Hacker News Reader

Comment by langsoul-com | Hacker News Reader

Is it really worth it to invest that much money into self driving. Like the sheer numbers is out of this world. Surely there's better investments.

View on HN

There are two drivers here:

1) Non linear payoff if it does work. A billion seems like a lot of money but at the volume of cars that Hyundai produces (millions), it's not a lot per vehicle and as a percentage of their revenue it's also pretty modest (about 1%). This stuff could add a lot of value to their cars.

2) Fear of missing out if somebody else manages. The nightmare scenario for Hyundai would be Tesla or some other competitor managing to launch something that actually works and then having to license their solution just to keep up; probably at great cost.

There are a lot of nay-sayers out there proclaiming at length how all this is a mission impossible. But the reason these investments are happening is that there are by now quite a wide range of autonomous taxis driving in a growing number of cities in the world. And they mostly seem to get their passengers from A to B without a driver. So, it's clearly not that impossible for this to work. Going from hundreds of bespoke cars operating for the likes of Waymo to millions of autonomously driving cars is the game here. Somebody is going to get there eventually. But not without a lot of investments.

Hyundai is opting to hedge their bets here. They are investing a lot more conservatively than e.g. Tesla. But they are not completely ignoring it either.

> Somebody is going to get there eventually. But not without a lot of investments.

There's still a bunch of major technological problems to be solved.

Taking Waymo as the best right now: 1. Their solution requires detailed mapping for all areas before it works 2. It doesn't appear to work with rain 3. Or snow

Like, none of these challenges are insurmountable, but as they say the first 90% takes 90% of the time, and the last 10% takes the next 90% of the time.

And more generally, even if the technological problems were solved, there's the regulatory concerns, in that who takes liability when (not if) there are accidents.

Given all that, I wouldn't expect to see full autonomy for about 10-15 years.

Mapping isn’t a major issue because there’s surprisingly few roads across a country. The interstate highway system is 47,432 miles in the US, you could send a car down every lane every day without major issue.

The US has 4.09 million miles of road for 333 million people. If you assume it costs 10$ / mile / year which seems excessive that’s still a trivial expense.

Why did Waymo survive and Cruise fail then?

Cruise severely botched an accident investigation which they might have been cleared of if they didn't try to hide evidence from CA regulators [0].

Follow the laws and it's fairly easy to operate.

I was a happy customer of Cruise's btw and they largely solved the self driving taxi problem at the city level at least, but Waymo does have access to better mapping because of Google.

[0] - https://www.latimes.com/business/story/2023-12-05/gms-cruise...

From the outside, with no insider information, my impression is waymo was under a lot LESS pressure compared to others to exaggerate their successes. This is just my impression but to me it feels like waymo was able to do a lot better self-policing or was much better at marketing to appear keeping safety at a top priority. For example, I remember reading articles that said the equipment on early prototypes were multiple times the cost of the Toyota Prius cars they were mounted on. Things like this make me feel safer with a waymo.

The real answer though is money. Since none of these investments are close to breaking even, the only obvious answer as to why one succeeds vs fails is the continued additional investment of money.

Execution, having a good idea and lots of money doesn’t guarantee success.

Waymo is 4 years older than Cruise

Theres a world outside US

The important bit is the ratio of customers to miles of road. The US has 80+ people per mile of road, that radio may be slightly higher or lower in each country but it’s never going to be prohibitive.

I think 10-15 years is extremely pessimistic given we have cars on the road now delivering passengers. Scaling that is the game.

I’d say 10-15 years is an optimistic outlook for ”full autonomy”. Maybe in some regions of the world in certain weather conditions. Full autonomy in northern snowy countries isn’t going to happen in 10-15 years, imo.

I think 10-15 years is a good guess because the issue is about regulation and infra and these move much slower.

> given we have cars on the road now delivering passengers

No we don't, only local roads, not fare-paying passengers on freeways, not in NorCal. In Phoenix AZ they are still testing self-driving on the freeway on employees, not fare-paying passengers on freeway.

Wait until you see self-driving taxis on freeway and to/from airports, with fee-paying passengers, in traffic, in bad weather, at night. See how many accidents there are per million miles. Then the discussion moves to "How many accidents/fatalities are acceptable per million passenger-miles?"

> See how many accidents there are per million miles.

It's hard to do worse than human drivers on that front. Especially under the circumstances you mention.

