Will we ever have all-weather self-driving cars?
Consider this: Waymo, Tesla and others have been diligently testing self-driving cars. Waymo alone has logged over 5 million miles. But most of the testing has occurred in California, Arizona and Florida.
What happens in a less benign climate? Suppose that it’s snowing a blizzard and your home weather station reports freezing temperatures. You instruct your all-weather self-driving sedan to drive you 5 miles to work. Will your car say in a soothing Alexa voice, “Sorry, Boss. I can’t see safely through this stuff. Why don’t you call in sick today?” Or will it say, “I’m required to tell you that it’s not safe to drive today. But – you’re the Boss, let’s go!”
Either way, there’s a dilemma. Perhaps you would be bold enough or foolish enough to drive yourself to work. But should all-weather self-driving cars be smart enough to beg off when they determine that it’s truly unsafe?
This blog addresses the question: is an all-weather self-driving vehicle a real possibility, or an impractical dream?
Concerns About Autonomous Cars in Inclement Weather
A few years ago the mere thought of owning a self-driving car seemed like something straight out of a science-fiction movie. Today, autonomous driving technology has advanced so much that we might see our first fully-automated car on the road within the next year. Despite these rapidly-evolving advances, there are still serious shortcomings in autonomous vehicle technology. And one of those is that to date, we have nothing approaching an all-weather self-driving car.
According to the U.S Department of Transportation, up to 22% of all motor vehicle crashes are weather-related. Autonomous cars have been extensively tested on normal roads in regular weather. However, testing still lacks for precarious weather conditions involving heavy rain, sleet, and snow.
To better understand the hazards of all-weather self-driving, we need to understand why the current technology lacks efficiency under certain conditions.
Why is All-Weather Self-Driving So Difficult?
At the beginning of this year Chris Gerdes, director of the Center for Automotive Research at Stanford University stated in an interview that bad weather poses a real problem for autonomous cars. Inclement weather seemingly makes all automated tasks slightly more difficult. In addition, heavy snow, rain and other fluctuating weather conditions confuse the cameras and LiDAR sensors.
LiDAR refers to the car’s light-sensing radar. It uses lasers to map the surroundings, enabling the car to ‘see’ around it. Tesla has chosen not to use LiDAR. However, other manufacturers consider it complementary to other sensors because of its high resolution and ranging accuracy.
Unfortunately, deposits of snow can hide lane markers and obscure road signs. As a result, LiDAR and other sensors cannot see well enough to guide the car safely. As a result, the risk of an accident greatly increases.
How Can an All-Weather Self-Driving Car Navigate in Really Bad Weather?
In conditions where LiDAR and video sensors cannot see clearly, radar sensors can help. Radar can sense moving objects including other vehicles and pedestrians as well as buildings, signposts, and snow. In order for these radars to function properly they would have to be secured either inside the body of the car or behind the windshield to ensure that ice and rain do not block the antennas.
One of the problems with radar as a weather-penetrating sensor is that it has inherently less resolution than optical sensors. In addition, metallic reflections can confuse radar.
Ford’s Map-Based Approach
Ford, in particular, has taken a novel approach toward all-weather self-driving. They have created high-resolution three-dimensional maps to guide driving. These maps not only provide information about the road but also what is above the road such as landmarks and sign posts.
In poor weather, the autonomous car can use radar signals supplemented by occasional signals from other sensors to position itself within a 3D model map. Thus even if snow covers lane markings, an all-weather self-driving vehicle could safely navigate. Of course, it would still need to sense the presence of movable objects such as pedestrians and other vehicles in order to avoid them.
Finland’s Integrated Sensor Technology
Developers from the VTT Technical Research Centre in Finland have created what may be the first all-weather self-driving car. This vehicle, named Martti, is a retrofitted VW Touareg. It has driven as fast as 30 miles per hour on a completely snow-covered road without losing control.
Whereas most other autonomous cars rely heavily on LiDAR technology to guide them, Martti uses a full suite of complementary sensors. It uses multiple radars, three forward-facing laser rangers, a rear-mounted LiDAR and a combination of antennas, sensors and cameras.
