There are more than 1,400 self-driving vehicles being tested in 36 states and the District. Some are delivering groceries. Others can be called like a taxi. The industry has been saying a techno-utopia is just around the corner, a transportation revolution that will make society better. But how close is that, really?
For now, most autonomous vehicles being tested in the U.S. require a backup driver. You’ll be the driver of this one — let’s just call it AV — on your way to the airport. The car is supposed to make the trip easy and comes with a custom-built, 3-D map to create the best route.
Your car can drive itself only within a fixed area and only under certain conditions. For example, Waymo permits its cars to travel in autonomous mode within a 100-square-mile area around Chandler, Ariz., a Phoenix suburb, but not necessarily at all times. If there is a dust storm, the cars are designed to stop. Then, the driver can take control.
Those odd protuberances on your car? They are sensors used to help the AV identify what’s around it and orient itself, down to the centimeter, on the road. The number, placement and kinds of sensors vary among companies and their cars.
Even with all this sensor coverage from cameras, lidar and radar, your eyes can see roughly four times farther in daylight than this car.
There are plenty of blind spots, and they set up many problems. In this case, the bus blocks the sensor’s line of sight.
Your AV is supposed to know where it is by comparing its sensor data with road features, like crosswalks and traffic lights, from its 3-D map. But if something blocks the sensors, the car doesn’t have enough data to sync with the map. Then it doesn’t know where it is — and the backup driver needs to take over.
How does your car know what’s around it? People, emergency vehicles, temporary detours? First, the car must identify all of that. Then, it determines what might happen in the next few seconds. Will that woman on the curb jaywalk? Might that sports car in front of you suddenly stop?
Does your AV need to swerve to avoid the car in front? After the car knows its location and what’s around, it must calculate how to react to the myriad scenarios of daily traffic.
Got all that? There’s no going back from here. The button at the bottom is now your responsibility. You can use it to stop or restart the car. Hit it quickly if you need to take control or if the car perceives a problem and disengages from autonomous mode.
Great job! You stopped the car in time. Luckily, the truck stopped in time. That was close!
The car didn’t see the light was red. You, as the backup driver, needed to take control of the car and stop it to avoid the crash with a vehicle that had the right-of-way at an intersection. Our reporting found several instances in which a car didn’t recognize the color of a traffic light. Let’s restart the car.
Good reflexes! Yikes!
Your AV failed to see the truck broken down in its lane until it was almost too late. There have been multiple reports of autonomous vehicles not detecting other objects in real time. The problem here is self-evident: If the car can’t see and avoid a broken-down truck in time, it can’t be counted on to safely navigate in routine traffic conditions.
The car came to a screeching halt.
Yes, but the car had no idea what it was or what it would do. Despite how much engineers train their self-driving cars, there’s always the possibility they’ll encounter something unexpected. For example, Volvo tested its vehicles’ Large Animal Detection System in areas with moose, but, during a 2017 test in Australia, a car detected a kangaroo and was confounded by its unusual hopping habits.
Whoa! The car stopped, even though the light was green.
You’re lucky. This car is equipped to hear sirens from the firetruck. Not all companies have audio detection systems on board. And the ones that do must adapt those systems to a variety of sirens that first responders use in cities nationwide.
The car stopped too soon for the red light.
Your AV didn’t differentiate fully between pedestrians and bicyclists, who move at different speeds and can take different paths. In the worst-case scenario, that could lead to deadly consequences.
The car pulled over. Let’s hope this doesn’t cause you to miss your flight.
ALERT: Severe storms warning issued for this area
Uh, oh. The car also got this information about the incoming storm. It’s still in the fixed area in which it can drive, but now the weather conditions have changed, and it can’t continue in autonomous mode. The bad news? You’re driving the rest of the way.
You grab the wheel and hit the gas to the airport. Think you’ll call a cab next time?
The autonomous vehicle industry has made lots of cheery projections: Robocars will increase efficiency and independence and greatly reduce traffic deaths, which occurred at the rate of about 100 a day for the past three years nationwide. But to deliver on those promises, the cars must work. Our reporting shows the technology remains riddled with problems.
There are flaws in how well cars can “see” and “hear,” and how smoothly they can filter conflicting information from different sensors and systems. But the biggest obstacle is that the vehicles struggle to predict how other drivers and pedestrians will behave among the fluid dynamics of daily traffic.
Like the kangaroo problem in Australia, new challenges pop up all the time. Street sweepers impose their own right-of-way, stopping to make giant circles and brush all four corners of an intersection no matter the traffic light. FedEx drivers jump on their carts and ride them like skateboards through blocked traffic.
What confounds the vehicles also complicates efforts to develop uniform standards among companies and establish regulations for safety, insurance rates, liability and traffic laws. For now, the Transportation Department, under Secretary Elaine Chao, is taking a hands-off approach on test programs.
That lack of federal regulation could “lead to disaster,” said Adam Scow, a senior advocate for Consumer Watchdog, a progressive nonprofit group in California.
“Safety and lives are at stake,” he said. “This is a no-brainer.”
While more than 40 states have issued some form of legislation or executive action related to self-driving cars since 2017, the regulations vary widely.
Some industry leaders now caution it may be decades before fully self-driving vehicles are on the road.
Gill Pratt, the head of the Toyota Research Institute, said in a speech earlier this year that it’s time to focus on explaining how hard it is to make a self-driving car work.
“How do we train a machine,” he asked, “about the social ballet required to navigate through an ever-changing environment as well as, or better than, a human driver?”
To construct this interactive graphic story and simulate the challenges the industry faces developing autonomous cars, The Washington Post reviewed voluntary reports from AutoX, GM Cruise, Nvidia, Uber, Waymo, Nuro and other self-driving vehicle companies that describe their autonomous technology and safety standards; autonomous car research from the Congressional Research Service and the University of Michigan's Transportation Research Institute; and video lectures on self-driving technology.
The Post also reviewed more than a thousand pages of autonomous vehicle disengagement reports, filed to the California Department of Motor Vehicles from 2015 to 2018, as well as crash reports involving self-driving cars from the National Transportation Safety Board and guidelines on automated vehicles from the Transportation Department.
The Post interviewed individuals from autonomous vehicles companies, academics and other experts in the industry. The Post also participated in self-driving car tests in Chandler, Ariz., and San Jose, Calif. Transportation Secretary Elaine Chao’s estimate on the number of autonomous vehicles in the United States came from her speech at an Uber Elevate Summit event earlier this year.
While autonomous vehicle makes and models, and the level of technological advancement, vary by company, the 3-D model of the car depicted in this graphic is based on a generic, four-door compact minivan and doesn’t represent a specific company’s vehicle. The representation comparing sensor capabilities with human vision is based on research from the University of Michigan’s Transportation Research Institute. The 3-D world doesn’t represent an exact location within the world. However, a one-way street to the airport — one that is unhindered by traffic jams, broken-down kangaroo trucks and other inconveniences — would be nice.
About this story
XX Data from (TK TK).