Apple Expands California Self-Driving Test Fleet to 27 Cars | Hacker News Reader

Apple Expands California Self-Driving Test Fleet to 27 Cars

bloomberg.com

64 pointsKKKKkkkk152 comments

Project Titan is such a funny thing. Back when Chauffeur, Google's self driving car project was the only game in town, automakers found it pretty easy to dismiss, but when rumours of an autonomous Apple car began to circulate, the tone amongst car people began to shift. It's what got the automakers thinking "Is this self driving car jazz for real? Omg it might be for real! The tech companies are going to eat our lunch".

It took the carmakers a while to figure out which way was up. The Germans spent $3 billion acquiring Here maps, which was a misfire. Tesla was quick to implement Mobileye's EyeQ3 chip and single camera system and then promptly burned their bridges with Mobileye by pushing the system's capabilities way beyond it's safe design constraints.

Ironically, Apple had succumb to irrational exuberance themselves in moving so aggressively on Titan. Analysts notes a $10 billion dollar bump in Apple's R&D budget during the 2 years when Titan was big. Supposedly the project was a clusterfuck. Steve Zadesky, who was put in charge wanted to make a fairly conventional car with autonomous features, but Jony I've and a group of engineers were adamant about developing a robotaxi, which offered the best opportunity to "reinvent the mobility experience". They couldn't come to terms and executive leadership of Titan was broken from the start.

So however much things were scaled back, the project definitely isn't dead. 27 test vehicles isn't small potatoes. Maybe Apple still intends to build their own vehicle, I guess we'll see.

They could've done a lot more if they were more focused, but that's not really how Apple works. They don't really research based on current technology, they like to spend time & money what could be invented. They even investigated things like "spherical wheels" [1]:

> Apple even looked into reinventing the wheel. A team within Titan investigated the possibility of using spherical wheels — round like a globe — instead of the traditional, round ones, because spherical wheels could allow the car better lateral movement.

[1] https://www.nytimes.com/2017/08/22/technology/apple-self-dri...

In case anyone is excited about spherical wheels, they have been looked at many times. They have the unappealing feature that the area of tread making contact with the road is minimized, giving bad grip and high pressure.

They are used occasionally in robotics on very hard, smooth, dry surfaces.

Wouldn't this be fixed by adding more "tires" then?

If adding expensive complex components is your thing, sure.

The more popular option is to use regular wheels. Cars go forwards a lot and other directions very much less.

You make some great points here. However, isn't 27 vehicles small potatoes? Tesla has almost 500,000 cars on the road and it also has 5 billion miles driven. That's a lot of training data. The approach to self-driving cars is largely training a model, so the data from 27 cars is actually extremely underwhelming.

Tesla doesn’t have 500K cars. I believe they have sold close to 150k cars till date. But your point remains valid. That is a lot of data compared to 27 cars

I had VIN #150,000 18 months ago, and they crossed 200,000 in May 2017 so it’s likely sitting at just shy of 300,000 right now. The 150k figure you are citing is US only (where there is a race to 200k when the federal credit expires).

You are correct. I was thinking US numbers due to the focus on Fed credit

27 cars is a start. In terms of fleet size, Waymo, Uber and Cruise each have about 200 test vehicles on the road, gathering in the ballpark of 4 terabytes of data per hour, and they currently have $150k or more of sensor+compute hardware on each vehicle, they use high definition 3D maps and are focused on validating their software to function safely within limited, geo-fenced areas; a realistic objective.

If Tesla were serious about developing an autonomous OS they would need to take a similar approach, but since all they're trying to do is summon the demon they can get away with a pentagram, a 3 horned goat, a copy of the necrinomocon and some candles.

since all they're trying to do is summon the demon they can get away with a pentagram, a 3 horned goat, a copy of the necrinomocon and some candles.

Any human sacrifice in there?

  Any human sacrifice in there?

Probably not until the next recession.

What about Joshua Brown?

https://www.bloomberg.com/news/articles/2017-09-12/tesla-pro...

Those Tesla cars aren't collecting training data, that's preposterous. They have a mobile phone connection and some flash storage, so where is all this data going to?

It's not going over the mobile link, you would need many magnitudes the bandwidth, and it's certainly not going on the flash storage, because if you continuously stream raw camera data to flash you will kill it in a year at most.

(That is before you consider that you want a $100k sensor suite with lidar, precise wheel odometry and a GPS/INS system to get a ground truth for training anything)

Hmm, I have not considered how they're retrieving driving data. I'm sure they're getting that data back somehow. It's a lot of data to waste.

That’s incorrect. Tesla’s connect to home WiFi and use that to transmit large volumes of training data. Owners have tracker huge packets being moved via their home connection

Man, I find this inside baseball kind of stuff fascinating. Where do I sign an NDA to learn the secrets of the past 10 years of Apple?

Car makers had self-driving research labs long before Titan leaked out.

It's kind of cool seeing tech behemoths battle each other.

They aren't strapped for cash, hire brilliant people, so you really get to see the impact of different strategies and other factors at play.

I'm, at least, enjoying the show.

It's interesting how the incumbent auto companies are competing with Apple/Google. They have a lot of ways to go to be able to compete for talent.

To be honest, I'd be pretty happy if apple would just let me use google maps with carplay. Can we make that happen before we figure out robot cars, apple?