Don’t we need data before we can say that? Human drivers are pretty bad but self-driving software is hard to compare appropriately since you can’t just use national averages which include conditions the self-driving systems don’t support.

> 2. It doesn't appear to work with rain

Waymos do drive in the rain. See [1].

[1] https://old.reddit.com/r/SelfDrivingCars/comments/126kznl/wa...

If you figure it out the reward is easily 100x that.

Not clear. Waymo has made it work, but cost per car and per ride is still very high. That should come down over time. It's not clear that this is a hugely profitable business. It's basically auto rental and taxi service, after all.

The way I see it is what Waymo + others (and even Tesla until last year pre v12) was the same as someone doing ChatGPT but not with generative transformers but with if statements. If user asks what is that, say this, etc.

And now Tesla changed the game completely and moved to a 100% NN based system which indeed requires huge amounts of data which they're the only who actually have. All the other companies have no chance of getting that unless they fit their cars with thousands of dollars of gear for free (because I think no one would pay extra to get it and they really need it in ALL cars).

So it's probably not a good idea in terms of monopoly, but I think it will be Tesla FSD that gets integrated in all the other brands.

Because I would want to put a NN as unreliable as GPT4 in control of my car ... They are struggling to put in guardrails and make it politically correct with text.

I guess AI truly is the new crypto because I'm seeing all the same hype patterns. And Tesla is always riding the wave.

I've heard from several folks working on self driving outside of Tesla that they're very concerned about how cavalier Tesla is being about safety. I recall hearing one story about an engineer who outright refused to ride in a Tesla with FSD enabled. All anecdotal, sure, but I've become very skeptical of Tesla's approach.

> They are struggling to put in guardrails and make it politically correct with text.

Are they? It’s quite hard to get politically incorrect outputs from ChatGPT nowadays.

How many things does it not talk about? One of the challenges of self-driving is that it has to fail safely so they have to handle all of these uncommon situations a chatbot can just give up on.

because waymo relies on a continuous mapping of the cities you can take it. that's why it is available in few US cities, but this does not scale at all. The point of those startups (and tesla's) is to produce something that can drive everywhere whitout prior information (fantasy world right now)

As far as the scaling, I think some would have said the same about Google Street View, but now the majority of streets in the country are mapped out. Google already has the infrastructure to be able to do the necessary mapping. And as Waymo cars become more common, I suspect they'll be able to fill in a lot of the gaps.

street view has images from years ago depending on where the google cars are mapping in that moment, waymo needs constant mapping of the cities it has cars in, therefore needing more man-hours. furthermore the driving laws differs from country to country, and in most of the world the road paint/signs are borderline indescifrable for a human, let alone a machine. i'm not saying that it can't be done, but it is already expensive in big US cities, it just do not scale well.

This argument keeps coming up, mostly from Tesla fans. Google does collect a lot of data, but it uses it mostly for two things: 1) location with loss of or errors in GPS info, and 2) street maps of where to go. The driving environment is too dynamic for static info to be useful for more than that.

Driving in static environments with GPS, maps and LIDAR worked two decades ago, after all. The hard problems involve the other players.

Yes, Musk is saying that Tesla will have robot taxis by August 2024.[1] He said that back in 2021, too.[2] And in 2016.[3]

Tesla vs bollard: [4] Tesla still has only a not-very-good level 2 system.

[1] https://www.cnn.com/2024/04/05/business/elon-musk-tesla-robo...

[2] https://www.youtube.com/watch?v=9vZ8eWBiWFE

[3] https://seekingalpha.com/article/4682448-tesla-sudden-third-...

[4] https://twitter.com/WorldBollard/status/1585626430154502145

It's not at all clear why that wouldn't scale, could you elaborate? Legitimately not being snarky. The existence of Google Street View seems to be a fairly strong counterpoint.

> The point of those startups (and tesla's) is to produce something that can drive everywhere whitout prior information

Nope. The point is to make asstons of money. One does not need to achieve this silly ideal in order to do so.

This is the Hyundai Motor Group that is making investments. They are the parent organisation of Hyundai, Kia and Genesis. They also make Commercial vehicles. FSD, at this point have a lot if challenges, technology and regulation wise. But it is clear that these companies see value in their offerings being able to be driven without full attention from drivers. For that, I think this is not that bad of an investment.

A billion is really just a drop in the bucket.

Reply on news.ycombinator.com