Hacker News Reader

Comment by mdorazio | Hacker News Reader

A few things of interest here.

1) They're basically duplicating the early Google cars with mostly the same sensor suite bolted on top and integrated in the side panels. This is good because it means that unlike Tesla, they're planning for fully autonomous.

2) Since the human driver is still responsible for switching lanes, it looks like this is very early stage testing and data collection.

3) I think it's great that they're taking the approach of giving customers a view of what the car is doing via the back seat tablet, and even encouraging them to share. This is a great way to get people comfortable with the tech.

4) It still saddens me that everyone is keeping their autonomous data proprietary. How much faster could development go if Google let anyone have access to their million+ miles of driving sensor data instead of needing to collect it all from scratch?

View on HN

>if Google let anyone have access to their million+ miles of driving sensor data instead of needing to collect it all from scratch?

You basically suggesting Google to give up their(only?) leverage. Open source is all good, but if they can't make money out of these projects, you will see less of moon shots.

Nothing stops them from actually selling their finished automation software.

This seems like the kind of domain where the government can be useful. Negotiate terms to buy the aggregate data off everyone experimenting with self-driving vehicles and make it all publicly available for them all to use. Its a safety issue more than anything else, because we have borderline genocide going on every year with tens of thousands of roadside deaths, making cars safer is not just some side project for profits sake alone.

> This seems like the kind of domain where the government can be useful. Negotiate terms to buy the aggregate data off everyone experimenting with self-driving vehicles and make it all publicly available for them all to us

Why should the tax-payers subsidize private actor's R&D? I'm sure Google is amenable to partnership/sharing their data with anyone for the right amount or under the right terms; free hand of the market and all - no need for government "meddling".

You could claim the data is a public good. If the data is useful for companies beyond Google or Uber, or even other industries, it is plausible that it confers a public benefit, a benefit that (at least theoretically) is not being accounted for when Uber and Google do their calculus about how much to invest in data collection. In that sense, government would be the perfect actor to support -- and perhaps even subsidize -- this data collection in exchange for having it shared with the broader community.

Because then you keep kicking the problem down the road. Groups with cash treasure troves can reap data, and certain actors will never release the data, and in practice you won't see many organizations negotiating such data because it is valuable but its value is highly volatile based on what comes of the R&D. I cannot imagine almost any department willingly just selling it under current conditions, or else they may face the reality they effectively sold the necessary work to get the first fully automated system to market for what becomes pennies on what the reaper of that gold mine receives.

Because this data is a matter of survival, where you can easily forecast at least several million vehicular fatalities in the next decade worldwide, it becomes relevant that the various states (I'm extending this globally since several European car companies are also engaged in automation development) it is paramount it be made accessible in the same way we seem to be finally having the epiphany that should have happened at least two decades ago that scientific research should be made openly available.

> Groups with cash treasure troves can reap data, and certain actors will never release the data, and in practice you won't see many organizations negotiating such data

You make it sound like it is such a travesty or a great injustice, and I cannot for the life of me figure out why. Generating this data is not rocket science; it is fairly well-documented, the components are commodities and the process repeatable, one of the big challenges is that it is capital-intensive to scale. If the data is valuable as you postulate, raising the required capital should be easy, capitalism has been solving this problem for centuries.

> Because this data is a matter of survival, where you can easily forecast at least several million vehicular fatalities in the next decade worldwide,

No, it really isn't a matter of survival (at nation-state level you've elevated the discussion to). There are things that kill more people than vehicular accidents, so if your angle is "saving lives", then world governments really need to step-up their health game and meaningfully fight heart disease and cancer; hell, maybe they could even buy all the pharmaceutical research data and open that up to the public while there are at it (good luck getting a good price from GSK, Pfizer et al).

What if they sold the data to Uber? Everyone wins!

Just because Uber starting field trials with passengers first, doesn't mean they are ahead of the game. Google have been doing it for years and you know how good at they are at this. So, Google will not sell the data, rather they will improve their offerings and just partner with car companies.

Well, you will run their software, and they can sell you ads in your car where you have nothing to do but wait.

My take on Google X projects is to do cool things that may find alternate revenue streams to selling ads (or just do cool things, without necessarily having a business plan up front). Sure Google makes most of their money from ads, but they are trying lots of different things to make money elsewhere. Self driving cars, Ara, Google Glass, and others weren't trying to push ads, they were trying to find new products to make money in some other way (maybe).

Google's push to expand their Cloud offering is also showing them trying to pursue non-ad based revenue.

Google knowing where you go could make their models of you more valuable. Now they can specifically target ads based on where you go and when. Sitting in the cab waiting to get where you are going what are you going to do? Pull out your smart phone and see an ad offering a sale at a competitor to where you told the cab to go.

Fuck that, I'll walk. I live a life with the least ads as possible and even then I am smothered by them.

I like open source and open data in principle, but hate how it is driving more of the economy to an ad-supported model.

