Even After $100B, Self-Driving Cars Are Going Nowhere
bloomberg.com
bloomberg.com
The two advantages I see is that you have a private space. And that they are point-to-point.
Those are important advantages. As, I suppose, is the fringe benefit of finding work for mathematicians.
But it seems to me you could do "self-driving" a lot cheaper and easier by improving the commons. Embedding sensors in the roads, street signs, and furniture - isn't that going to make more sense than turning ever vehicle into a super-computer?
Even after all that... buses and trams and light rail work absurdly well, at a fraction of the cost.
While we fucked around for a decade the CPC built out a high speed rail network on the mainland. Similar landmass, similar geographical challenges, less wealthy nation. Guess Americans will have to wait until public transit is more profitable than making everyone get their own car with its very own JARVIS.
You could slowly do it over decades in places primed for growth. Places like Arlington, Va are being urbanized.
I’d love automation of public transportation, but nobody wants to try because of the social hurdles.
Buses do not work absurdly well. They take three times as long to go the same distance and operate on a fixed and often infrequent schedule. They aren’t available all the time. You can’t easily transport cargo in them.
Ideally we’d redesign cities to be more walkable and bikeable so cars aren’t as necessary, but that’s not going to happen overnight.
Spend as much money on buses as the Boring Company wastes with silly unusable routes nobody wants, and you'll get great bus service.
If you had to commute regularly from, say the New Territories in to Central (about as far as you can go in HK), the bus/MTR takes 2-3 times as long as driving.
FWIW, Sai Wan Ho to the New Territories took me MTR and light rail, where I was ultimately picked up in a car by the person I was visiting. People living in the New Territories seem to find a car nearly-essential, but that seems to be because MTR doesn't run out there, and the bus service is not nearly as well-developed as Hong Kong Island or Kowloon.
Not if everyone has to spend $50-100K on a car + maintenance + insurance + fuel + parking every 10 years or so and live in cities optimized for those. Regardless how much it costs to invest in infrastructure it will cost far less than equipping everyone in America with a new car at their own expense and socializing the cost of maintaining roads, highways and interstates.
Replacing each car in America today with an electric, self-driving car (using a Tesla ballpark and factoring in the aforementioned additional expenses) is $10-20 trillion dollars. Imagine the kind of public transit we could have for twenty trillion dollars invested every 10 years. China put in 24,000 miles of high-speed rail for $500B.
Spending that $10-$20 trillion by the government would mean collecting it all at once from people via taxes, not them spending it at their own discretion. That's a big difference.
Remember, America is spending money hand over fist on the socialized roads, freeways and interstate all so people can then spend a ton of private money on a car averaging 1.5 person occupancy that spends 95% of its time sitting still not doing anything.
We should absolutely invest in public transport, but let's not pretend it isn't going to take making spending money on it a priority. It doesn't come for free and it doesn't completely eliminate the need for cars.
It also requires the entire development to shift to be much, much denser. It's not like "oh we just issue some bonds to build a rail line and then nobody ever has to buy a car again".
Self driving has the potential to provide a middle path. Vehicles with 95% up time instead of down time.
This is what Uber and Lyft claimed they could solve over the last 10 years and made zero progress.
In theory the economics would work out so you wouldn't want to own a car anymore, just use these robo-taxis, as it would be cheaper and more convenient. We'll see if it really plays out that way, and how long it takes.
Does that assume there is no traffic? I feel like the space savings from buses (in terms of space on the road) ought to mean that they would be much quicker than cars for most trips if they are given priority as they'd eliminate traffic jams, significantly increasing the road speed. This applies only to dense urban areas of course. For rural areas, self-driving cars make quite a bit of sense.
In theory you could save time in special circumstances on a bus, but I don't think I've ever done it in practice. One exception could be if you need to search for parking at your destination. A robo-taxi wouldn't have that short-coming.
