BYD says self-driving tech is more valuable in factories than cars
cnbc.com
cnbc.com
So I was being polite leaving that conclusion for the reader. But I can see I wasn’t clear enough.
Given we cycle cars much more often than we cycle roads (globally) makes sense to update the technology on the car.
https://www.nytimes.com/2001/07/26/technology/what-s-next-it...
But, I doubt they'd be much help for passenger cars on diverse urban per-trip routes:
* roads are constantly wearing & being trenched or resurfaced; the lines would often be broken or need refreshing, necessitating an effective 'lineless' mode
* cars often need to improvise (including briefly into oncoming lanes) to get around temporary blockages, and need to pull away from the major maintained lanes – again requiring confident off-line operation
* after car sensors/decisionmaking can handle all the other challenges & collisions/traps to avoid, merely following lanes (when needed) is not hard at all: guidelines help with an easy, essentially-solved problem but not the actual potential deployment blockers
Full-unattended self-driving cars are working pretty well in SF right now - on a nighttime walk you'll see many, which are behaviorally only a little more cautious/tentative than human drivers. So, it's only the branding & empty driver's seat that distinguishes them.
There are roads out here that don't even have lines, because they go so long between maintenance cycles that if they did paint them, they'd fade fast enough that they'd still spend more time without lines than with.
And how do you handle complicated intersections? Ones where there's 6 different roads all meeting in one place, with three separate pedestrian crosswalks, and traffic lights that handle directing the traffic to various different combinations of possible outflow roads?
Are most trains self-piloting?
[1] https://en.wikipedia.org/wiki/List_of_automated_train_system...
If 100 people are on a train, 1% of them are driving.
My point here is that even trains with drivers can be considered close to driver less vs people driving cars.
Cars are really only 'door to door' in the suburbs when commuting between two homes with driveways. Otherwise they're no different than a decent transit system.
I do find it interesting how people automatically assume all driving is door-to-door, forget about parking lots, forget about finding street parking, etc. and immediately talk about how transit can't work based on the same issue.
American suburbs are incompatible with trains. In cities, Uber is door to door.
I get the broader point. But at this crossroad, re-designing our cities for trains is a moot point. We could have, we should have, but we didn’t. Similar to African and Southeast Asian countries leapfrogging copper for mobile and fibre, electric self-driving point-to-point (perhaps with sub regional rail and more-efficient air) looks like the clear future for America, a rich and spread-out country.
Absolutely, in part because they don't have trains. You have to put the trains in first, and the density will follow. However, I actually picked my wording pretty carefully in the original post. Most people live in cities, and the fraction that lives in urban areas continues to grow. I don't think trains solve all problems but I reiterate they solve the problems of most people.
That's before we even get to the problem of Uber and Lyft being utterly unsustainable businesses.
> I get the broader point. But at this crossroad, re-designing our cities for trains is a moot point. We could have, we should have, but we didn’t.
It's not nearly as bad as you think. You drop the transit down, and the buildings around will redevelop absent punitive zoning. No better time to start than today!
Given a choice, many are choosing point-to-point. Even where there is mass transit, American cities aren’t as dense as European and Asian centres. (Particularly post Covid.) The solution will be hybrid. Planners who pivot hard for trains make the same mistake as those who went hard for cars in the 50s.
> before we even get to the problem of Uber and Lyft being utterly unsustainable businesses
In suburbs, yes. Cities have had cabs for ages.
> drop the transit down, and the buildings around will redevelop absent punitive zoning. No better time to start than today
Now isn’t good. We’re on the precipice of a regime change. Self-driving cars port economies of scale from Phoenix to New York and vice versa. That isn’t true of trains given the amount of local planning required*. We should deploy artery routes. But the end game will be point to point, aided with trains for efficiency. There are many ways that could develop; it seems silly to bet on the legacy model this late in the game.
* Yes, Wayne must tune. But that team can move on once done. We don’t ship Chicago’s urban planners to Seattle; there is a strong element of re-learning everything every time with trains.
