It's time to admit self-driving cars aren't going to happen | Hacker News Reader

It's time to admit self-driving cars aren't going to happen

techcrunch.com

75 pointsgeorgecmu154 comments

I think that we won't be able to have full self-driving on standard public roads, for a while, where non-self-driving vehicles (and pedestrians and cyclists) mix with self-driving cars.

I think, if we had fenced-in and dedicated roads for self-driving cars, we could have had them yesterday.

But that ain't gonna happen anytime soon.

And we definitely won't be getting flying cars (for the general public), until we have full-self-driving.

> I think, if we had fenced-in and dedicated roads for self-driving cars, we could have had them yesterday.

We could put them on rails and have overhead cables providing electricity efficiently, and make them go up to 300kph too, damn the future looks bright !

Make them bigger so they can hold many people for one trip, while we are at it.

It’s likely profitable and better for the environment if we get electric self-driving trucks on private highways. 70% of freight is moved by trucks.

https://www.trucking.org/economics-and-industry-data

This analysis is immensely US-specific

It would also work in China, of course. In fact, I would expect them to implement it first.

https://www.cnbc.com/amp/2022/07/28/self-driving-tech-compan...

Trains are an order of magnitude more efficient, however, and there’s a huge cost & environmental impact to building those extra highways. What we should be doing is more intermodal work where long-distance trucking is replaced with rail & local delivery.

Yes, everyone knows trains are more efficient. They’re at full capacity and have been for decades.

It’s very difficult to add train capacity.

So, let’s not derail the conversation with your better idea that’s never gonna work

This is not even close to true unless you’re going to rescope your comment to focus on a few very specific chokepoints. There’s tons of unused mainline capacity because years of crew efficiency improvements mean the railroads are running fewer longer trains.

More importantly, however, climate change means that we can’t continue to ignore externalities. Redirecting road expansion projects into rail construction would be necessary even if it wasn’t cheaper.

Good one!

I've always been a fan of trains. I seem to be in a minority, in the US, though...

Trains are awesome. My thinking is that the implementation sucks.

I've been on a train twice in the last ten years.

On the one ride, it was hot so it expected to arrive about 45 minutes late. You can't go full speed on hot tracks.

On the same ride, there was construction, so we sat and waited for about 45 minutes in the middle of no where, in the dark. It was long enough they told us to just let our kids run around and brought the snack cart around for free.

That's a lot of delay for a 3.5h ride.

Ontario, Canada.

My in-laws use trains to travel between cities a fair amount. More delayed trips than not.

If you’re not talking about long distances, don’t get me started on the LRT in Ottawa and the months of downtime it has had.

I was on a trip to Switzerland last month - across a couple of dozen train, tram, trolley-bus and regular bus rides, I had just a handful that were delayed - three to four minutes late (mostly on intercity trains that I'm going for an hour or so journey on, so it was basically nothing). But mostly they were bang on time. It's ridiculous, they're absolutely nailing punctuality, cleanliness, frequency, interchanges, mode changes etc.

And if you happen to miss the connection, no worries there's the next train in 30 minutes.

If you are in Tokyo, it's more like 3 minutes.

Oh by I mean anywhere in the country, not only in the capital. But maybe it's the same in Japan, I don't know.

Depends on where you are.

Tokyo is like Type A++, so the trains are constant.

Other areas, probably not so much.

But you can get pretty much anywhere in Japan, via train (and maybe a short-haul flight in a 747).

Unless you live somewhere that the next one isn't coming at all, and now you have to wait twice as long, and jam twice as many people in.

Yes, if trains had roughly the frequency of hopping in your car whenever you want, and the density of roads, they would be used a lot more and many more people would not have cars.

Cf. The London underground.

I traveled to Tokyo for twenty years.

The Tokyo rail system is an official Wonder of the World.

I never looked at schedules, and the trains were reliable and cheap.

Crazy good.

Switzerland too - I was there for the first time for a couple of weeks last month. Didn't even think of hiring a car, using Uber or anything - the public transport is just too damn good. Even though I had a pretty good idea what to expect, I was still amazed.

The subways in Japan are fantastic. Interesting that they make a loss on the trains themselves and more than compensate in profit from retail and real estate.

>I traveled to Tokyo for twenty years.

Longest. trip. to. Japan. evar.

NYC subway used like that too. Didn't look at schedules, just hopped on and off at whatever station. Didn't feel like crossing under the east river in a tunnel so hopped off a red train and switched to a yellow train to use a bridge instead. All on a whim.

I wish public transport providers would take advantage of the possibilities of apps more aggressively. I'm in London and I rely entirely on CityMapper to use public transport (mostly) and some Uber. I don't have a drivers license - never got around to learning to drive (I'm 47... maybe one day) as I've always lived places with excellent public transport.

Apps like CityMapper revolutionised that in making me resort to Uber/cabs far less, but there's some really obvious improvements in terms of leveraging journey data to both offer "virtual" bus services where you can offer higher frequency than you have buses to service by filling in with on demand minibuses and cars from car hire companies, and in using the end-to-end journey data to adjust schedules more dynamically.

There are plenty of on-demand shared services (e.g. London's Dial-A-Ride) but most seems to be operated in isolation and/or targets people with limited mobility for end-to-end transport rather than to fill in smaller gaps in other services.

I saw one of these the other day in Cambridge, MA. Looks like they’re out of service here, but that kind of 10 passenger small electric shuttle bus seems great for filling gaps in transit networks: https://www.ridecircuit.com/about-us

I suspect the cost structure isn’t that great compared to a full size 60 passenger bus, though. The drivers per passenger ratio is pretty low, which probably more than counterbalances the lower operational cost of the vehicle.