Good public transit cones every 15 minutes, so on average you'd be waiting at the stop for 7.5 minutes. Most bus rides are longer than that
Trains on the other hand can be a lot better if the stations are close to where you live and where you're going, they don't have traffic and they tend to come every 5 minutes or less during peak hours.
Busses aren't faster.
In fact, as someone else mentioned, being able to utilize the existing infrastructure in new ways will likely be cheaper.
> But it seems to me you could do "self-driving" a lot cheaper and easier by improving the commons. Embedding sensors in the roads, street signs, and furniture - isn't that going to make more sense than turning ever vehicle into a super-computer?
This would require a broad agreement on a global standard for Infrastructure to Vehicle (I2V) and huge investments by local and national governments around the world. It could technically happen, but likely will proceed at a snails pace. As you mention, it would be much better than relying on-board sensors like Lidar and cameras as centimeter accurate GPS coordinates can be embedded into road signs etc, but it will take decades.
The logical thing would be to make cities less car-dependant but I'm sure there is strong economic interests that would like to continue selling cars.
There's really no advantage to self-driving cars or electric cars. They're dramatically less space-efficient, they're incredibly expensive (car + gas/electric + maintenance + insurance + parking + time spent trying to find parking), highways ruin cities and even car-first streets hurt communities. They promote unhealthy cities and lifestyles. Also, they're super duper complex in comparison.
The optimal solution is rail, followed by bikes and e-bikes then by busses.
The "optimal solution" to the housing problem would be to have everyone live in dormitories and eat in communal chow halls. For extra efficiency, you could hot-bunk the beds three shifts per day.
But, oddly, no one wants to do that.
There is more to life than "efficiency".
[1] https://twitter.com/Caltrain/status/1559318073940647936/phot...
What's silly is suggesting that "efficiency" is the only metric for quality of life.
> Hong Kong MTR and Singapore MRT.
Hong Kong has a population density of 6,801/km^2. Singapore has a population density of 7,804/km^2.
The United States has a population density of 33.6/km^2. It turns out that this matters.
The parts of the United States that have high population densities (though still not nearly as high as Hong Kong or Singapore, in general), such as the large East Coast cities, Chicagoland, etc. already have public transportation.
That's a straw man you introduced. Cities with fewer cars are more efficient and have a higher quality of life.
That comparison re: density is meaningless, and that number doesn't matter at all. Nobody's going to put Hong Kong type subways in the middle of the Nevada desert where that average gets pulled down. Only New York and Chicago really have proper transit and that's only downtown. There's no reason for car ownership to be higher in SF or NY or DC or Chicago than in Hong Kong except inadequacy of transit.
You'll notice that all the most liveable cities in the world have excellent transit and none of them are in America [1]
[1] https://www.visualcapitalist.com/ranked-the-most-and-least-l...
Sure thing. That's why everyone who can possibly manage it moves to the suburbs when they have kids. Because they're looking for "lower quality of life" and a less "livable" environment for their kids.
Note that Europeans do this too.
P.S. I can assure you that Calgary, #3 on this list, has plenty of cars.
> There's no reason for car ownership to be higher in SF or NY or DC or Chicago than in Hong Kong except inadequacy of transit.
Except that people enjoy having a personal vehicle that goes where they want, when they want, and also enjoy not dodging junkies and drunks pissing in the subway cars and shaking them down for money.
- On the one hand, "self-driving cars" is a general problem of transporting anything from point A to point B without direct human intervention in a very complex space, we'll probably see a lot of things getting transported with cheap point to point solutions post the tech being widely available with some people aggregating pickups and drop offs for efficiency a la trains etc.
- On the other hand, I expect a wildly high spike in traffic post these vehicles being more accessible, currently the largest limiting factor to a car being on the road is someone driving it, what happens exactly when practically everyone who can afford a car now can just send them off to do their daily todo list stuff like picking up groceries. There was a fun study[0] that showed car usage increased[0] on average.