There are no plans to put Martti on the streets of the world anytime soon. Instead, the company is looking to sell the software to other manufacturers, helping them improve their safety ratings and ensuring a safer, smoother ride for all.
All-Weather Self-Driving Will Not Arrive Soon
Despite all the hype surrounding autonomous cars, they still have some way to go before they can be deemed completely safe for the general public to use. However, the technology is advancing rapidly thanks to GM, Waymo, Daimler-Bosch, Ford and many other companies.
A safe transition to completely autonomous cars may come surprisingly soon, provided that we limit them to fair-weather use. Certainly, an all-weather self-driving solution may require much more time. Nevertheless, Ford’s 3D mapping and VTT’s multi-sensor integration offer promising ingredients for an eventual solution.
Regardless of how the technology evolves, companies will have to go to great lengths not only to demonstrate all-weather technology, but also to convince regulators and the general public of its safety.
Would you consider buying an all-weather self-driving car? What would it take to convince you that using it is as safe as driving yourself?
– Photo by Josep Castells on Unsplash
Intuitively, the idea that a self-driving car could be better (safer) than the “average driver” makes perfect sense. It would obviously result in lower accident rates if only because of avoiding human vulnerabilities — distracted-driving and inattention, not to mention degraded physical conditions such as sleepiness or being under the influence of some substance. Most of us would agree with the desirability of that concept as long as it relates to “other drivers”.
The hitch occurs when we start thinking about ourselves, because unfortunately studies have confirmed that if you ask drivers how competent they are relative to other drivers, the answer comes back that almost everyone believes they are “better than average” … which is, of course, statistically impossible. (google the simple search phrase “studies average drivers”)
It leads to the amusing situation where most drivers would like to force other drivers to give up control to a self-driving vehicle so they can feel safer on the road, while at the same time wanting for themselves the ability to have full-manual control of their own vehicle any time they want it.
Or you can look at this the other way around … if people were truly competent in driving in bad weather, the accident rate should not increase just because of weather. People who are really experienced with bad weather know to dramatically increase safety margins, and in the limit recognize that they shouldn’t be driving at all. I am confident that self-driving cars will eventually do exactly that — become MUCH more cautious in bad weather or bad visibility conditions, and refuse to get on the road once conditions deteriorate to a threshold where the danger is unreasonably high.
Another advantage of a self-driving car is that you can raise competency uniformly across the entire fleet of cars simultaneously with a simple software upgrade. By comparison, it would take hours (or days or weeks) of training in conjunction with specialized training facilities to teach one person how to better handle a specific vehicle in various degraded conditions like ice or snow or limited visibility. Very few people have adequate all-weather skills when it comes to driving, but they don’t have the time and/or interest to work at improving their skills, much less the feeling that it is necessary.
I don’t believe the deployment of self-driving cars is a technical issue. It is instead a people issue. Art is quite right that the laws will follow (not “lead”) the deployment of such autos, and will adapt both with the evolving abilities of such vehicles and also with public opinion (whether it makes sense or not). In the end, lawmakers will respond to the prevailing opinions of whoever gets them elected (whether their constituents, or their political party, or the PACs that give them money, or lobbyists who “encourage” politicians to slant their vote in particular directions).
One way or another, self-driving vehicles are inevitable, and in my opinion desirable. But the path to get there is going to be circuitous and much slower than necessary because of people-issues holding the technology back.
Thanks for your always-thoughtful comments, Charles.
It occurs to me that the way that gas-driven vehicles gradually gained acceptance and displaced horses might give clues as to how autonomous vehicles might evolve. Though of course there are many differences that will not replicate.
In the event of a accident who gets sued?
Driver for not paying attention, the maker of the electronic gizmo, the company that assembled the vehicle, etc., etc., etc.
Bob, at first “everyone” will get sued – especially those with deep pockets. If the defendants can show that the self-driving vehicle is safer than an average driver, damage awards may be reasonable. As case law evolves some jurisdictions (cities, states) who value the potential for safer traffic and less congestion will pass laws spelling out responsibilities and liability limits, to protect the self-driving industry. Other countries may pass laws first, providing examples we can learn from.