It is technically possible, but every time someone suggests self driving trams/trains/metros here, the public transport union threatens a strike.

Find a way to eliminate mode switching and you might be on to something. Even transit activists recognize the high cost of mode switching and how much of a negative it is for practical use.

> And we definitely won't be getting flying cars

Seems I'm going to have to downgrade my New Year disappointment about "still no flying cars" to "still no self driving cars" (for the general public).

I think we already have a couple of viable flying cars, but the issue is that we really don't want Bubba Beer Blaster driving in 3D, when he can't even handle 2D.

Even if you have fenced in and dedicated roads, there are system control problems that will always arise. The more vehicles the greater the opportunity for failure of the system. Even today, simple systems like automated industrial plants have to have humans monitoring them closely to ensure that any errors that arise are actually handled properly.

Once you put a large number of vehicles in play, the entire dynamics of the system would become unmanageable.

You would require multiple hot redundancies at every level with human monitoring to at least minimise the things that could go wrong. It would only take one coding error for the entire system to fail and that coding error might only ever be found after some disaster has occurred.

It might be nice to have self-driving vehicles but the cost may be so prohibitive to make it infeasible.

This always had the taste of being too US-centric anyway. European countries (and I guess the rest of the world as well) has very steep, narrow and weird roads and don't run on neatly organized grids. So it's even harder to achieve then just fencing cars and separating traffic.

> And we definitely won't be getting flying cars (for the general public), until we have full-self-driving.

The irony here is that "flying cars" (which really means eVTOL planes) is in many ways an easier problem to solve in terms of navigation, because the notion of strict air corridors and tracking of every vehicle coupled with fewer annoying pedestrians wandering around makes it far easier to make those accommodations (e.g. virtual "fenced-in and dedicated roads")

Flying cars aren't and won't ever be viable because of fuel consumption. We're already worried about ordinary cars in this regard.

There has always and will always be a high end of the market, and that high-end will pay extra irrespective of fuel/electricity consumption.

So in the sense that they'll likely never replace "ground cars", you're right. But that does not mean there won't be some space for them.

The far greater problem for eVTOL companies than energy use is noise and safety which for the medium term at least will exclude them from the most attractive end-to-end options and prevent them from being a real car-replacement option.

E.g. Lilium originally touted themselves as a sort of "flying Uber" on the basis that it could in theory land almost anywhere. But they dialled it back on the basis that being able to is very different from being allowed to e.g. land in a residential area without a dedicated, secured platform and following all kinds of regulations on both noise and safety. Now, they're instead focusing on partnerships (e.g. with Ferrovial for Florida[1] and Saudia for Saudi Arabia [2]) willing to operate networks of what they call "vertiports", so more of a replacement of a segment of helicopter services than cars.

[1] https://lilium.com/newsroom-detail/ferrovial-and-lilium-deve...

[2] http://www.tradearabia.com/news/TTN_402471.html

There will never be a regulatory framework in which flying cars are allowed in urban airspace/ground space except perhaps in certain very special locations (like Saudi Arabia and maybe Florida) where there is rule of the rich rather than rule of law.

My take on this is always: we will see first fully automated airplanes and railroad trains before we will see a widespread of autonomous cars.

Airspace is an highly controlled area. We have many years of actual autopilots. We have many years of guided take-offs and landings. But still we relay on the skilled human.

The same applies to railroads. Here in Nuremberg we have an highly automated train running on a railroad track that is even partly shared with trains operated by a driver. But the costs of running the automated trains seems not to benefit so that further development has been stopped, because of costs reasons. Even has we have a huge shortage in drivers, which would suggests that automation would be the solution. AFAIK there are only isolated airport shuttles that run fully automated.

That means, if can't run automated systems in highly controlled areas. Then why would it work with cars in highly uncontrolled areas?

AFAIK there are only isolated airport shuttles that run fully automated.

There are several fully-automated rail systems worldwide. Paris M14 and the Copenhagen Metro among others are fully automated; things like the DLR in London are level 3 (driverless with a staff member on board).

It doesn't make economic sense to automate trains and airplanes because 1. Much smaller part of the cost and 2. Very high stakes.

Driving is high stakes too of course but two orders of magnitude less so.

That said, we are seeing massive amounts of money being made in ADAS systems (Tesla FSD, MobilEye) while robotaxi plays are losing billions. Self driving cars will grow out of increasingly capable ADAS systems rather than robotaxis because they are already profitable and there's a gradual path to full autonomy. In the coming recession I expect most robotaxi companies to go under (Argo being the first yesterday), with Waymo being perhaps the only player left. Even Waymo may have to reduce spending quite a bit if Google is facing headwinds.

And what is the reason why most rail transport in the world is not already 100% autonomous?

Not saying this as a negative thing. Because having well paid, healthy and rested employees looking after stuff like maintenance, safety, stuff like trying to stop the train when some idiot parks the car over the tracks, or when a badly loaded rail car breaks loose and derails suddenly, or just generally driving in places where fully autonomous operations can’t happen for various reasons.

But it’s the Unions. World wide one of the major things train worker unions have pushed back hardest on is smaller train crews for both passenger and freight… and it should be obvious why. It means companies that move to smaller crews would lay off union workers.

Plenty of this is actually justified by good arguments put forth by the unions, particularly in places like America where railroads have basically refused to spend money in capital expenses and try to run with the oldest railroad hardware possible in order to maximise their profits per ton per mile shipped, regardless of environmental impact, crew fatigue, or the long term degradation of the entire strategic value of a national rail network that can move the heaviest military goods efficiently in times of war.