I suspect peoples opinions may shift as roads become more train-like with ever growing slow moving traffic of either empty cars doing tasks or people reading as they slow travel to their destination.
Trains / buses on bus only routes / lanes restricted from being self-driving / premium lanes may be the only fast ways of getting around.
- [0]: http://www.joanwalker.com/uploads/3/6/9/5/3695513/mustapha_e...
- [1]: "while total VMT decreased slightly during the chauffeur week for the first subject, and hardly changed (on average) for the second subject, the remaining 11 subjects increased their auto travel. The increases in total VMT during the chauffeur week ranged from a low of 4% for one of the Millennials (from 532 to 554 miles) to a high of 341% for one of the Retirees (from 117 to 516 miles), with an overall increase of 83% for the entire sample (from 3,344 to 6,118 miles)." where VMT is "vehicle-miles traveled".
"Absurdly" is far from true. They work barely at all if your time is valuable to you.
I can take a train into a nearby CBD - avg 15m wait time for bus (to train station), 10m bus transit time, avg 7m wait at train station, similar waiting times at destination, all multiplied by 2 because I return at some point.
It's usually between 45m and 60m.
In about 15m I can drive there. I'd rather spend the extra time with my kid.
Self-driving cars are neither better or worse. They do not exist in reality, only in our imagination.
Such things are not bound by mundane constraints. They can be orders of magnitude safer, they can solve congestion, be fast cheaper, not use parking, and even somehow solve climate change.
I think this is the way to go. Could maybe improve safety which was is holding autonomous back.
There would still be challenges: how to handle safety issues, crashes, unruly passengers, etc. without a driver, but worst case you're paying for an attendant now instead of skilled labor.
Skilled labor usually refers to individuals who work in jobs that require experience, and education, like college degrees and advanced degrees, and who are highly trained professionals in a specific field.
This is in contrast to driving a bus, where anyone can walk in off the street, go through 1 to 2 weeks of training, and then pass a certification or licensing test
Bus drivers are unskilled labor.
They only make decent money since its a pubic union job. Good for them. But if the government contracted it out at the lowest bidder, they wouldn't be making that much.
Much like how the NYC subway now has different gauge tracks after all the competing companies were merged. Or the multiple cell networks we have now.
In Canada, as with a lot of the US, the roads are asphalt. These tend to get resurfaced once a decade.
All we need is a cheap rfid type device, with unique serial numbers, and they can just be dumped into the asphalt, at least a few per sq cm.
Just mixed in, no specific placement, as long as the density requirement is met.
This allows for redundancy too.
After the road is done, a mapping device can drive along slowly, and map sensor location to road location, and create a database.
This may seem like a lot of data, but you only need this data for the area you are in, and if we started today, it would be a few years before roads were created this way.
Memory and storage grow constantly.
My biggest concern is ists of some type. Anti-technologists, some radical *ist group, dragging some high em device behind them, to blow all the rfid devices.
But anyhow, road sensors aren't that hard, or even expensive. Not compared to standard resurfacing costs, and that is done frequently enough that natural repair cycles can get things embedded.
There is almost 4 million miles (6.5 million kilometers) of roads in the United States, with just over 45,000 miles [73,000 kilometers (km)] composing the Interstate Highway System (fig. 2) (Federal Highway Administration, 2004). The total distance of U.S. roads combined could circle the Earth at the equator 160 times; the interstate highways, almost two times.
The analysis shows that a total of approximately 1.5 billion metric tons (Gt) of aggregates, 35 million metric tons (Mt) of asphalt, 48 Mt of cement, and 6 Mt of steel is in place in inter- state highways. These estimates illustrate the large quantities of these materials used in the Interstate Highway System.
This doesn't translate exactly to how much surface is covered, but I know locally all the highways are concrete. Also, placing sensors in 4M miles of roads is no small task.
Asphalt is cheaper, and replaced more often, so degradation due to frost greaves and such is not as fiscally damaging.
But my entire point is, by placing sensors during normal resurfacing, it literally is a small task.
