Photo: Google
There’s this thing called “autonomous driving.” You might have heard of it since... it’s been all over the frigging news lately. Everybody’s doing it one way or another, but Google is the only company that’s constantly sharing its findings with the public.
It’s actually pretty interesting to see things from both perspectives. We’ve heard about a Google car being pulled over for driving too slowly and another one getting rear-ended by a distracted driver as the autonomous vehicle was waiting to make a right turn, but we’ve only had one side of the story so far.
Google currently has 23 Lexus RX450h hybrid SUVs and 30 prototypes (these are those too-cute-to-be-true tiny white vehicles) driving on the public roads of Mountain View, California and, to a smaller degree (only 12 of the total 53), in Austin, Texas. These cars are now averaging between 10,000 and 15,000 autonomous miles (16,000 - 19,000 km) per week, which gives them plenty of opportunities to learn about all kinds of situations.
And learning is exactly why they’re out there. The technology for autonomous vehicles is already here - if you place enough sensors, cameras, radars, lasers and whatnots on a car, and stick a processing unit with sufficient power to make the most of this abundance of information, it is absolutely impossible for it to hit anything. But that’s not the point.
What Google is trying to do now is make its cars smart enough to cruise alongside the more unpredictable human drivers. And let’s not forget pedestrians. The rear-ending episode is the perfect example of how differently a computer and an impatient driver judge the same situation.
We’re not siding with the robots here, but you have to take into account that had a human driver covered the total of two million miles that Google’s autonomous cars drove over six years, statistics shows that, besides being extremely tired, he would have also been involved in a lot of accidents. Some of which might have even been caused by his actions.
By contrast, the Google self-driving cars have only been involved in 17 minor accidents, and they weren’t at fault in any of the cases.
You can read the full Google report for November here, or just run through the more interesting bits below.
On driving too slow:
From the very beginning we designed our prototypes for learning; we wanted to see what it would really take to design, build, and operate a fully self-driving vehicle - something that had never existed in the world before. This informed our early thinking in a couple of ways. First, slower speeds were easier for our development process. A simpler vehicle enabled us to focus on the things we really wanted to study, like the placement of our sensors and the performance of our self-driving software. Secondly, we cared a lot about the approachability of the vehicle; slow speeds are generally safer (the kinetic energy of a vehicle moving at 35mph is twice that of one moving at 25mph) and help the vehicles feel at home on neighborhood streets.
On making a right turn on red:
Our vehicles can identify situations where making a right turn on red is permissible and the position of our sensors gives us good visibility of left-hand traffic. After coming to a complete stop, we nudge forward if we need to get a better view (for example, if there’s a truck or bus blocking our line of sight). Our software and sensors are good at tracking multiple objects and calculating the speed of oncoming vehicles, enabling us to judge when there’s a big enough gap to safely make the turn. And because our sensors are designed to see 360 degrees around the car, we’re on the lookout for pedestrians stepping off curbs or cyclists approaching from behind.
Google currently has 23 Lexus RX450h hybrid SUVs and 30 prototypes (these are those too-cute-to-be-true tiny white vehicles) driving on the public roads of Mountain View, California and, to a smaller degree (only 12 of the total 53), in Austin, Texas. These cars are now averaging between 10,000 and 15,000 autonomous miles (16,000 - 19,000 km) per week, which gives them plenty of opportunities to learn about all kinds of situations.
And learning is exactly why they’re out there. The technology for autonomous vehicles is already here - if you place enough sensors, cameras, radars, lasers and whatnots on a car, and stick a processing unit with sufficient power to make the most of this abundance of information, it is absolutely impossible for it to hit anything. But that’s not the point.
What Google is trying to do now is make its cars smart enough to cruise alongside the more unpredictable human drivers. And let’s not forget pedestrians. The rear-ending episode is the perfect example of how differently a computer and an impatient driver judge the same situation.
We’re not siding with the robots here, but you have to take into account that had a human driver covered the total of two million miles that Google’s autonomous cars drove over six years, statistics shows that, besides being extremely tired, he would have also been involved in a lot of accidents. Some of which might have even been caused by his actions.
By contrast, the Google self-driving cars have only been involved in 17 minor accidents, and they weren’t at fault in any of the cases.
You can read the full Google report for November here, or just run through the more interesting bits below.
On driving too slow:
From the very beginning we designed our prototypes for learning; we wanted to see what it would really take to design, build, and operate a fully self-driving vehicle - something that had never existed in the world before. This informed our early thinking in a couple of ways. First, slower speeds were easier for our development process. A simpler vehicle enabled us to focus on the things we really wanted to study, like the placement of our sensors and the performance of our self-driving software. Secondly, we cared a lot about the approachability of the vehicle; slow speeds are generally safer (the kinetic energy of a vehicle moving at 35mph is twice that of one moving at 25mph) and help the vehicles feel at home on neighborhood streets.
On making a right turn on red:
Our vehicles can identify situations where making a right turn on red is permissible and the position of our sensors gives us good visibility of left-hand traffic. After coming to a complete stop, we nudge forward if we need to get a better view (for example, if there’s a truck or bus blocking our line of sight). Our software and sensors are good at tracking multiple objects and calculating the speed of oncoming vehicles, enabling us to judge when there’s a big enough gap to safely make the turn. And because our sensors are designed to see 360 degrees around the car, we’re on the lookout for pedestrians stepping off curbs or cyclists approaching from behind.
Google News