You could not physically do automatic freight rail in America without getting people killed. The hardware and infrastructure are literally not capable of it. For all the automatic signalling systems in place, it’s all duct taped together around having a human in the loop that can be made responsible for decision making in the train driver cab.

In more modern environments like say Switzerland, they probably could get much closer, but they also have a well regulated railway system with incentives that ensure the companies don’t need to be so brutally effective to make enough money to meet corporate profits. Then there’s places like the trans Siberia railroad… good luck keeping an automated network of that size operating within safe automatic bounds over winter, or the parts of the UK where there are still physical mechanical based rock fall “sensors” in place to alert drivers to their being rocks on the track ahead that may have only landed there minutes ago… and they don’t run enough trains on the tracks to make maintenance of network of cameras and motion sensors worth it.

Railways are, when run well, very automation friendly environments. But that breaks down the further you get from a tidy metro area and surrounding country/farmland, further out into long stretches between cities or states where you would need to maintain thousands of kilometres of fences to avoid accidents, loss of property (stray farm animals) and the larger the network gets the harder it is to have it all owned and operated by a single company that could even begin such a project… sometimes you have a nationalised owner/operator but then it’s government budgets deciding if the actual rails and signals need the upgrades vs the potentially private companies that may own the trains and haul freight on the government’s rails…

The real world is messy. So the union does kinda have a solid point. But that doesn’t eliminate the fundamental fact that at the “big moving objects on continuously welded rails” level of controls theory, railways are very automation friendly and are in theory able to be nearly fully automated.

Stolen stuff is a common problem. People having fun putting rocks in the railway or emergency cases where you need a human controlling the behavior of a group of panicking people are still not fully resolved problems

It’s really difficult and I think we can afford one employee driving a vehicle transporting thousands of people.

Just as a note, one of the largest iron ore mining operations in the world has been runing autonomous supervised Trucks and trains since ~2015:

> Autonomous trucks [1]

> We run more than 130 autonomous trucks, part of our Autonomous Haulage System, across our Iron Ore operations. The trucks are operated by a supervisory system and a central controller, rather than a driver. The system uses pre-defined GPS courses to automatically navigate haul roads and intersections and knows actual locations, speeds and directions of all vehicles at all times.

> In 2018, each truck was estimated to have operated on average 700 hours more than conventional haul trucks, with 15% lower costs – delivering clear productivity benefits. They also take truck operators out of harm’s way, reducing the risks associated with working around heavy machinery.

In Australia though .. starting from sheep shearing robots in the early 1980s

[1] https://www.riotinto.com/en/about/innovation/automation

I'd like little Jetson pods on some sort of maglev tracks... But we'd need a lot of infrastructure to make that happen.

Separating traffic feels so 60-80s.

I think there won't be a big bang where we go from no self driving to fully self driving vehicles at once. It will be many small changes, that taken together will slowly transition towards full self driving cars. Since quite some years we've had cruise control, now there's adaptive cruise control. Then cruise control became able to overtake other vehicles. Next up, they may be able to take a certain exit based on satnav, etc.

It's hard to say when, but chances are that at some point in time we'll simply have FSD, while nobody realized that's where we were headed because it was all marketed as small(ish) individual features.

> It will be many small changes, that taken together will slowly transition towards full self driving cars.

The problem is that there is no smooth transition: there is a "valley of failure" in the middle of it, where there is enough automation that the human can disengage most of the time, but not good enough automation to avoid all errors, so the human operator has to step in to correct errors quickly, and doing do "cold" (i.e. from a disengaged state) is not good enough.

Dekker touches on this in "The Field Guide to Understanding 'Human Error'"

Good point. I think it will get even worse as automation improves. Consider that problems beyond thew ability of the automated driving system are delegated to the human driver. As automation improves, the human driver will have to be even more alert to resume control.

Yep, the sibling comment ( https://news.ycombinator.com/item?id=33370118 ) explains it well:

> humans are really bad at co-operating with machines. We are asked to pay full attention to everything around us and take over at a moment's notice, while doing nothing >99% of the time. Humans simply do not work like that.

I have been trying to find some refs.

here's a couple: https://www.skybrary.aero/articles/cockpit-automation-advant...

https://themobilityforum.net/2021/03/15/the-potential-pitfal...

Like most of the 'Human Error' research, it focuses on Aviation, but is worth a read anyway if you care about topics such as e.g. SRE and Incident response.

It seems to me that the biggest obstacle to that scenario would be legal. At some point the law would need to change to allow the "driver" to be asleep/watching TV/etc. behind the wheel.

The biggest obstacle is "if compiles, ship it" mentality. Just like when a human drives: cutting corners increases the risk of accidents.

I do not believe this will happen.

What we are seeing right now is that humans are really bad at co-operating with machines. We are asked to pay full attention to everything around us and take over at a moment's notice, while doing nothing >99% of the time. Humans simply do not work like that.

Meanwhile the AI acts like a black box, giving very little information about its inner state to the human. It can fail at any moment, without clear prior warning. It will fail in situations which look completely normal to the human. It will fail in situations it seemingly has successfully driven hundreds of times before. The AI is making decisions, but the human does not know why it is making them - or even that it is making them at all.

Humans are really bad at judging an AI's performance. We have already seen people falling asleep behind the wheel of their self-driving car, and that is not going to stop happening.

