[agildehaus]

It's important to note Google/Waymo's tech is limited to places that have been heavily manually mapped and test driven. They're basically creating a virtual track for their cars to follow, and presuming that changes in the real world can be detected and distributed to the entire fleet.

Other companies seem to be taking a generalized AI approach. Their cars work everywhere, but how well?

It'll be exciting to see which approach works best.

[wstrange]

This comes up on HN quite a bit, but I have never seen anything that suggests that Google doesn't also use a generalized AI approach.

Having high res maps certainly helps, and if you have that data, then why not use it?

It is no different than what people do. If you drive the same route all the time, you have some expectation of what lies ahead.

[jayjay71]

This information is not current, but I know that they used to detect traffic lights by looking for them in their maps. By that I mean they knew where the traffic light should be, then did the basic geometry to figure out where that is in their image, and then determined the color of the pixels to figure out what the light was saying. That absolutely would not work without the maps. I am willing to bet that their approach fundamentally depends on the maps. That's fine, it just means they can't work in areas that aren't already mapped (and you have to have one of the maps stored locally in the car, as those files are gigantic).

[magicalist]

> they knew where the traffic light should be, then did the basic geometry to figure out where that is in their image, and then determined the color of the pixels to figure out what the light was saying.

This has to be really old information, though. They have that video of the car detecting school bus stop signs and a police officer directing traffic. A stop light is child's play after that.

[jayjay71]

Why do you assume those other things are harder to detect? Computer vision is tough, and human detection is one of the most researched problems in the field. Uber's self-driving cars were running red lights not too long ago. And it's important to understand that a video is not a live demo.

If Google has an approach that works for them and it depends on the maps - that's fine. I'm just pointing out that they (probably) made a decision a long time ago that they are going to use those maps to further their self-driving technology. It's a design choice, with pros and cons, like any other decision.

[birdman3131]

And in steps google street view. They have been planning this for years.

[jayjay71]

Google Street View contains images, not point clouds. And based on my usage, it's updated roughly every year. This is not even close to sufficient for their self-driving cars.

[sangnoir]

> Google Street View contains images, not point clouds.

You are wrong[1]. Streetview vans have been outfitted with LIDAR for years and generate 3D models as they drive. Just because you do not see the point-cloud data in your browser doesn't mean Google doesn't have it.

1. http://google-street-view.com/about-google-street-view/

[ZeroGravitas]

You can see the basic shapes that the images get mapped to in the browser.

I'm sure someone built a hack that let you explore that directly.

http://callumprentice.github.io/apps/street_cloud/index.html

https://flowingdata.com/2014/03/26/reconstructing-google-str...

[jayjay71]

Neat! I stand corrected. I wonder if the map data they have all over the country/world/wherever is comparable to the cities they test their cars in. If so, that would be fantastic.

[foobarian]

Perhaps if the tech is good enough for now, it will become somewhat widespread, and then force a de-facto redesign of the traffic "API." E.g. maybe lights will have hidden IR identification beacons intended for self driving cars. Maybe traffic police will have special manual beacons that work on those cars while directing traffic, etc.

[ndm00]

This is a great point. Additionally, in a world where there are many competitors with self-driving cars, I can't imagine roads not having and API of sorts to simplify the work cars have to do. Instead of painted line markers, why not IR beacons or something similar. Could be the same for road work signs or temporary barriers.