Don't hold your breath. We'll all be forced to use apple maps until the end times.

I mostly like CarPlay, but the navigation is the sticking point, not because the directions or metadata are bad, but because the search is so, so, so bad.

I've rented cars with CarPlay before; I find that the ideal way to go about it is to do a location search as I'm walking to the car and start navigation; once I plug in the phone and enter CarPlay, it picks up seamlessly onscreen.

Agree, getting to where you are going is much easier if you get into Apple Maps prior to engaging CarPlay. Recently I got trapped in a location that CarPlay would not let me out of but when I disconnected the phone from the car I could get back to entering my desired destination.

I am not sure if UI elements went off screen or there is a bug/quirk when its active. However it selected a destination from ages for me when connected.

I just find it odd than when I use CarPlay to play music the UI is actually worse than if I use the car's built in display to show what the phone/ipod is playing. It was like Atari 2600 vs modern display type differences

This is actually what I do.

It works so well for text message interactions though (for me and my accent). I’ve had whole conversations with my wife only through Siri dictation and have come away mostly impressed.

Agree, search is basically unusable, (drive off the road, dangerous unusable!)I was surprised when recently started using on our new development Honda Civic.

I can't speak on car play, but I switched back from GMaps to Apple.

I hadn't used it since it first came out (when it sucked) but have found it's a lot better now. I especially like that it gives me the speed limit, a feature I wish Google had. I have removed Google Maps from my phone altogether now.

I think outside of the US Apple Maps is still super average.

In Germany it works well. I actually think the lane indicators for navigation on the Autobahn are superior to Google Maps. If only I found a way to always have north up. I get super confused and lose orientation if my map is not properly aligned with north up.

The last few times I've used it in Europa it's given me the wrong location for an address or missing places. I agree that seems better than it was originally, but it just can't compete in a lot of locations.

At best! I’ve tried for a long time to use it, but given up. It fails to find even well known locations in densely populated areas, directions are terrible, and in many cases you just get gray blobs where detail is crucial.

I use the web version of google maps more often than not.

The biggest problem with Apple maps is that it is terrible at working out where you want to go. You have to know the correct name that they have in their database. Google tends to have a lot more context, and so can often work out what you want, even if you don't know the exact name, and so can get you there.

Assuming that you can find the address that you want, then Apple Maps if good enough.

It doesn’t even show subway station when unzoomed. And it’s one of the most important way to understand what part of the city you’re looking at. At least where i live.

It’s just catastrophically bad.

Waze shows speed limits.

I've seen speed limits in Google Maps.

It's only shown in the Bay Area and Brazil. Really.

The API itself has more data, with varying levels of quality around the globe.

I'd see them in Atlanta.

They’ve shown up in Detroit.

To me its not the quality of the maps, but trying to extend their walled garden to my car

I liked Google Maps on my iPhone right up until they started nagging me to sign in before I could get directions. The minute I saw that screen, the app was gone.

On the flip side, if I weren't able to type "hom" and have it autocomplete to my home address I would be pretty pissed.

(I'm a dev on google maps iOS) We have never required sign-in for directions, you could always just tap the destination field and start typing.

For a long long time that particular screen had no history whatsoever if you weren't signed in, but that's been fixed for a while now as well.

It's possible I misread the screen that popped up when I tried to search that said something along the lines of "Choose an account to sign in" instead of showing me the map I asked for, but if you're looking for product feedback: whatever screen it was, it was enough to get me to stop using the app.

Where I live Google Maps is just so much better then Apple maps that never use Apple. Plus Google is just doing so much with maps then Apple is doing.

https://www.justinobeirne.com/google-maps-moat/ Google Maps's Moat - Justin O'Beirne

As much as I want to agree with you, from a product perspective, it's problematic: do you offer only one option and give Google virtually a monopoly in the navigation market, or do you offer multiple options and potentially confuse users (a la the Paradox of Choice?)

What do you mean? I think it should offer you whatever navigation apps are installed on your phone that support it.

Google Maps has been awful for me lately. Missing turns, not being forward enough (as in, telling me "turn left/right after the current turn). It also has given me completely wrong instructions on where to go at the end of the route to get to the destination, one time even getting me there, having me drive by it, then go a few blocks down a residential street, and say "you've arrived".. when I was clearly not at my destination.

Might try Apple Maps again.

I've seen a similar problem with Google Maps navigation. Several times recently instead of being directed to a particular store in a strip mall, I'm directed to a residential street a block behind the strip mall and told I've arrived.

Not sure why you got downvoted. Perhaps some Google devs can't take the heat.

I think they have a problem with the location database. For instance, the location of my home, according to GMaps, is not on my street but on the parallel one. They have probably tried to do some smart optimization and f*d up. The only issue is that there's no clear way to tell them they have made a mistake...

Try the use Apple maps from time to time and always go back to Google. Plus Google keeps adding and improving while Apple maps barely changes.

https://www.justinobeirne.com/google-maps-moat/ Google Maps's Moat - Justin O'Beirne

I might be wrong, but I thought the issue was the Google Maps haven't built a CarPlay UI for their iOS app? Is it actually Apple that's preventing them from doing it?

IIRC, there is no api for third-party maps on iOS+CarPlay.

Reply on news.ycombinator.com