Self-driving vehicles are to a large extent an all-or-nothing thing. Until we can fully rely on it, it will only make the driving experience worse. Let's just stick to adaptive cruise control for now.

This paradox is already evident in commercial flight: as automation has increased, flying has become safer, but pilots themselves understand less and less about the machines they operate, and when automation fails they are ill-prepared to get themselves out of messes.

The solution is VR training, not reduced automation.

I agree that this is unlikely.

I have a new model car with all the lane-assist and following whizz-bangs. I'll never use them. When the lane markers get challenging and difficult to understand is when the assistance turns off and reverts to the human. Typically about 1 second before disaster were to strike. If it were even 10 seconds beforehand, then I'd maybe try using it again. But the 1 second window is too short for my personal limits. The assistance is completely backwards to me. I don't need it when lane markings are clear and easy to understand, I need it when they are unclear. Like, I don't need a translator to Swahili for most of the day in Mombasa, but I do need one when going over legal documents.

Personally I'd be satisfied with just being able to hand the highway driving over to the car. Keep me in the middle of the lane, don't let me hit another car or any object in the road, and sound a chime when I'm a mile from the exit.

Same, and current tech is more or less doing this. Even my 2016 Toyota Rav4 has lane departure assist and dynamic cruise control. I still have to keep a hand on the wheel and pay attention to the road of course, but road trips are quite nice just with those 2 features.

Now if only those systems would dodge debris on the road, like a truck tire or a piece of lumber.

Therein lies the long tail of challenges to be truly self driving, right?

If AI dodges the debris 99/100 the whole world will claim it as a failure because that 1 miss ended in someone's death - even if human drivers are only able to dodge 90/100.

The problem is 100% of economical, and not technical nature.

The car manufacturers are afraid of being made responsible for that deaths, even when you could prove a human would not have been smarter.

and also not suddenly dodge things that are not there

This happens to humans, too.

Ya, that would make long distance Highway trips more bearable. For city driving, public transit is probably the better option. Although FSD can be a great way to optimize limited city road resources, I don’t expect to see that in the USA anytime soon, but it could happen in China where the traffic is much worse.

Yes - I don't understand why this hasn't been the focus. It would certainly beat airtravel for certain city pairs in the US. Anything in a 800 mile zone I would do a self driving car on interstate vs a plane.

Seems a much simpler problem to solve than full self driving everywhere!

This is not a great article. It's based around Ford admitting they're not a software company, that they won't be able to compete effectively in this space. This shouldn't be a surprise to anyone.

This does not mean that self driving cars "aren't going to happen" -- in fact, the progress by industry leaders has been methodological and constant. The evidence clearly shows that it is happening, and has every indication of continuing to happen. Five years ago people were incredulous that a computer could pilot a car at all. Now, everyone has conceded that this is possible and the criticisms are about how a computer pilots a car and whether it should make more risky maneuvers. This is exactly what we would expect progress to look like.

As progress continues, companies will learn whether or not it is profitable to develop competing products, or concede to market leaders. Ford has conceded. Rivian has conceded. They will license technology developed by other companies with more capable technical teams.

The author seems to think that all self driving companies are basically the same, so Ford must just be facing the harsh truth that everyone else is ignoring, vs Ford being outcompeted by companies with much stronger ML chops

I don't know if Tesla originated the approach, but they're the first I'd ever heard of doing it by my camera or radar or anything like that. I remember my first reaction was befuddlement, when highways cost like $xxx million per mile, why would the trivial cost of dropping electronic friendly markers for lanes and such be anything but a trivial 0.1% cost that should facilitate self driving from the road construction end as a primary means of situating self-driving vehicles?

Instead you have cars with that forever risk of death to their passengers when self-driving due to the non deterministic factor of not having a highway equivalent of train rails.

exactly, NFC-like technology would be extremely inexpensive in the scheme of things and could be deployed into road surfaces without much difficulty. I remember seeing a video of self driving technology from the 90s that used a similar approach - magnets in the road IIRC

This is what I wanted to say.

Autonomous driving without changing a thing, will not happen soon. But changing some things, is totally possible.

It's inevitable that roads will eventually be built with electronic assists. However, there are good reasons to develop cars that can still drive without them.

Given that the government would have to do it, it's going to be a really long time to completely retrofit our roads with these devices. We're still not putting them into new roads right now, nor am I aware of any international standard for doing so. A common standardised database of road works would have been useful for decades of GPS navigation but we still don't have that either - most countries don't have one at all, and the ones that do aren't standard (often even varying from city to city) and tend to be incomplete. Relying on the governments of the world to sort this out is a risky strategy.

Secondly, even assuming we do get these systems in place, you can bet good money they will frequently be wrong. Some contractor will come and dig up the road but forget to move/disable/update the electronic markers. So having a parallel visual self driving system to help identify these situations is essential.

Third, pedestrians, bicycles, kids, animals, etc. are unlikely to contain these markers. For trains, we just ban them from the track, put fences around it, etc. but that's clearly not going to work for most roads.

They actually did try this in CA in the mid 90s: https://www.latimes.com/archives/la-xpm-1996-06-28-mn-19436-...

And a follow up story from 2016: https://laist.com/news/self-driving-cars-debuted-on-californ...

Actual self driving is likely to happen, but it's going to need some accommodations.

They might be high-resolution mapping with the ability of (local) government to mark changes as they occur. Road workers with devices that can direct automated vehicles to avoid holes in the road and the like plus other special cases.

Along the way there will be many stand-alone (self-parking) and in between (freeway/highway autonomous driving) functions that will be very useful and will be attractive to customers.

