Hacker News Reader

Comment by BigBubbleButt | Hacker News Reader

> 2) $2.25B is a big round, and speaks to the level of confidence in Waymo as well as the massive cost to get to market

It's a lot of money, but self-driving cars are notoriously overhyped. As someone who used to work in the industry and has worked at more than one of the "top" companies (Waymo, Zoox, Cruise, Argo, Nuro, etcetera), you don't have to talk to many engineers to realize there is a looooong way to go before a meaningful product is ready for market.

I'm not saying it'll never happen, but it's not going to happen before my kid is old enough to drive - and I don't have a kid yet. Before people bring up "we overestimate 1 year out and underestimate 10 years out," I was actually working on self-driving cars over a decade ago.

View on HN

Waymo is already doing rides without a safety driver to passengers who have signed their NDA. Which means they're confident of their behaviour on some really difficult problems, like pedestrians.

Sure, it'll be at least a decade (if ever) before they can do everything a human driver can do, but a "meaningful product" they pretty much already have.

They don't have to support driving in blizzards or on unmapped roads to be a meaningful product. Uber is useful even though it's only available in large Metro areas, Waymo can be a meaningful product even if it doesn't work in many places.

> Waymo is already doing rides without a safety driver to passengers who have signed their NDA. Which means they're confident of their behaviour on some really difficult problems, like pedestrians.

If they were confident they'd lift the NDA. I really can't say much more, other than I disagree.

> Sure, it'll be at least a decade (if ever) before they can do everything a human driver can do, but a "meaningful product" they pretty much already have.

As per the blizzards, I'm not talking about a driverless car that can do anything a person can do. I'm talking about a driverless car in any environment (let's say Phoenix) that can compete with Uber or Lyft and possibly maybe someday somehow make a profit.

Another question to ask yourself - why is turnover so high if things are going so well? Nobody wants to stick around to become filthy rich and work on an awesome product that can change the world? Or maybe reality is a little different from the blog posts you read every few months.

> If they were confident they'd lift the NDA. I really can't say much more, other than I disagree.

Please say the little more you can (and thank you for sharing what you have already!) - there's been something overhanging Waymo for nearly 2 years and no one seems to want to even hint at what it is

(also, if you feel comfy, I wonder when you were last privy to the internal narrative there)

> why is turnover so high if things are going so well?

There's _so_ little external information other than Urmson saying this is far away, Cruise is pumping near-production status, Waymo (confusingly) is in production yet if they were, then they wouldn't been doing what they're doing now. Yet all the ride reviews in the latest PR round sounded awesome!

He’s said what matters. I’m adjacent to the field and agree. The main issue is that getting to 5 nines of reliability in SDC is extremely challenging AND it isn’t good enough. There are infinite random issues to deal with from a simulation/modeling perspective (construction, weather, rain/snow/mud/dust weirdly shaped vehicles, random Uber eats cars stopping in the middle of the street etc) then you also have issues with vehicle and sensor maintenance, maintaining enough bandwidth/compute to support a larger network of SDCs, change management in the vehicles that sees challenges w scale, HD map issues etc. All of these problems are solveable but there are a lot of them.

He and the public are familiar with Waymo's solution to edge cases, but after thinking on it, it's unacceptable to get into an accident every 20,000 miles just because the car doesn't realize it needs to dial out to a remote operator

Honestly they don't even have to support most roads and road conditions and still build a useful product. Thinks of malls, sports complexes, university complexes, airport etc. where you can have Waymo shuttles to help ferry people from fixed point A and point B.

Having self driving cars plying from airport to nearby transportation hubs alone could be pretty useful.

I also see the potential for building infra that is specifically for automated vehicles (private roads and such).

Waymo can still be 10 years away from a good autonomous vehicle and still deliver value to consumers.

> Waymo is already doing rides without a safety driver to passengers who have signed their NDA. Which means they're confident of their behaviour on some really difficult problems, like pedestrians.

Unless the NDA is to keep customers from talking about when the remote safety monitor takes over.

For many people the difference between a prototype and a product is very small. In this safety critical business, the difference is huge.

It can still be a useful product even if it often engages a remote driver.