Cars are an awful model because they spend 95% of their time idle. They do that because most people only want to commute to work and from work. Which means you need to have enough capacity on the road for 1.5-pax average vehicles for everyone. That's why adding one more lane never solves anything. They're fundamentally a choice which does not scale no matter how many more lanes you add. No matter how much automation you add. And if you choose to play this game you just induce more demand.
They're also a regressive tax on the poor since you're shouldering the average person with hundreds to thousands of dollars in car payments, maintenance, registration, licensing, insurance, etc. Plus the socialized cost of 'free parking'.
Cars are a bad model, period.
> In suburbs, yes. Cities have had cabs for ages.
Well sure, but that's because they pool their leases and insurance, which isn't something you can do in a gig economy. Which means uber and Lyft have fundamentally worse unit economics than cabs. So you should assume you'll in the fullness of time pay at least as much as cabs cost.
> Now isn’t good.
Now is the best time since yesterday.
Private cars, sure. I doubt a New York taxi gets that down time.
> uber and Lyft have fundamentally worse unit economics than cabs. So you should assume you'll in the fullness of time pay at least as much as cabs cost
But higher utilization. Uber is profitable in New York. I think Uber is actually more expensive than a yellow, now, because the TLC hasn't been great about inflation adjusting rates. But it's more convenient, more reliable, and so, again, utilized more. (Wayne was like $5 per ride in Phoenix.)
> Now is the best time since yesterday
You're purposely ignoring the point. If you're getting all the requirements tomorrow, tonight isn't a smart time to build.
> Of all people who commute to work in New York City, 39% use the subway, 23% drive alone, 11% take the bus, 9% walk to work, 7% travel by commuter rail, 4% carpool, 1.6% use a taxi, 1.1% ride their bicycle to work, and 0.4% travel by ferry.
> Uber is profitable in New York.
And if that's all it took to be Uber they'd be in great shape but again, most people in New York take the metro. And Uber's business model relies on people taking Uber outside of Manhattan.
> You're purposely ignoring the point. If you're getting all the requirements tomorrow, tonight isn't a smart time to build.
This doesn't make sense. I'm telling you if you build rail density will come. Like it always has. So if you want a chicken, you better get started on the egg.
I assume this is your source [1]. It's from 2017, and only measures commutes. As of February, total subway ridership was down a quarter to a third; busses have stabilized at -35% [2]. Notable absentee: Manhattan, the densest of the lot, where we're stubbornly below -50%.
I'm not arguing against trains. Just against deploying new trains at the precipice of a game changer.
> if you build rail density will come. Like it always has
You're quoting sources from before remote work. (I'm ignoring the increasing automation, particularly in Manhattan.) And, again, before we know how self-driving cars will change transportation.
[1] https://archive.today/20200213125620/https://factfinder.cens...
[2] https://comptroller.nyc.gov/reports/riders-return/#:~:text=M....
Why not? Cities were redesigned from 1940s-60s to be compatible with cars. It took an enormous amount of capital, but it was done because of the promise of a new technology.
Most of the infrastructure in suburbs currently under construction will be tear-downs in 30 years. The only redesign that needs to happen is letting current developments age out, removing restrictions on denser and multi-use architecture closer to the city center, and pricing utilities by effective utilization (suburbs use more utilities but don't pay more for them). Denser architecture and urbanization will naturally re-emerge because it is more economically competitive.
Mass transit can then be added in piecemeal, first with busses, then light rail and street cars, then underground trains.
I'm not arguing against redesign per se. Just the timing. We don't know what self-driving cars will mean for urban transport, but we know it's going to be impactful. Building out rail infrastructure now is like building the best piston engine on the eve of the dawn of the jet age. It's probably still very relevant. But we don't know in what form.
