Hacker News Reader

Comment by msoad | Hacker News Reader

TLDR: It's not self driving. It's data collecting cars for now.

I know a little bit about this. It's not "self-driving" car per se. You know with current technology we can't have completely driver-less cars. But to get there we need a lot of data. Uber and Lyft saw all the rides on their systems and thought if we had a lot of sensors in cars and recored all of that we could have the winning factor for self driving cars which is data.

In order to record that data they need customized cars. They also need to own the car to own the data associated with it (this is my assumption).

View on HN

That's not the TLDR.. that's not in the article at all. You are correct however that it's unlikely that Uber would commit a large order for a car that doesn't exist yet.

It would make more sense if they were ordering their own fleet of standardized, sensor-laden, optimized UI cars. But that's not what this (dubious) news report is about.

I meant my comment's TLDR. Sorry for confusion

Strangely, it seems the etiquette would be that a TL;DR at the beginning of a post suggests a summary of the article. TL;DR at the end summarizes your comment.

Do you write this based on the article (where?), or on information from some different sources (private?)? I couldn't find such information in the article, while it mentions that e.g. "In August 2013, Mercedes-Benz [...] drove [103 km] between the German towns of Mannheim and Pforzheim without any driver input", which seems to have happened indeed (https://www.mercedes-benz.com/en/mercedes-benz/innovation/au...)

I meant my comment's TLDR. Sorry for confusion -----

Or, if you're Tesla, people pay you for the privilege of collecting data for you. [0]

[0] https://backchannel.com/how-elon-musk-and-y-combinator-plan-... "Probably Tesla will have more real world data than any other company in the world."

Interesting from a couple of angles.

First, there's this widespread handwaving around getting the very detailed map data that is at least useful for even limited autonomy. At one level, it's a solved problem that that doesn't mean it won't take a huge investment to collect all this data.

The second is the idea that this mapping data may be a source of competitive advantage for individual firms working on improved assistive driving systems followed by autonomy--initially under limited circumstances. On the one hand this makes a lot of sense. On the other hand, it means that a lot of work probably ends up getting duplicated.

> You know with current technology we can't have completely driver-less cars.

Why not? Doesn't the Google car have a near-spotless record with minimal human intervention?

Under a very limited set of conditions. Anyone who believes that this translates into door-to-door general driving without a human even present anytime remotely soon is... optimistic.

Living in Mountain View and driving with Google's self driving cars every day, I have to say I think they are quite a bit further than you give them credit and those conditions appear to be less limited than you are letting on here.

Something I've long wondered: howwell do they deal with inclement weather?

They do well in rain but snow remains a problem, not due to icy roads (at SXSW Chris Urmson said they do well on slick roads) but that the "better than GPS" LIDAR-based location tracking stops working when snow is on the ground.

The car simply cannot figure out where it is when there's a significant layer of snow on the ground because the reference maps stop looking like what the car is seeing.

https://youtu.be/Uj-rK8V-rik?t=46m22s

I could be wrong, but last I heard google's self driving cars still required a high-precision 3D map of any route they operate on. They can drive on marked roads in mountain view and a couple other areas Google has mapped, but they're a long way from sensors that let cars drive an arbitrary route on any road.

Reply on news.ycombinator.com