PDA

View Full Version : How does Google's self driving car park itself?


Emerald Hawk
08-07-2014, 03:01 PM
This has been bothering me ever since I first heard about Google's new self driving car, the one that has no steering wheel or other manual controls. I assumed that some journalist would ask this eventually but I can't find the answer anywhere.

How does the car park itself in a new, unknown underground parking garage? This seems like an absurdly hard technical challenge and I would love to know how they do it.

Every GPS I have ever used stops working completely within a few feet of entering a parking garage. All garages have special rules for Handicapped vs non-handicapped parking. Many also have numbered spots for residents, rules about parking at certain times of the day, reserved spots for takeout customers, expectant mothers, etc. Does the car have cameras that can read and interpret all this information, find a legal spot, and park without human intervention? Or do you sign a waiver when you buy it that says you only park in per-marked spots with clear skies?

I also wonder about parking in open fields, as directed by people waving you to a certain spot, like I just did for our county fair last week. Does it come with gesture interpretation too?

Full disclosure: I used to work for Google, but I have no idea who I could direct this question to.

OldGuy
08-07-2014, 03:29 PM
Isn't the answer it doesn't and if this must be done a human takes over.

Snarky_Kong
08-07-2014, 03:34 PM
Lasers.

I don't believe the Google car drives via gps. Route planning, sure, but the driving is image recognition and sensors onboard.

Dr. Strangelove
08-07-2014, 03:40 PM
They're a ways away from the kind of navigation you describe, but on the specific point of GPS in a parking garage, you can use dead reckoning--basically, using accelerometers and other sensors to figure out where you are based on where you started from and how you moved since them.

Dead reckoning is more prone to error than GPS, but there is at least one company (U-Blox (http://www.motorauthority.com/news/1090212_u-blox-3d-automotive-dead-reckoning-offers-improved-urban-navigation-video)) with a system that's at least accurate enough to be useful.

Incidentally, Google's system uses more than just a camera--it uses a LIDAR (laser depth sensor) system to build a 3D map of the surrounding environment. However, IMHO, LIDAR systems will never be practical for production vehicles. Instead, they'll use multiple cameras and figure out depth via parallax (just like our eyes).

Emerald Hawk
08-07-2014, 03:52 PM
Isn't the answer it doesn't and if this must be done a human takes over.

How do you take over if there's no steering wheel?

http://www.nytimes.com/2014/05/28/technology/googles-next-phase-in-driverless-cars-no-brakes-or-steering-wheel.html

on the specific point of GPS in a parking garage, you can use dead reckoning

I understand that the car can drive around in a parking garage. It can sense the pillars and other parked cars, and stay in its lane, etc. That's no problem. But where does it park? Does it look for white lines? Car-sized spaces between other cars? How does it know if it's a handicapped spot? Maybe I've answered my own question and it has cameras and computer vision algorithms to attempt all of the above.

Shalmanese
08-07-2014, 04:23 PM
Currently, Google's car can't go anywhere that's not been previously comprehensively mapped. So, presumably right now, Google engineers would take a pre-existing map of the garage and manually mark out handicapped spaces, zones etc.

Dr. Strangelove
08-07-2014, 04:35 PM
How do you take over if there's no steering wheel?

http://www.nytimes.com/2014/05/28/technology/googles-next-phase-in-driverless-cars-no-brakes-or-steering-wheel.html

The article also says they're limited to 25 mph, so clearly they are already missing big segments of what normal cars do today.

If the cars behave like automated taxis, there may be no need for the cars to ever park other than at the main taxi lot, where they can install extra equipment to help out the vehicles.

Maybe I've answered my own question and it has cameras and computer vision algorithms to attempt all of the above.

It definitely has that already. The hard part is dealing with ambiguous input. Maybe the handicapped spot has an oil stain over the marker. A human can still see a blue spot and the general outline and know that it's a handicapped spot. A computer has a harder time making these guesses. I think it will be quite a while longer before computers can handle all of this.

Fortunately, parking lots aren't a big cause of auto injuries. It's boring highway driving that kills the most people, and that's what computers are good at. Hell, we're already pretty much there. Just look at this idiot (https://www.youtube.com/watch?v=zY_zqEmKV1k) (the video is self-referential proof that we need full automation).

TriPolar
08-07-2014, 04:46 PM
It works from a pre-programmed map now, and Google hopes it will be able to figure it out itself in the future. Either way, when it brings your car back it's changed the seat position, played with the radio, and stolen all the change in the console.

Chronos
08-07-2014, 05:12 PM
While, yes, image interpretation is in general a difficult problem... we're a lot further along in it than most people in this thread assume. The handicapped symbol is a simple, consistent image, printed on a nearly flat surface, and computers nowadays are easily capable of correctly recognizing it. You want a really hard problem? Try facial recognition, which computers nowadays are also doing.

dstarfire
08-07-2014, 05:45 PM
I'm betting that big thing on the roof is a lidar array(like radar but with lasers instead of a radio). Other autonomous car designs typically use lidar to continuously build a 3-d map of their surroundings. That would let you navigate through the garage. Add in a couple of cameras to read signs and spot dividing lines and you're all set.