If we're only talking about "the holy grail" unlimited full autonomy, then yes, it's probably not possible, since it approaches general intelligence.

Self-driving cars will definitely happen. It won't happen next year, but it will absolutely happen. That's the way humans are. If they can do something and make some money, they will do it.

Your last part is important, however: it’s quite possible that full self-driving vehicles might end up like supersonic passenger air travel where the cost is too high to be viable.

For example, Uber’s valuation was based on the idea that self-driving cabs would be cheaper than paying a human driver. It would be very easy to imagine that if the vehicle costs $100k it’s cheaper to pay a driver minimum wage instead. This is especially worth thinking about in light of the climate crisis: single-passenger vehicles have much higher emissions than almost anything else, even for EVs, and if the users had to pay the accurate cost the break-even point would be even further away than it is now.

Talking to a developer about self-driving cars is a good litmus test of sincerity and/or competence.

Utterly unbelievable that Tesla is still selling "Full self driving" cars to customers.

It could probably be done today with richer sensors on the car than just visual and with "smart" roads outfitted with specialized symbols and transponders to guide the car.

Pure self drive like humans do requires a cognitive model of cause and effect and the nature of the world being navigated, not just pattern recognition and prediction.

Enormous advances in predictive and pattern recognizing AI have been made in recent years, but these things still don't "understand" anything. I put understand in quotes because I still don't think we understand understanding.

The mistake of those who have bet the farm of self drive (Musk/Tesla being chief among them) is to jump the gun and think that the AI advances of the last decade have made it just a problem of scaling a technology. It's not. It's still a fundamental invention problem, not a scaling problem, but what we have is good enough to fool us into thinking we can get there.

Musk's other big venture SpaceX can successfully build and scale rockets to send humans to the Moon and Mars. That's because that, unlike full self drive, is just a matter of scaling already thoroughly understood physics. We know exactly how to build such things. We just haven't done it yet (beyond Apollo).

Q: Is this the same article style like the one from New York Times about humanity not able to fly in a million years 9 days before Wright brothers did it? (https://bigthink.com/pessimists-archive/air-space-flight-imp...)

A: Yes, it is.

It's more like predicting humans won't be able to fly a few months after the Wright brothers did it. Exhibit: Cruise cars driving happily in SF: https://youtu.be/DNOJFcwXdec?t=61

OK they are not at "omnipresent scale" and aircraft took decades before we had proper flights with queues for security and baggage reclaim but I think tech moves faster these days.

> Ford CEO Jim Farley justified this by saying on the company’s earnings call Wednesday evening that “profitable, fully autonomous vehicles at scale are a long way off and we won’t necessarily have to create that technology for ourselves.”

i.e., “Tesla has won, we’ll license the technology from them.”

...

tesla doesn't have self-driving cars and doesn't even appear to have an actual plan to get them?

they're in fact under investigation for their existing "autopilot" system killing people: https://www.theguardian.com/technology/2022/oct/26/tesla-cri...

More that self-driving is not coming anytime soon and when it comes it will be commodity technology. So you are better off just buying it from someone else.

Yep. Nowhere near. Good chance nobody here sees it in their lifetime.

We have already reached the point where the best AI driver is better than the worst human driver, so I'm hopeful.

What does that mean exactly?

Presumably it's better than someone who is blind. That's something I guess /s

The bar to legally drive a car is very low in many parts of the world. As such there are many people who can legally drive, yet do it worse than the best AI, and on the whole people seem to be OK with that.

Yeah, I'm hoping I could use it to get to the shops in a few decades when I'm 70+ or so.

That is a good age to stop driving by yourself. With the machines you will keep your independence and won't need the help of another human to go wherever you want. It's a win for me.

The worst human driver is blackout drunk. I’m not sure that’s such a high bar to clear.

I should have clarified, the worst legal human driver, I thought that was a given. I saw old people driving way worse than any AI. Distracted people. Plain dumb people. There are a lot of varieties of really bad drivers. And the best machine is better than them.

> the worst legal human driver

Maybe they should be illegal if they are worse than this: https://youtu.be/3mnG_Gbxf_w

That happens all the time. Humans get distracted, machines sometimes fail. But machines will eventually get better, while humans will always get distracted sometimes, because it is part of ourselves.

On a straight highway in sunny california maybe, in the city, on a mountain road, anywhere with snow or heavy rain or fog, it's game over and will always be if we keep using the current techs

https://www.youtube.com/watch?v=sbSDsbDQjSU

high tech sensors can see right though through fog when human eyes can't, so I'm not sure humans really have the advantage in that case.

When Elon Musk says this, he's flat-out lying. When Waymo says it, they're talking about an extremely narrow set of circumstances[1].

Regardless, it's not true in any practical sense.

1. https://jalopnik.com/waymo-claims-its-driver-ai-is-safer-tha...

Watch this video: https://youtu.be/6A0qutTwYRA

Deep learning didn’t take off until 2012 - so a little over a decade ago.

Do you think it won’t be solved in another decade?

Will completely self driving cars be a solved problem in a decade? Absolutely not.

>The fact is that these existing services are extremely constrained in terms of geography and operating hours (though the latter is arguably a regulatory issue)

Is it though? Cruise would like you to believe that the only reason they don’t have fully driverless cars in SF during the day is strictly regulatory. However, it is far easier to drive in a dense urban environment at night when roads are empty than during the daytime when roads are crowded. I bet the disengagement rate is so high during the day (when the cars have backup drivers) that fully driverless operation would be impossible.

the self-driving myth has pumped countless startups in the past 15 years. i guess the downturn is a good opportunity to move to some different myth (i suggest longevity, which is more likely to happen).