> signed their their NDA

> they're confident of their behaviour

pick one

Killing a pedestrian without a safety driver behind the wheel will kill the company. They're risking that, so they're confident.

but you don't have any first hand data on how much time it's spent with the remote safety driver engaged, while the NDA supports the idea that the ride is not ready for public scrutiny.

Killing the company? Really?

Your decade of experience doesn't support your projection. 10 years ago, AI didn't show a clear path toward meaningful products either. IMO product innovation isn't on a curve, it's a staircase.

> 10 years ago, AI didn't show a clear path toward meaningful products either

That's not entirely correct. It's just that, as techniques become mainstream, they tend to fall out of what's considered to be "AI". As an example, speech recognition used to be firmly in the AI camp. Now that it is commonplace, it is only ever mentioned in the AI context if it is paired with ML. Regardless, there are lots of useful products using that technology. Same goes for many others.

Both Waymo and Cruise have received California’s approval to remove the safety driver. Both can go tens of thousands of miles without driver intervention from the safety driver. Sounds like it’s right around the corner for level 4. Throw in remote drivers for the extremely rare scenarios where the car can’t figure it out. It’s less than 5 years away for city driving in the major city areas that have lidar maps. 10 years for nationwide.

You realize you're talking to a guy who's actually seen the internal deadlines, all of the secret data, and a whole bunch of other stuff. And you're telling me the PR talking points as if it's something I don't already know to be complete and utter bullshit.

So how about this: wanna bet?

Needs more data, you're asking us to go with the antagonistic anon account saying ~nothing new versus $3.5 bil of new money. Something is happening in the next decade or you have the responsibility to blow the whistle, and you gotta pick one

> Something is happening in the next decade or you have the responsibility to blow the whistle

Blow the whistle? There's nothing unethical here - I'm just saying it won't make money. It's not like children are being exploited for slavery and it's my moral duty to raise awareness to this issue. Why would I, or anyone, open myself to getting sued into bankruptcy over this?

There's a big difference between making some vague comments about how I know this won't work, and releasing trade secrets. I don't have to release any more data than I am comfortable with (not to mention it would be illegal), and you don't have to take me seriously if you don't want to. But if you think I'm full of shit, you have the opportunity to make a bet with me and make some money.

It's fraudulent to raise money and make the statements Alphabet has been making if there's no path to profitability in the next decade. I'm guessing you have a more limited proposition (won't be a profitable passenger carrier in next decade) and more perfect information on that (no way it can work in rain or snow; fixed cost of mobile compute > $XX,XXX)

It's not fraud, it's just a bad investment. Unless Waymo is saying "we guarantee to have this technology doing this impossible thing and making a trillion dollars in revenue in 5 years" there is nothing illegal about it. Being disingenuous is almost never the same thing as fraud. By simply prefacing each statement with "we hope to" it removes legal liability. It's not illegal to be optimistic.

Staging demos where the cars perform better than average isn't illegal, it's just misleading. Likewise, making projections about revenue isn't illegal, it's just not grounded in reality.

I'm not accusing Waymo of fraud. I'm saying their technology isn't going to be ready anytime soon, and they aren't making any legally binding statements to refute that - so why would it be fraud?

Just for fun I looked up the dictionary definition of fraud.

fraud, n., wrongful or criminal deception intended to result in financial or personal gain.

When Dmitri Dolgov said that Waymo's cars are safe enough to drive his children, I think that he committed fraud. Maybe not in the legal sense, but it's wrongful and deceptive and results in financial and personal gain.

https://www.dmv.ca.gov/portal/dmv/detail/vr/autonomous/disen...

Just came out. Is the data correct or are companies underreporting disengagements? Cruise is at 20,000 miles per disengagement. 4 years ago they were only at 50 miles.

Those disengagements are not representative of all the disengagements the vehicle makes. Basically, the company gets to decide for itself whether or not an incident is severe enough it needs to be reported.

https://www.bloomberg.com/opinion/articles/2018-02-02/safety...

Well I can also make an anon account and make contrary statement to all that you've said so they're not in a position to place more weight on a comment than they already have.

It's just a post by an anon person on a public forum, nothing more.

People underestimate how the exponential improvements in computing hardware creep up though. It's like computers were rubbish at facial recognition, chess, go and the like for years until suddenly they were not.

Reply on news.ycombinator.com