Google almost specialise at crunching massive data into useful products, and building reliable supersystems from millions of unreliable subsystems.
Car companies might have some software knowledge these days but they really are a long way from the capabilities Google has. They solve car problems, but it might turn out that this is an information system problem.
But with self-driving cars, it's different. A car landing on a highway, when it's not supposed to be there, and then getting into an accident, is not good news. So that's why I think accurate map data will be very important for this.
So, let's say you are driving a vehicle that isn't allowed on the highway but incorrectly takes one. How do you think maps will help with the "and then getting into an accident" part?
Of what relevance is Google's map data (which they buy from vendors)? Google's map data doesn't tell the car if a pedestrian or another car or a fire hydrant is where it is about to go. The car has to figure that out all by itself and in real time.
Map data is neither necessary nor sufficient to drive a car.
At the scale where a driver-less car operates, the map data is only relevant as to whether it's managed to get its owner where he wants to go. If that question is still relevant, then it has succeeded.
Street View is YEARS out of date in most major international cities and doesn't exist in the majority of smaller cities. It would be dangerous borderline criminal for a self-driving car to be relying on it for anything important.
Google's cars start with an extremely detailed map of the road, including where the lanes are and where everything should be. And, every time a Google car drives the road, that map can be updated.
This obviously isn't ready to deploy nation-wide right now, but there's a pretty obvious growth path here. If you were to buy a Google car for your commute, you might have to manually drive it for several days in order for it to learn everything it needs to know. Eventually, the other self-driving cars on the road will also be helping build a map for you, and you for them.
No they don't.
Google licenses the majority of its data from third parties. Or have you never seen a sat-nav system before ?
Basically Google has recorded all the driver's reactions to whatever is happening on the road, and this allows them to teach the bot to drive. Using this data the bot can virtually drive the roads again and again, until it doesn't make mistakes. There's no way car companies can compete with them.
The best Audi could show at CES was self-parking, and even that needed infrastructure put in place in the parking lot for it to work, so it's not even in the same class as Google's self-driving tech.
Also as others have said, Google might have the most accurate data in the future for such technology, which means cars using Google's self-driving tech will be the safest out there.
I read a claim[2] a while back that the coming S-Class could drive itself, but they chose to not sell it because of concerns about lawsuits, etc.
[1]: http://en.wikipedia.org/wiki/Autonomous_car#History
[2]: http://www.businessinsider.com/mercedes-benz-2014-s-class-se...
They partnered with Stanford AI Lab , won the Darpa Grand Challenge, climbed Pikes Peak, etc.
The key issue being overlooked in these discussions of 'Autonomous vehicles at scale' is that all of these projects revolve around Lidar - and I'm unsure how well that scales with multiple laser sources throwing off 'point cloud' data. :shrug:
[1] Volkswagen Electronics Research Laboratory (ERL) http://www.vwerl.com/
http://www.eastbayexpress.com/ebx/the-future-is-in-the-past/...