Electric and self-driving cars are not the solution.
The word needs trains, trams and [e]bikes.
They seem to be doing alright for a technology which is supposedly "going nowhere". (Yes I know these cars sometimes stop, but that's better than having them crash).
To counter the expensive point(s) that LIDAR critics have mentioned you then have Tesla's Fools Self Driving (FSD) demoware, which still doesn't work and even if it does it works worse than LIDAR, hence why Elon's 2020 robo-taxi prediction was completely off especially even if you gave him more time.
But at least Lidar works better here than vision, only for this robo-taxi use-case here and not for selling such cars with expensive LIDAR systems to individuals at scale.
I'm pretty convinced if we wanted to make 100 million lidar units in 2023 or 2024, the cost would be approximately* the same as it is today.
* From a computer science big-O perspective. Simple volume economics might drive the cost in half, or even a quarter, but not down by an order-of-magnitude.
The Audi A8 used Valeo's (with Ibeo) first generation low resolution LiDAR Scala 1 from the automotive supplier Valeo. Mercedes new models will be using Valeo's second (or third) generation LiDAR. All these are used for L2/L3 assistants. Valeo is a traditional large automotive supplier.
Luminar, a public company from the US, cooperates with Volvo. Some models will come with a LiDAR in the base configuration. These are "new LiDARs" with high resolution.
Innoviz, a 'startup' from Isreal, will supplies LiDARs to VW. Its angular resolution is (in its focus area) about 0.1 (or 0.2) degrees, which is sufficient for higher levels of autonomy and surpasses/equals the resolution of the expensive Velodyne sensors of the past. They will probably be in the same price range. Due to the limited FOV due to the technology, you will need multiply LiDARs.
Many new models from Chinese car brands will also ve equipped with a LiDAR. Most of them with Chinese LiDAR manufacturers like RoboSense or Hesai. Some are equipped by European manufactures like Ibeo/ZF. For example, there is the automotive sensor AT128 by Hesai. It targets normals vehicles (see price range above) and claims a similar performance (except for FOV, so you need multiple) like the Velodyne Ultra Puck (~$50000).
So costs of LiDARs are a not the very expensive obstacle they were in the past. The only problem could be that the new LiDAR manufactures cannot scale up series production. For example, Ibeo just filed for insolvency because they could not close another round after aggressively increasing spending in the past years.
https://velodynelidar.com/media-coverage/ieee-spectrum-highl...
Tesla taking longer than Elon thought doesn’t mean they are still not miles ahead… who else is even close? (On a rainy day)
It kinda depends where and when it stops. If it panics in the middle of an intersection with an incoming car, you get this:
https://www.cnbc.com/2022/09/01/gms-cruise-recalls-and-updat...
These things are not robust or reliable. These are publicity stunts.
The whole article reads like free publicity for him and his new company, and is light on data backing up the claims (beyond usual anecdotes).
He never spent a day in jail. According to the article:
> Levandowski pleaded guilty to one and was sentenced to 18 months in federal prison in August 2020. Because of the pandemic, the sentence was delayed long enough that he never served a day before his pardon, which came on the last day of the Trump presidency.
And let's be honest, if you want to see videos of cars doing funny or stupid things, the surest bet will always be the idiotsincars subreddit for human drivers. Cherrypicking incidents is easy.
As I'm getting older, I think I'll soon fall into the category of "human drivers with kinks to work out" too.
I'm pretty sure my city would, on the net, be safer with self-driving cars. I think 90% of human drivers here will be better drivers than a self-driving car, but and it's not even the next 9% which scare me, but that 1% of least-safe human drivers.
From a purely technical and apolitical perspective, if we could magically replace every car here with self-driving, I'd do it in a heart beat.