As for the types of parking spaces, the easiest way to handle that would be by having a digital map of the parking lot that includes data about the acceptable uses about each space. It'd be a simple matter have the car make sure it has adequate parking data about the destination before beginning the trip. And when you actually have all the data you need, determining appropriate permissions for any given set of circumstances is old hat for programmers (and legal systems were doing it long before computers existed).

Emerald Hawk
10-22-2014, 08:38 AM
It can't park itself, according to this slate article:

http://www.slate.com/articles/technology/technology/2014/10/google_self_driving_car_it_may_never_actually_happen.single.html

It canít currently find a space in a supermarket lot or multilevel garage.

I knew it. I don't know why it took so long for reporters to discover this.

Chronos
10-22-2014, 09:02 AM
From that article:
It can't consistently handle coned-off road construction sites, and its video cameras can sometimes be blinded by the sun when trying to detect the color of a traffic signal. Because it can't tell the difference between a big rock and a crumbled-up piece of newspaper, it will try to drive around both if it encounters either sitting in the middle of the road.
And this is any different from a human driver how?

zut
10-22-2014, 11:24 AM
It can't park itself, according to this slate article:

http://www.slate.com/articles/technology/technology/2014/10/google_self_driving_car_it_may_never_actually_happen.single.html


I knew it. I don't know why it took so long for reporters to discover this.
This isn't an answer to your original question, but Audi has demonstrated a prototype self-parking system (http://blog.caranddriver.com/audi-demonstrates-driverless-self-parking-a7-in-vegas-parking-garage/), which relies on transmitters located in the parking structure to communicate with the vehicle.

So that's a way to implement self-parking, if not Google's way. It does require infrastructure upgrades, and so doesn't seem as attractive as a magic car that can drive everywhere, in all conditions, on today's roads, all by itself. Still, IMHO, it's at least a feasible way to solve the problem. And you're more likely to see slow and steady implementation of autonomous systems (sold as safety and convenience enhancements) than to see a quick bold jump to a steering wheel-less car.

Shodan
10-22-2014, 11:33 AM
And this is any different from a human driver how?It's more expensive, and less fun to manufacture.

Regards,
Shodan

Senegoid
10-22-2014, 11:51 AM
This isn't an answer to your original question, but Audi has demonstrated a prototype self-parking system (http://blog.caranddriver.com/audi-demonstrates-driverless-self-parking-a7-in-vegas-parking-garage/), which relies on transmitters located in the parking structure to communicate with the vehicle.

So that's a way to implement self-parking, if not Google's way. It does require infrastructure upgrades, and so doesn't seem as attractive as a magic car that can drive everywhere, in all conditions, on today's roads, all by itself. Still, IMHO, it's at least a feasible way to solve the problem. And you're more likely to see slow and steady implementation of autonomous systems (sold as safety and convenience enhancements) than to see a quick bold jump to a steering wheel-less car.

I suspect that this will be the way it's done in the longer term.

An analogy: One of the problems that has kept electric-powered cars from taking over the world has been the lack of recharging stations on every corner, in contrast to gasoline filling stations, which are very common.

There has been little demand for charging stations because there are few electric cars, and little demand for electric cars because of the lack of charging stations. But this has been changing very gradually: In more recent years, there has been gradual increase in the prevalence of charging stations (most often seen in public parking lots, like at public libraries for example). Thus, electric cars and the necessary public infrastructure will proliferate very gradually in tandem.

Same with self-driving cars and the attendant infrastructure. Little by little, some technology will emerge, and some standards and protocols will be developed. There will be RFID chips, or something similar, embedded in parking lots (perhaps in each individual parking space), and also in any number of other places where a robo-car might go.

Not sure how they will deal with parking in dirt fields at the County Fair, though.

Chronos
10-22-2014, 01:27 PM
Make the transmitters cheap enough, and you can just build them into the cones that the county fair workers are putting out in the dirt field anyway. And if there are enough other autonomous cars in the parking lot, they can all communicate with each other to give the vehicle a very good idea of where the parking spots are.

One other place where these vehicles would be very useful, even with the current tech, would be in school buses. There's a huge shortage of school bus drivers, which self-driving buses could fill in. You'd still need an adult on the bus, of course, to monitor the students, but that adult would no longer need a commercial driver's license, widening the pool of applicants considerably, and would also be able to devote their full time to watching the students, a huge improvement over a human driver who must split their attention between the students and the road. And school buses travel a set, consistent route every day. Unfortunately, it probably won't happen, because people are stupid, and prefer the illusion of safety over the real thing.