Comment by jayjay71 | Hacker News Reader

Comment by jayjay71 | Hacker News Reader

Parent Full thread

This is the way Google does it as well, and in my opinion is a fundamental architectural failure. There's a reason after several years Google still has tested in so few locations.

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.

You could drive down every road in the US for under 10 million dollars. So as long as you can keep them updated I don't think highly detailed maps are a problem. Especially if this is supposed to be a backup for GPS.

There were 4 million miles of road in the USA in 2013. Sure you could drive them - but drive them in a special sensor-equipped vehicle? In a fleet of such vehicles? In a reasonable time? Its a big job, and $2.50 a mile seems pretty low.

2.50$ is more than double what trucking costs per mile that should cover every lane mile at normal speeds. Now building 100 cars with this equipment is not going to be free, but trucks also burn far more fuel. Anyway, you doing this is hard, for a multi billion dollar per quarter company it's pocket change.

Add in hiring, benefits, vehicle transport and repair, sanitizing the data, hiring managers to manage the folks that do all that... any large project is unexpectedly expensive.

Still point taken that its just millions.

"benefits"

This is Uber we're talking about.

The drivers of the fancy LIDAR vehicles are direct hires:

https://www.uber.com/careers/list/20169/

> "Sure you could drive them - but drive them in a special sensor-equipped vehicle?"

Street view already did it... Not 100% coverage to be sure, but pretty darned close.

Any idea what that cost them?

The maps Google uses for their self-driving cars, as I understand it, require constant maintenance. Also just the amount of data is huge. If this weren't the case, surely they would have driven in many other places by now?

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.

I think they could do it every week, compared to the value of completely self driving cars. Especially if they get say UPS to add sensors to their trucks. The post office does something similar every single day.

Every self-driving car on the road is automatically remapping everything. Any discrepancies, like a new pothole or a disappearing tree, can be reported to the mothership for resolution and pushing out to all the other cars in the area.

Considering how often GPS gets confused here in DC, I'm amazed we are suggesting relying on it to drive the car.

They are using it for navigation not lane following. You can supplement GPS with dead reckoning for tunnels etc, but high quality maps let's a car be aware of landmarks and then use them to figure out where in the world they are. EX: With enough precision there is plenty of information in the picture to figure out where it's located. http://socialforest.org/wp-content/uploads/2014/04/Forest-Gr...

Considering how often human drivers get confused here in DC, I'm amazed that we rely on them to drive cars.

How did you get to this number? I'm familiar with a few "map the world" efforts, and this seems an order of magnitude too small. I'd love to hear about a breakthrough.

"Fundamental architectural failure" is a very bold claim to make given that absolutely no other approaches have worked as well as Google's for urban driving.

How do you know that? There hasn't been a competition since 2007, and that was before Google was involved.

* That we know of. Tesla et al. have not publicized urban driving videos/examples, so I'm assuming they're not capable of it as well as Google, right now.

https://www.wired.com/2015/04/delphi-autonomous-car-cross-co...

That article has 0 information about Delphi's approach. In fact, the Delphi car has LIDAR too. For all we know they could be taking Google/Uber's approach.

I used to work on this - the approach is very different. That's a pretty hollow statement on the internet though, so take it as you will.

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).

What is the approach, then?

At a very high level, instead of comparing everything against a map, we try to figure out what it is. For example, the way Google detects traffic lights is using geometry to infer where a traffic light will be. Their maps are so good that they know where the car is, and where the traffic light is, and just use math to figure out where in the image there should be a traffic light. Then they just look at the pixels and figure out which color is lit - red, yellow, green. The way we did it is constantly scan for traffic lights (you can do lots of sanity checks, for instance using GPS are we near an intersection?), which means figuring out whether or not a traffic light is in the image, and then once we detect one figure out what color is the light.

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.

There's a serious problem at the heart of "throw a bunch of data at it" strategy - driving has an extremely long tail that's relatively fat. You can hit the "average case" relatively quickly (you can assume that Google's done that for the SoCal environment), but being able to extend that dataset to other, less likely scenarios, like "heavy downpour in SF" or "freak snowstorm" or "sudden construction due to water main break" is really hard.

None of what you said has anything to do with my point that the pre-mapping and then LIDAR-localization approach taken by Google is the only approach shown to work.

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.

Reply on news.ycombinator.com