I don't see the connection. The core code and data of ad-supported companies like Google and Facebook are closed, while the open source companies and projects like Red Hat and OpenStreeMap are not usually ad-supported.

If that's the end game google can license their software to other car vendors. No need to giveaway the data.

And they run MY software, without giving much back.

-- disgruntled open source dev

I can eventually see regulations forcing companies to share particular types of data, such as that associated with accidents. Especially more unique accident scenarios. It'd be pretty limited in scope, but sharing that sort of data would have a pretty clear public safety rational even if they're just using it to build up simulations. No doubt there would be a lot of difficulty in just finding a way to make it portable.

I don't have the familiarity with what they're collecting (and how) to make any actual guesses on how you'd go about it.

Beyond that, no company is going to want to share their general data. There's too much time and expense behind its collection, and just sharing it wouldn't be enough; to make sense of it they'd have to share a lot of insights into their proprietary systems as well.

Different sensors may have different responses. A separate collection would likely avoid any particular biases that Google picked up, as well.

On the flip side, the different biases of different manufacturers might not go so well together. Toyota couldn't get the software for an accelerator pedal right, so I have zero confidence that the 2030 self-driving Prius won't just e.g. drive 15 mph under the speed limit everywhere, camp out in the passing lane, stop for at least 30 seconds when switching from reverse to forward, and gridlock in shopping center parking lots.

Then again, as someone who won't buy a self-driving car for as long as possible, it might be fun to learn the driving styles of different self-driving cars and use them to my advantage. E.g. If I know that tailgating a Prius which is going the speed limit in the left lane will cause it to move over in self-driving mode, or that a Mirai will always brake-check itself when it gets cut off, it might be possible to exert some control over them if it's advantageous in traffic to do so.

> Toyota couldn't get the software for an accelerator pedal right, so I have zero confidence that the 2030 self-driving Prius won't just e.g. drive 15 mph under the speed limit everywhere, camp out in the passing lane, stop for at least 30 seconds when switching from reverse to forward, and gridlock in shopping center parking lots.

All the things you describe are as likely to cause accidents as wrong turns or over-acceleration. If Toyota can't get this right (my bet is they will) then they won't and shouldn't be doing autonomous cars, despite their "change of heart".

I agree the data should be open, I just don't think Uber, or anyone else, should attempt to piggyback on top of it.

Pushing other peoples' cars around does sound kind of fun, but I hope we'll have something better to do than frustrate commuters.

Number 3 I totally agree. I don't agree with number 4 though. I think if you spend millions of dollars, time and energy to amass data then you shouldn't share it with your competition. You'd just make everyone's job easier.

>You'd just make everyone's job easier.

How is that a bad thing?

1,300,000 people die each year in car wrecks and another 20,000,000 - 50,000,000 more are injured and/or disabled.

Google's motto is "Don't be evil" after all. Is valuing short term profits over millions of lives not "evil"?

>Don’t be evil. We believe strongly that in the long term, we will be better served — as shareholders and in all other ways — by a company that does good things for the world even if we forgo some short term gains.

-Larry Page

You can't win any argument with

1) Google said don't be evil 10 years ago

2) I THINK that X is evil

3) Therefore, Google should not do X

Killing puppies is evil. Is not open sourcing sensor data they spent their money money to get evil? I don't think so.

Neglecting to save millions of lives in order to make some extra money is okay, but killing a dog is not?

Who's to say that opening up the data will save more lives than Google continuing on their current course?

If google opens the data (that I would argue is still incomplete), and irresponsible actors are given a means (via incomplete LIDAR maps, algorithms, etc) to mislead people already screaming TAKE MY MONEY, that would likely lead to a TERRIBLE image for self-driving cars if and when those race-to-the-bottom companies cause injuries and deaths.

I could foresee that those irresponsible actors could then scapegoat Google's data as the source of issues, whether or not that is true.

BUT I suppose injuries, deaths, and ensuing lawsuits are what prompts stricter licensing and oversight committees. I believe however that strict enforcement of self-driving cars will be an eventuality, no homebrew kits I'm afraid.

Killing a dog is a direct action. Yes it is wrong, at least unless there was a shortage of food.. then I guess it beats cannibalism.

Running a private experiment to gather data.. that data is private. Sharing that data is opt-in, not opt-out. Google has no obligation to release it. (If they do, cool. But not evil of them not to).

Now it is valid to say the government should run more studies on self driving cars and share that data. But I am not sure any business would exist in this country if every private R&D effort they did got open-sourced.

What if there was a middle ground where you could run your self-driving software in a Google simulation data set pulled from real world situations, so Google could essentially hide their data behind a testing black box? (Think codingame, etc). They could charge companies for the privilege of testing their software against that data.

It's interesting to consider Google's data from another perspective:

> “Uber logs as many miles in 24 minutes as Google’s autonomous cars have logged in their existence,” they wrote. “While none of these miles are fully autonomous today, we just point out the scale of experience that can accelerate the development of the AI, mapping and learning for autonomous cars.” [1]

[1] https://www.bloomberg.com/news/articles/2016-09-12/google-ca...