I wish some of the investment was used to create semi-autonomous vehicles for the disabled instead.

Recent and related:

Even after $100B, self-driving cars are going nowhere - https://news.ycombinator.com/item?id=33212850 - Oct 2022 (481 comments)

Even After $100B, Self-Driving Cars Are Going Nowhere - https://news.ycombinator.com/item?id=33106739 - Oct 2022 (107 comments)

I beg to differ. Self driving cars are unlikely to make it to the West but the outlook for the East (China) is totally different.

In the West there are economic woes that will make the dream of everyone having 2.2 Teslas in the driveway not happen. There is no getting around this, the West is in a bit of a pickle.

Meanwhile, in China they have an entirely different way of achieving self driving. That is to put 5G (6,7 or 8...) on everything. Therefore, when you approach a junction, the 5G gadgets at that junction with their all seeing eyes and ability to communicate with vehicles just tells your car where to go and when.

By the magic of 5G it will also know what the deal is at the next point of decision, so your car can arrive for when the lights are green and you can be on your merry way.

The sensors on the car can do the lane keeping assist and slam on the anchors should an unexpected object appear on the highway.

There are aspects to this approach that are un-American. So it is not going to happen, even if a magic wand magically fixes economic woes that are far too structural for America to fix.

You could say that the Chinese approach is not 'self driving' but 'networked driving'. But what is the big deal if your car whisks you off from A to B without having to touch the wheel?

Do you have any pointers to this actually happening, or is this just your prediction?

FWIW, I would not start with China if I was looking for an easy place to test self-driving cars. There's plenty of crazy drivers, remarkably heterogenous vehicles on the road including suicidal cyclists and all sorts of weird three-wheeled moped-scooter-truck things, signage is inconsistent and missing, road quality is widely variable, traffic jams can be epic, the list goes on and on.

> You could say that the Chinese approach is not 'self driving' but 'networked driving'. But what is the big deal if your car whisks you off from A to B without having to touch the wheel?

Would this before or after they've locked everyone under indefinite COVID detention, err... quarantine?

Even if this were the way they justify the massive panopticon, there would be so few people that would benefit from this tech that one has to ask what was gained from doing so? The way QR codes made it's way to China isn't entirely just weibo or wechat alone, Xianjing relies on them as does all the AI facial recognition tech and is the very definition of an Authoritarian police-state and if that is what it 'takes' to get to mas adoption of Surveillance Tech than I think we are better off.

Public transportation mixed wit WFH solutions seem far more favourable.

You are right about public transport, as a cyclist I like the Dutch solution.

I don't believe you have been to China recently. There is no need to offload your assumptions about what Chinese people have for breakfast or how they put their socks on. This thread was all about self driving cars and they have a different approach to you.

> I don't believe you have been to China recently.

Given they've had a zero Covid-policy, it's safe to say that is the case for so many others as well.

> There is no need to offload your assumptions about what Chinese people have for breakfast or how they put their socks on. This thread was all about self driving cars and they have a different approach to you.

I think this is what bothers me most about those trying to sell AI solutions to anything and everything, self-driving being the most ubiquitous, is that their is almost no actual self-reflection as to how this will play out.

Rather than consider the stark implications of the reality you've laid out, of which there are many, you'd rather go in blindly proclaiming a technical solution is at hand and move headlong towards that end without realizing that for all of that to be a possibility you'd have already established a very dark dystopia, when in reality the goal should be to get as many off the road as possible in order to make any actual headway in curbing GHG emissions and thus climate change.

I study AI and ML and the truth is that FSD seems like an unattainable panacea, as in completely autonomous and non-Human guided, as it relies on so many more problems to be solved: including, but not limited to, having complete global connectivity and with 100% coverage and no outages, assuming you don't want to see a massive pileup of cars that temporarily went offline ran a stop/redlight and collided into a building.

If you haven't been to China then just say so.

Since you are a student then why don't you lift a finger and type "china self driving 5g" into your favourite search engine? You might find their approach to solving problems that you claim to be interested in to be quite educational.

These fears of yours may be something our Chinese friends might have considered. They might have even developed solutions.

Regarding topics of save the earth, climate chaos and emissions, why don't you study what their public policy is and check their track record on previous goals?

Negativity and an unwillingness to learn from other cultures is a dystopia that you are already in, judging by your comments. I promise you that you will be a happier person to put the bile aside. Admittedly it is hard to do given state policy in the UK and USA to be positively sinophobic.

I'm part Han, that side of my family is from Guangzhou; so what Culture is it exactly do you think I don't know?

My fears, I'd say well founded apprehensions, are stemmed from how they have dealt with Tibet, Xianjang, HK (proving even ethnically Han aren't exempt from their iron fist) and how they are trying to impose their will on Taiwan. The CCP is the issue, I'm sure Chinese devs would rather build those tools, or borrow it from Western IP as it is often the case, but as Jack Ma found out: you don't get to decide and the party will always have the final say.

I'm well aware of what they're doing with their AI Industry and it's mainly surveillance, specifically in computer vision (face recognition) who they sell to around the World now, and use it widely domestically.

I'll do you one better than use a search engine and show you how well documented that sector is [0] even in Academia.

I'm no fan of the Military Industrial Complex in the West, but the two aren't even on the same level of egregiousness.

0: https://journals.library.cornell.edu/index.php/ciar/article/...

