go. some people Point out quickly People often kill others while driving, and even if self-driving cars are much safer than humans, there are some fatal accidents that can happen to self-driving cars. For now, this is true. But this misses the point. Human driving produces Fatal accident every 100 million miles. Waymo is the leader in mileage, with 20 million miles recently. They have n’t had a fatal accident yet, but considering how many miles they have driven, too early prove They are safer or safer than human drivers.
Uber is not that far A fatal accident occurred. The company did not release specific figures, but last year's IPO documents showed that the company had driven "millions of miles". It's hard to tell without specific numbers, but would like to know if Uber's driving record is much worse than humans.
In addition, a review of Herzberg's death shows that many preventable errors have occurred. of accident report A report released by the National Transportation Safety Board in December 2019 found that proximity cameras and ultrasonic sensors were "not used at the time of the crash."
In addition, the system has a false alarm problem-a dangerous situation is detected when it is not present-with "programmed in ADS (autonomous driving system) in one second" Suppress planned braking while (1) the system verifies the nature of the detected hazard and calculates alternative routes, or (2) the vehicle operator controls the vehicle. "So even if the car detects a danger, it doesn't brake-it can avoid collisions and even reduce fatal collisions-but it continues for exactly a full second.
The design of the system assumes that a pedestrian can cross the road only at a crosswalk, so when a person does not use the crosswalk, she cannot be identified. To make matters worse, when the system does not know whether an object is a bicycle (like Herzberg), it cannot retain any information about how the object moves. The system sensed her presence six full seconds before the impact, but did nothing (except possibly braking in the last tenth of a second) before colliding with her at a lethal speed.
These are avoidable failures.
Hope they never happen again. In response, Uber parked the car on the road and returned to the driverless car trial phase a year later with a radically changed plan. “We have implemented key safety improvements from two safety reviews, shared our experience with the larger autonomous driving industry, and accepted the NTSB ’s recommendation to implement the safety management system that is being implemented today,” Uber ’s Automated Driving car Nat Beuse safety chief tells Vox In a statement in response to a request for comment. "As we look to the future, we will continue to put security at the center of every decision we make."
Such as I have written before Using quality self-driving cars on the road can save hundreds of thousands of lives. But getting a car to a level that saves lives requires a lot of engineering work.
6) Are self-driving cars good for the environment?
Some advocates believe that driverless cars will be good for the environment. They claim this may Reduce travel by car By removing ownership of a car and transitioning society to a model where most people do not own a car and only require one When they need one.
In addition, others believe that the human driver's driving style is wasteful, including hard braking, acceleration, and idling the engine, which consume fuel, which is what a computer can do. avoid.
But as self-driving cars get closer to reality, most of these claimed benefits seem to be starting to become impossible.
There is not much evidence that computers are more fuel efficient than humans.. Have A small study It is suggested that adaptive cruise control can improve efficiency slightly (5% to 7%), But other than that, nothing else. In addition, researchers studied the impact of fuel-efficient vehicles on mileage and found that in many cases, people Driving more when cars are more fuel efficient —As a result, self-driving cars with higher fuel efficiency may not mean that they will produce lower emissions.
A study attempts to assess the impact of autonomous vehicles on car use behavior Simulate a family with a self-driving car and they spend money to get them a free driver for a weekAnd tell them to treat the free driver service like driving their own car for free.
result? They made more car trips.
Some major shifts to the world of low-speed travel are still possible. Just studying a week's driving habits is not enough to solve this problem. The researchers conducting this study are preparing future studies, and these comparisons may lead to more encouraging results.
7) So if they are not necessarily safe and not necessarily green, why should we do this?
The above sections may cause some pessimism, but there are many reasons to excite self-driving cars. They may make life easier Suitable for elderly and disabled people who cannot drive safely. they Might provide a better, safer, and cheaper option for people currently forced to own a car to go anywhere. Additional research and development will make them safer-once the problem is solved, autonomous vehicles may be safer than human-driven cars.
In a sense, we are in an embarrassing transition period when we want driverless cars but have not become a simple positive thing.
In any case, research and development is underway, mainly because autonomous vehicles may be gold mines Be the first company to get them on the road. They may build their position in the carpool, taxi, and truck markets while competitors are still struggling to catch up, and then they will benefit from further mileage to further improve their cars.
It is not uncommon for a technology to be dangerous and hardly worth it when it is first invented, and to be refined into a valuable part of modern life. The original aircraft was dangerous and commercially useless, but we have made major improvements since then.
8) What role does policy play in the development of autonomous vehicles?
There are no federal laws on self-driving cars. Many of the actions on the policy front are mostly carried out at the state level. The laws surrounding self-driving cars vary from state to state, 29 states passed legislation.
Most of the development of driverless cars has occurred in the states that are most friendly to it, especially California and Arizona, and it is easy to think of some states that banning self-driving cars for a long time after others have become commonplace is not a slam dunk .
When driverless cars were first proposed, I heard many people worry that regulators would unnecessarily delay their implementation. By 2016, it was clear Did not pass. Indeed, in some cases, regulators may be overly tolerant-for example, given Uber's withdrawal of its car and new security procedures, it appears that the car that killed Elaine Herzberg should not be on the road at all.
Policies may also affect the quality of self-driving cars For the environment. For example, under high gasoline taxes, the social cost of carbon emissions can be reflected in the price of use Self-driving cars-money can Spending on climate adaptation and clean energy. However, at present, our transportation policy has little effect on the social cost of driving, and this problem will only become more serious when driverless cars bring more people to the road.
9) So-when can we get a self-driving car?
In a sense, we have been "contacted" in the field of autonomous vehicles for many years. Waymo conducts driverless test drive in Arizona, They have been doing this since 2017. Cruise postponed the launch of its autonomous taxi service in 2019, but they think it could happen in 2020. Earlier this year they launched Car without steering wheel … and no timetable. Tesla's regular software updates make Autopilot's autopilots perform better, but they are still far from fully autonomous.
Of course there are skeptics. Recently, Volkswagen CEO stated that fully autonomous vehicles could "Never happen. "
Given the progress that has been made, this may be an overly harsh forecast. However, for a typical American, it is difficult to accurately estimate how long it will take for self-driving cars to be realised, both because no one knows and because the company is motivated to make publicly optimistic estimates. These companies boast Their progress, but Don't post their unfortunate experiences. Timelines have been delayed, and plan changes are generally not publicly acknowledged until it is clear that they cannot be completed on time.
At the same time, the company is likely to actually put the car on the road before it is ready. They know that, like Uber, killing someone is not only terrible, but it can also bring doom to their business. As a result, people are motivated enough to say optimistic words instead of actually launching.
It's not hard to imagine that, at least under limited circumstances, they will arrive later this year. It is not hard to imagine that the deadline will be postponed for another three to four years.
Self-driving cars are on the road. They are closer than a year ago. The time they actually arrived here was anyone's guess.
Subscribe to the Future Perfect newsletter We'll give you an overview of ideas and solutions to the world's biggest challenges, and how to do things better.
Part of Future Perfect's funding comes from personal donations, grants, and sponsorships. Learn more Here.