Lots of hills, and will have plenty of ice and snow in a few months. It seems to me like a logical place to test the resilience of a self-driving system against some predictable corner cases.
> The company has been creating extremely detailed maps that include not just roads and lane markings, but also buildings, potholes, parked cars, fire hydrants, traffic lights, trees, and anything else on Pittsburgh's streets.
If one city requires that much work, they're not scalable. Even if we assume the LIDAR on every Uber car collects the new data, the car will trip sooner or later if it can't detect potholes.
Concerning the bridges, can't they do like the Kinect: Match the image with depth-sensing lasers, so they can detect the boundaries of the bridge, stray away from pedestrians, filter out fire hydrants and trees but slow down when encountering other objects?
For example, "don't take 5th street, it's full of potholes" or "we can't drop off the passenger on this block, there's a fire hydrant – will have to drop them off on the next block, so plan accordingly".
On the other hand, if the cars won't drive properly without these maps, in which case I too would be extremely hesitant to trust them.
Here's a typical example, rather on the roomy side: https://www.google.com/maps/@40.4435279,-79.8936687,3a,75y,9...
One of the tricks of driving in an older city with on-street parking is weaving in and out of unused spots to make way for oncoming vehicles that have no means to move over themselves—or won't, for fear of clipping a side-view mirror, or because they can't bring themselves to care about how little space they've left for others. Sometimes you have to stop and wait for one car to make its way through a tight stretch before going through yourself, rather like pulling up to a one-lane bridge that has been generated at runtime.
The trouble is that the lanes have to be imagined with respect to the current conditions, bearing in mind what other drivers are likely to expect of you; knowledge of the road is not enough.
This is a social problem as well as a technical one, and I will be interested to see the solutions.
Delphi has a more robust system (in theory), but the accuracy isn't quite as good. To the best of my knowledge Ford is also taking a different approach which doesn't depend on super-detailed maps to operate. The technology to compare the surroundings against a map is far easier than figuring out the surroundings and then navigating within that (not to trivialize this - both are extraordinarily difficult and nobody has made a fully self-driving car yet).
But hey, like they always say in YC - do things that don't scale. Uber is clearly pushing ahead faster than anyone else, and they have a hell of a team behind this project.
Regarding bridges - to oversimplify the simple answer is they're far noisier for the sensors (all of them, including cameras) and it just wreaks havoc on the sensor fusion system.
Still point taken that its just millions.
This is Uber we're talking about.
Street view already did it... Not 100% coverage to be sure, but pretty darned close.
I would actually love for someone who works at Google to chime in here. It'd be great to find out their system is more robust than I think it is.
An interesting aside - it seems like most of Google's team left this year. Chris Urmson left, Anthony Levandowski left, and many of their engineers are gone as well (many went to Otto). It'll be interesting to see what happens now that they've lost all of the original leadership (Sebastian Thrun left long ago).
Google's approach is simpler but is highly dependent on the maps. The other approach is easier to scale, but is obviously harder and likely takes more work to get the same results. Something I think a lot of people gloss over is that these maps take up huge amounts of data - not something you can stream over a cell network and probably not even high speed WiFi (maybe you'd download a map for your trip the night before). And also, as I understand it, the maps require regular maintenance.
The real question is how will this scale? And honestly I don't know. My theory is the tech is so hard and the rewards are so great that infrastructure will change to make it easier. By installing smart traffic lights, having cars talk to each other, and adding various road marks specifically for self-driving cars you can drastically reduce the complexity of the problem.
You're making a point about end-to-end trained ML systems; if anything, that's an indictment of Tesla's approach, not Google's more traditional sensor-based-approach.
It's not a question of detecting the potholes, but of localizing the vehicle and tracking other objects in the vicinity (like other cars in traffic). High-fidelity maps make the system more robust because it can eliminate lots of measurement noise by averaging the readings from a constellation of landmarks.
> can't they do like the Kinect
This is essentially what they are doing. The problem is more difficult when the Kinect is moving.