>A couple caveats for those going apoplectic over the headline: I mean self-driving isn’t going to be a thing A) in our lifetimes and B) with any kind of omnipresent scale

Sure, you cannot deny the fact that it already happening. There's a surprising number of driverless cars on the road already.

In terms of meeting the 'omnipresent scale'. This is much harder. There's a ton of people whose life choices mean their car budget is <$5000. Which means they can only afford 20 year old death trap beaters.

So how does society get to the point where $150,000 driverless cars can be afforded by those people? It's from the amortization table and depreciation. It might be decades before there can be any major proliferation. That isn't the goal of driverless. The obvious caveat is most manufacturers are working on subscription models.

The interesting thing about the current system. A tremendous cost to society is the poor investment into transportation. You drive around a city and you'll see millions of $ sitting in driveways not doing anything. Whereas public transit is certainly not the alternative to this poor investment.

This is where the driverless taxis come in. The people who genuinely use their vehicles to a significant extent can still do so, but most people really just need low cost transportation. Which is indeed why public transit must exist and also why large cities are repairing that huge wealth damage.

In the near future public transit like buses will go away and be replaced by driverless taxis. Not exactly nostradamus to see that coming. This will be what presents driverless cars at an omnipresent scale. This absolutely will be happening very soon.

Teslas (while much maligned and deservedly so) can already drive pretty well on most well painted and high traffic roads. Granted it is very conservative and often I take control just to avoid other drivers honking at me for taking too long to turn.

But this article mentions L4 autonomy. L4 still allows for geofencing and human control. Tesla is already pretty much L3, borderline L4. I think it is realistic to say L5 won't exist until radical change to our infrastructure is made but to say L4 won't exist in our lifetimes... I mean, I suppose if you're 80 years old already.

I also think it draws too radical a conclusion in the article. The conclusion is from the author. The people from Ford explicitly said they think that it will be developed independently and Ford will be able to buy it instead of having to make it themselves. Which is very different than "it won't happen in our lifetime." It seems they didn't decide to stop research because it is not possible but because oters can do it cheaper.

Tesla is L2 and nowhere near L3.

Self-driving cars were always overhyped in the US because roads are the only decent transportation infrastructure.

A future where we have better cars is a waste. For long distances, we should build trains. For short distances in cities, we have buses, trams, bikes. This technology already exists and is very good, it’s just not supported by the government-business hybrid that runs the US.

> For short distances in cities, we have buses, trams, bikes

Although I agree, as an urban dweller, I am quite anticipating a possible future with the total proliferation of self driving cars. If self driving cars reach a large majority of the road, I can anticipate local and state governments outlawing the manually-driven kind on urban streets. I personally would really like this, because human drivers are an absolute nuisance to urban life in many ways.

You can't use laws to make humans drive 25 on that road where everyone routinely goes 40. It just doesn't work, and the people whose houses are on that road suffer the noise and danger that results. You can, however, use laws to make algorithms respect the speed limit, because big companies will otherwise face big fines. That's why I want to see self driving cars.

You can also make the road narrower and put obstacles on the road to force to reduce the speed: it is very effective, and very cheap to install with little technology.

Here is a typical example of narrow street with obstacles in Switzerland: https://www.google.com/maps/@46.8111357,7.1507033,3a,75y,167....

Correct, but that takes active effort on a road by road basis, which becomes a political problem on a road by road basis.

In a world where all cars are automated, establishing speed limits is presumably something you do once and auto manufacturers will be forced to comply.

Similarly, issues with cycling safety and pedestrian safety are instantly mitigated without complex study groups to examine why this or that particular intersection has seen three cycling fatalities.

All of this is assuming the creation of currently-mythical technology, or course

The US still hasn't finished industrializing/urbanizing, so we still have relatively large rural populations that drive longer distances to go to and from stores schools etc. The schools can't even cover the area they serve with bus service for children. Those areas will continue to meed cars, and they aren't a corner case, they are still roughly a fifth of the population.

And buses dont work that well in suburbia and that is over half of the population. Only about a quarter of the US lives in denser urban environments.

I think you hit the nail on the head: instead of large, sprawling suburban cities built around the car, the US will need to rebuild cities around humans who walk and take bikes.

Completing the automation of agriculture is the only thing that would drive a significant fraction of the rural fifth into cities, and rebuilding suburbia to be denser is fighting against market trends. I'm betting we get self driving cars first, and I'm not at all sure I disagree with the article's title.

Cruise and Waymo are the clear market leaders in Autonomous Vehicles. Tracking the progress of these companies in the past year alone where they are operating with no safety driver in dense San Francisco is incredibly impressive. The field of machine learning in general is advancing at a great pace as well. I think by the end of the decade in places with limited weather events AV will be as common as Uber. At some point people will reconsider buying a new car and just rely on a robotaxi subscription along with walking, cycling, and transit to get around. It will be survival of the fittest though, only few companies will succeed. Tesla 'Autopilot' on the other hand doesn't even use Lidar, seems to disconnect frequently and is much more dangerous. I really think they use deceptive marketing.

I think it's more likely that human-driven cars will become outlawed and infrastructure modified to accommodate fully-autonomous. It's not what I personally would want, as I simply enjoy driving cars and motorcycles, but it's what I think will eventually happen.

Self driving cars will happen but they will need a bit of help from their environment, just as human drivers need stop signs, light signals etc.

I can see urban streets and freeways being first to succumb to self driving. Remote, lightly traveled roads will take a long time ie decades before self driving is safe there.

The real conundrum of self driving is that if it becomes capable for 99.9% of situations people find when things go really bad, heavy snow storm, earthquake, floods, forest fires etc drivers will have lost their touch for driving just when they need it most and the self driving abandons them.