I won't pretend, though, that technical perspectives are the most important. The financial, political, legal / liability, and similar organizational and human issues are important to get right. I'd like to see that before mass adoption. Today, I wouldn't buy a Tesla since I want to _own_ my car. If there is a new piece of shared infrastructure, I'd like to make sure it's run _in the public interest_. I don't see subscription-based, car-rental, cloud-based, algorithmic, freemium, network-effect tech-style models as something I'd want in my city.
If you've been following the public announcements in self-driving for the last decade closely then you don't assign them much value, particularly with respects to timelines.
I follow the public announcements very closely, and have seen your sentiment about "ready in five years" expressed since at least 2014.
According to the report sited in the article [1] $100B is nearly a third of the total research into automation, connectivity, electrification, and smart mobility (ACES). Is this sustainable?
[1] https://www.mckinsey.com/industries/automotive-and-assembly/...
It’s interesting to see how much people get wrong about autonomous vehicles (AV), mostly because of how little these companies can share publicly. IMO, geofenced driverless vehicles are closer than you think and AV are getting more affordable to manufacture as we progress thanks to economies of scale. Can it replace all cars immediately in the next decade? No. Can you use it like a Uber in SF after a long eventful night within this decade? Probably.
People get it wrong because its the marketing departments who are doing the sharing in these companies.
1) So hard, even with incremental progress, we can't predict the problem will be solved even in a decade or two, e.g. Nuclear fusion, General AI (AGI).
2) Hard, but the progress over decades has been notable, implying convergence in a decade or two. Voice recognition and Natural language processing seem to fit this category.
I'd put self-driving in category 2, but I can see why hubris around projects like nuclear fusion and AGI might make you think this is of the same type.
This is why there is a lot of companies that did (2), but we will not have self driving car in the next 10 years.
I have lane assist and drive assist in my new car and I really do love it for highway driving, it's pretty suttle overall but very useful.
Feels like a solid backup and less of self driving but I am enjoying it as a user.
It will follow the bends hands free and keep you centered but doesn't intrude the way lane assist does by itself.
Self-driving is being extensively used all over north america. I'm in the detroit area, so I've seen them. Haven't had the chance to be driving near one yet :(
Kind of impossible to say 'going nowhere' when there's popular youtubers with videos showing it working in real life.
The thing that's happening right now is the big shift in transportation. It's better for the wealth of our countries to have driving services at a monthly cost rather than some large cap-ex purchase that mostly sits in a driveway taking up space.
The problem? Boy do we have a ton of lazy people working for insurance companies who are about to lose their job.
For example when real new markings contradict old markings? E.g. just recently had to do a left turn on a recently re-opened road, which required me to cross several solid lines, because previously there was no turn there. Or in another example when the road has double sets of markings - regular, and temporary, for the maintenance period. What happens then?
In the cars you can't do that in, there's teleoperation as a backup; the car stops until someone can manually resolve. https://www.motortrend.com/news/mira-self-driving-car-remote...
( Leaving a note on HN just to say "I told you so" ;-)
>In the view of Levandowski and many of the brightest minds in AI, the underlying technology isn’t just a few years’ worth of refinements away from a resolution. Autonomous driving, they say, needs a fundamental breakthrough that allows computers to quickly use humanlike intuition rather than learning solely by rote. That is to say, Google engineers might spend the rest of their lives puttering around San Francisco and Phoenix without showing that their technology is safer than driving the old-fashioned way.
The view that something fundamental is missing has been prevalent amongst Google's best self-driving engineers since sometime in 2015. Howver, at that time they still entertained optimism that the AI field was moving fast enough that the fundamental missing thing could be discovered.
When ex-google self-driving project director Chris Urmson, Drew Bagnell and Sterling Anderson founded Aurora in 2017, they touted the mantra that "you can't build a ladder to the moon"
Meaning, that, while self-driving industry can make continual incremental step-by-step progress with the technology they have, they will not within a reasonable time frame achieve their objective without a fundamental change.
Also Tesla is profitably improving at this point. Even though I don't have a Tesla, the fact that they are able to increase software margins means that they are noticably improving step by step.