Driverless Cars? One Skeptical California Resident Asks Google to Dish
Robocars take human error out of the equation, but can they tell a tree branch from a pedestrian?
LA drivers have a bad enough reputation, but as a resident of California, what I worry about more are cars without any driver at all.
Google has been testing driverless cars on public roads in California since 2009, according to the company, and it just rolled out its latest version — one with a removable steering wheel, accelerator pedal and brake pedal — on the streets near its headquarters in Mountain View.
Some people have no trouble with the concept of driverless cars.
In an April 2014 blog for Slate.com, technology writer Will Oremus noted that a self-driving car errs on the side of caution and that its decisions “are rigorously data-driven.” And unlike cellphone-texting human drivers, he said, a Google car never gets distracted or impatient. He contended the cars will likely prove safer than those driven by humans. “In fact, given that Google’s cars have now logged nearly 700,000 miles without causing a single accident, it seems that they already are,” he wrote.
Actually, Google cars have been involved in minor accidents, although perhaps not by the time Oremus wrote his blog.
In May 2015, Consumer Watchdog, a nonprofit public interest group, learned there had been accidents involving Google’s robocars and filed a Public Records Act request with the California Department of Motor Vehicles (DMV) seeking communications between Google and the department.
DMV releases accident reports
Initially, the DMV refused to release accident details, claiming they were confidential. But on June 18 of this year, it announced it had reversed its policy and would release accident reports about crashes involving self-driving cars tested on public roads.
You can check out the six driverless car accident reports in California filed since last September, when the DMV began requiring them. The accidents appear to have been caused by other drivers, often by rear-ending the Google robocar when it was stopped at a traffic light.
Asked about the accidents, a Google spokesperson explained the company is still in a pilot phase of the self-driving car project.
“Every vehicle has trained safety drivers as we learn to understand the challenges and surprises of city and highway driving,” she told SafeBee via email. “Safety is our highest priority and in nearly a million miles of autonomous driving, we can report that the self-driving car has not been the cause of a single accident.”
Discussing Consumer Watchdog’s campaign to highlight potential safety problems with the self-driving cars, the spokesperson acknowledged some concerns were valid, but said a lot were “wrong or exaggerated.”
“Consumer Watchdog’s website makes it sound like the car will go charging through the street uncontrollably if it misses an important piece of information — but that’s not how the car has been designed to work," she told SafeBee. "If it’s not sure what to do, it’ll do the safe thing: slow down and stop. Unfortunately, humans don't do this enough." She noted the robocars’ speed is limited to a modest 25 miles per hour.
How safe are driverless cars?
For its part, Consumer Watchdog contends Google’s long-term plan for the cars is inherently unsafe.
“Unbelievably, Google is planning to offer its robot cars without a steering wheel, brake pedal or accelerator, so there would be no way for a person to take control in an emergency,” said Consumer Watchdog’s John M. Simpson in a news release.
The Google spokesperson confirmed Google is working on a car that needs no human intervention. “Our long term goal is fully self-driving vehicles that can take you from A to B at the push of a button, because we believe this is essential to providing the greatest benefits to people in their daily lives,” she said. “This was one of the most important decisions we made — if our goal over the long run is to get to a fully autonomous vehicle, at some point you have to try to really do it.”
What about bad weather? Consumer Watchdog charges, for example, that rain interferes with Google car sensors and that they don't work at all in snow or heavy rain.
“Extremely heavy rain, snow, or thick fog can be a challenge, just like it is for human drivers," said the Google spokesperson. "We’re working on improving this, and the good news is that our cars recognize when they have limited visibility and will make the safe decision not to drive."
The cars' ability to process and respond to human hand signals is another concern raised by the public interest group.
"There’s a difference between the social signals humans rely on to read another driver’s intention (like someone waving you on at a stop sign) and signals that are used to indicate rules of the road (e.g. construction worker or policeman)," said the Google spokesperson. “In the second case, we can detect and understand cyclists using hand signals to show that they’re turning left or right, and we can detect someone like a construction worker or policeman holding up a temporary stop sign."
"Definitely tricky scenarios"
“These are definitely tricky scenarios, but we can navigate them; we continue to polish our capabilities so we appear smooth and confident to other drivers. Worst case is that we slow down or stop until we’re sure what to do....We’ve been very honest that we’re still developing the technology, so there are still rare situations that might stump the vehicle," said the spokesperson.
Because Google's robot cars rely on detailed mapping of routes, some critics worry the cars might not recognize potential hazards such as potholes. Google says its driverless cars are aware of road changes.
“Because we map all roads in great detail before we drive on them, we can compare what our sensors see in real-time to what we’ve mapped, and can detect when the environment has changed," the Google spokesperson said. "Additionally, one of the benefits of our technology is that we can share the learnings of one car with the entire fleet. So if one of our cars detects a change in road conditions, it can alert the rest of the fleet."
Consumer Watchdog has also charged that the self-driving cars' sensors can't reliably distinguish between a tree branch blowing in the wind and a pedestrian, another allegation Google disputes.
“In our 1.7 million miles of autonomous and manual driving, our cars have gotten really good at detecting different categories of objects, such as vehicles, pedestrians, bicyclists and static objects that we don’t expect to move into our path, such as trees and mailboxes.”
“I'd add that we have a special team whose entire job is to develop our software's ability to handle rare and weird situations on the road, since we need to be prepared for the .001 percent edge cases that could pop up on the roads, even if we’ve never seen them before in our real world driving.”
“Baked-in defensive driving”
The Google spokesperson noted Google has added sensors to eliminate blind spots and that the car can “see” out to 200 meters in all directions. “As a result, this vehicle can avoid many of the sticky situations that human drivers often get into,” she said.
The cars also employ “baked-in defensive driving behavior: We stay out of other drivers’ blind spots, we nudge away from lane-splitting motorcycles or wobbling vehicles, and we pause 1.5 seconds before proceeding into an intersection after a red light turns green,” she added. “We’ve also built backup systems for all the major systems, including the steering wheel and brakes” for added safety.
At least three other states — Nevada, Michigan and Florida — have passed laws allowing self-driving cars onto their roads. As these vehicles motor closer to reality, questions around their safety will likely take a front seat.
Related: How to Make Your Next Car Safer