Self driving cars can react on what is happening around them (if sensor suite will catch that information) but are unable to predict what is happening around them, because they don't understand situation they are in.

They're also unable to predict what the human drivers next to them will do because the human drivers don't understand the situation they are in.

Anecdotally, the most common flaw I experience while driving (self or not) is me reacting to traffic in front while another car is only reacting to me, the car most immediately in front of them, passes (often erratically switching lanes), merges with less than a car length, only to slam on their brakes because the traffic in front of me has no openings.

Like when children are approaching the pedestrian walk, then you can predict that they are going to cross. If not, then please return your driving license.

I mean, what’s wrong with trolley cars and light rail? There’s your self driving car. I don’t think automated transportation is a technology problem, it’s more a political problem. Nobody wants to put money towards a system that is for use by the common public.

Think about this: trolleys and light rail don’t require batteries, they are more efficient, the maintenance is way lower, they reduce traffic, what’s not to like? And it would be trivial to make them self-driving. But it’s politically unthinkable.

It's time to admit this is a game: journalism plays with society by projecting dream-like nonsense on the technology industry. It's time to admit we have large institutionalized portions of society whose purpose is basically immature nonsense, such as pushing idolization of the entire landscape of fashion and celebrity.

Just use trains and bicycles. We've had the answer for 100 years.

I live where driving is a requirement. Self driving cars will not handle rural roads for a LONG time. And I really don't want something the size of a Cybertruck hurtling along the roads around where I live without someone actually doing some thinking.

I have always been skeptical of FSD because of liability. The system we have today distributes liability to the driver in almost all circumstances – 94% to 96%, according to one quick statistic. It is difficult to imagine this liability being redistributed to a handful (or even 1 or 2) auto manufacturers who remove the driver from the equation.

In other words, I am not sure that auto manufacturers can afford the liability assuming their system reaches a state in which it is 'safer' than a human driver under any condition. The cost of proving that in courts across the world may not be economically feasible.

Mercedes already assumes liability for their L3 system, let’s see when others follow

More like - let's see when Mercedes will stop selling self driving systems or renounces liability for L3 systems when their self driving car will actually crush somebody.

Yet.

Define what safety wise and engineering wise things need to be solved to give substance to that claim.

Notice we do not have self piloting planes, or train engines ( ah ah, the non manned power units always run coupled to manned train engines)

> train engines

Singapore MRT is completely driverless. Over 200km of track and 120 stations serving a city of five million people, intrigued how you wouldn't deem it fully automated?

London's Victoria Line has been driverless for decades.

It's not a hard problem. You solve it by putting linked activators along the track to control the train speed and acceleration. RMT has backup sensors for train position and radio control for emergencies.

It's plain old engineering. No AI required.

It's a completely different problem to automating a mixed use national high speed rail network which has to deal with a huge range of contingencies.

Only the new fully underground lines are driverless. The aboveground North-South and East-West lines, which are also the backbone of the network, have human drivers.

> Notice we do not have self piloting planes, or train engines

Wait, what. We absolutely have both of these things.

We don't have self-piloting planes which routinely take off, fly a route, and land.

Setting an autopilot for a route is relatively trivial. Auto-landing in heavy crosswinds or landing on water because of an emergency isn't.

Not take off, but we do have many planes which routinely fly a route and land. It depends on the airline and route, but some airlines now have changed their standard company procedure to engage autopilot as soon as you're about 400ft off the ground all the way through to autoland. Some airlines only use autoland when the landing is too difficult for the pilot due to poor visibility (though as you say, high crosswinds usually mean a manual landing). On top of that, autoland is very old technology - ILS is from the 30s and autoland not much later.

We don't have automatic taxiing or take off in any common airplanes yet. This phase involves a lot of ATC back/forth so it doesn't really make sense at the moment, but there are some changes coming to ATC and transponders which might make this more feasible in the medium term.

Docklands light railway in London is completely driverless

why is it time to admit that? it may not make anyone as rich as they hoped but I cant imagine we won't solve the problem

pretty sure the universe will ensure it happens in the the next 5 yrs just because this guy said it will never happen.

Ah yes, the scam companies that care more about raising vc money than solving the actual problem obviously would never make it happen.

Then there's comma.ai that's the actual leader in the space having production units you can buy and fit into your car right now.

Something the "experts" that seem to be expert only at reading company financial numbers seem to have missed.

The article is mostly talking about Ford.

fsd beta looking cool with output of what it sees https://m.youtube.com/watch?v=Ui9qjDgVXco

Zoox has made a good start. They're technology-intensive with multiple sensor modes integrated with some heavy CPU power.

Fortunately the system has been prepared for remotely piloted vehicles using partial AI and mostly remote humans over 5G.

Archived copy without GDPR nag screen:

https://web.archive.org/web/20221027152317/https://techcrunc...

It boils down to the question whether AGI will happen during our lifetimes or not.

> Argo AI was considered a leader with solid technological fundamentals by most experts in the field, so its shuttering is a strong signal not to be ignored.

I thought Cruise was the leader, with Waymo next.

Waymo already went public several times since 2017 or so saying true self driving would be impossible without major infrastructure changes and might be impossible in bad weather conditions with the tech they use

Cruise seems close technology wise but they're hitched to a parent company that is unlikely to do very well in the coming recession and transition to EVs. They're also spending billions with zero revenue. There's a big risk they'll face the same problem as Argo where GM will try to find a buyer but nobody is interested.

Reply on news.ycombinator.com