> Uber logs as many miles in 24 minutes as Google’s autonomous cars have logged in their existence

Hmm… Let's try comparing 24 minutes of miles logged in Google Maps (and Waze) with Uber's data and see where they stand then…

>It's interesting to consider Google's data from another perspective:

>> “Uber logs as many miles in 24 minutes as Google’s autonomous cars have logged in their existence,”

Google's data is not limited to its autonomous driving project, despite what the cleverly-worded sentence implies. Off the top of my head: Google also has 3D LIDAR data from human-driven StreetView vehicles, GPS and accelerometer data from millions of mobile Waze and GMaps users.

The 24 min quote is strange, they only have 14 Ford Fusions equipped with sensors, so probably not more than Google, so I am not sure they are comparing LIDAR data. It seems this is about smartphone GPS timestamped coordinates of their taxis versus the google LIDAR data, which seems a strange comparison to me. Google also has a billion of unsuspecting people having their smartphone working for them, but the data quality is probably lower than Uber GPS coordinates.

4) Maybe data is useless if you don't use the same sensor system.

c'mon, that the race for competition ! The great competition that drives the world forward ! I for one welcome my new competition overlord, blah blah blah

Based on the article, it sounds like the driver is only responsible for changing lanes when all normal lanes are blocked: i.e. a truck is blocking the only lane on a 2 lane road, and the driver has to cross over into oncoming traffic to get around it. While this is disappointing, it's very understandable that they haven't automated this process yet.

Tesla is also planning for fully autonomous.

Not anytime soon. And not with their current sensor suite.

As soon as they can.

https://www.tesla.com/en_GB/blog/master-plan-part-deux?redir...

As the technology matures, all Tesla vehicles will have the hardware necessary to be fully self-driving with fail-operational capability, meaning that any given system in the car could break and your car will still drive itself safely. It is important to emphasize that refinement and validation of the software will take much longer than putting in place the cameras, radar, sonar and computing hardware.

Even once the software is highly refined and far better than the average human driver, there will still be a significant time gap, varying widely by jurisdiction, before true self-driving is approved by regulators. We expect that worldwide regulatory approval will require something on the order of 6 billion miles (10 billion km). Current fleet learning is happening at just over 3 million miles (5 million km) per day.

Here Elon mentions a milestone of 6 billion miles. At their current rate, Tesla will hit that in 5 years, but clearly we can expect the rate to increase rapidly.

What struck me as interesting about Tesla's approach is they are using a much cheaper sensor suite than Google/Uber. If they succeed it means they'll have developed algorithms (patents) to achieve a self-driving car on cheap hardware. Giving them 2 competitive advantages.

Definitely not with their current sensor suite, but I don't see how anyone is gonna get there any sooner than Tesla.

However, their cars are primarily for personal use, not on demand travel

Not for long https://www.tesla.com/en_GB/blog/master-plan-part-deux?redir...

These guys are sharing their initial dataset, though it seems limited and may be a recruitment gimmick: https://github.com/commaai/research

A million miles? That doesn't sound like a lot. Uber currently only has 14 self-driving cars but they are approaching 200,000 drivers. In 2-3 years, they could require a certain level of "automation" for all their drivers in order to gather data. They'll have hundreds of millions of miles.

How many Tesla cars have with their self-driving feature? 20,000 cars? They'll have 100,000 cars RSN.

By 2020, we will have 2 orders of magnitude more data.

> By 2020, we will have 2 orders of magnitude more data.

Sure, but so will Google. Their latest claim is > 1.5m autonomous miles and thanks to their Streetview project they have a ton more detailed mapping data than that.

Google doesn't currently have a plan to scale.

Btw, I'm not saying that Google didn't do a great job, and make valuable progress. The game, however, is about to take a huge leap. Google was an early investor in Uber and many thought they were going to cooperate.

I'm not sure why you're being downvoted... Tesla already collects as much sensor data in a day as Google has in the entire history of their program.

Because it's wrong.

First, tesla collects zero lidar data. So no, they don't, because they don't collect the same kind of sensor data.

The second assumption is that the sensor data google is collecting, that is similar to tesla's, can only be collected by google's self driving fleet ,and not any of the other ways google has to collect data.

Because, as above, it's not the same kind of sensor data, this is also false.

And Uber? i didn't make any claims about the current status, but a few years out. Tesla is using radar, instead of lidar, and cameras. Their goal is fully autonomous and they believe they can do it with their current set up. Their equipment could change, of course, but they will soon be shipping 100,000 cars a year.

While you're picking nits and calling things wrong, note that a lidar-equipped Tesla has been seen at Tesla HQ. So it's not zero, but that's OK, I won't refer to your statement as being "false", just over-confident.

One could say something like 'Reddit produces more words in 24 minutes than Shakespeare produced in his career', but there is a substantial qualitative difference between those two datasets.

Reply on news.ycombinator.com