I doubt it will be mass transit a la Europe. But I also don't think we'll be in an LA universe. Cars that can seamlessly deliver people to train stations is an option. But at that point, why not directly to their train car? These seem like simple modifications, but they impact whether you build out lines and arteries with waypoints in the suburbs.
At this point, given the challenges that we can see in front of us, we do not know that self-driving cars will ever be genuinely practical. Not in the "can drive anywhere, under any conditions, with no more danger than humans" sense.
That’s fine. I’m limiting scope to urban and suburban environments where they interface with other modes as a universal last mile. That definitely looks proximate, given there are working examples already deployed, albeit unsustainably.
Because my understanding is that current self-driving technology is also nowhere near being able to handle inclement weather, especially snow.
In a sense. That also applies to planes and cruise ships. All remain labor constrained.
That’s the vision of self-driving cars. Door to door and scale free. Trains approximate that for the masses, most of the time. But they don’t for the long tail.
Not when the train drivers go on strike. In the uk we know all about driverless trains...
People used to own 'vehicles' that were capable of automatic obstacle avoidance and fully autonomous navigation to predefined locations. These vehicles were even capable of refueling themselves, self-repairing minor damage, and even self-assembling new units!!
Occasionally the they would glitch and cause unexpected behaviors, and given the technology of the time, the waste products from the 'fuel cell' would cause problems, but they worked pretty well...
Eventually they were replaced by the non-intelligent automobiles, but they are still the preferred method of ground transport in rocky/uneven terrain, and can occasionally be seen in urban environments during parades, or being used for crowd control.
We called them horses ;)
In fact the etymology of the words "driver" and "driving" may surprise you.
> The origin of the term driver, as recorded from the 15th century, refers to the occupation of driving working animals, especially pack horses or draft horses. The verb ' to drive ' in origin means "to force to move, to impel by physical force".
It's literally the job you're implying doesn't exist, heh.
Further, the next iteration of transportation after the horse was actually the electric car in the early 1830s, long before the gas car (1886). The arc of history is surprisingly bendy.
Kind of. But it doesn't get more self driving than this:
https://www.foxnews.com/us/ohio-deputy-drunk-amish-guy-buggy
Ohio deputy encounters 'drunk Amish guy' slumped over in moving buggy: 'The horse knows how to get home'
Also, Didn't Tesla Autopilot originate from factory automation/self QAing car in factory? (Maybe I misread that, and I can't find reference now).
This feels like a very big "We are giving up on that" call from BYD.
I can’t imagine this is going to age well.
But boy it’s taken a lot of time and money to get from “mostly there” to “a bit more mostly there”.
The 1st "DARPA Grand Challenge" was in 2004 – when not a single entrant made it further than 8 miles on a 150 mile course of some early stationary obstacles then some straightforward highway driving.
Now, just 19 years later, unattended driverless cars are offering reliable commercial service in a real, challenging urban environment. Of course, at times hyperoptimists & promoters had expected this milestone ~5ish years ago. But that's still pretty quick progress!
They don't have to work everywhere to be incredibly useful. I'm pretty sure they'll have problems on things like snowy roads, & exceptionally poor roads or extreme-outlier traffic tie-ups, for perhaps another decade. But for many uses, they're already "here".
How much is it going to cost to make them work in London? Tokyo? Beijing? Sydney? Delhi?
* 24/7 in San Francisco, over most of the city for their test riders, over part of the city at night for paid customers. (I see their cars all over at night, and occasionally during the day, through every kind of street & congestion.)
* evenings in Austin, downtown/campus central area
* evenings in Phoenix, central area
They've given hints Dallas is next. The maps & hours declared on their website have sometimes lagged actual availability in their app.
GM doesn't quite have "limitless" financial backing – it doesn't have its mid-20thC dominance any more! GM was bankrupt as recently as 2009, and has been lagging other companies in some important ways. But yes: it's an important project for GM, which strikes me as a reason to expect rapid progress in tech and rollout will continue, rather than a reason to expect a slowdown.
Though SF is where they have the largest fleet, & most testing, bringing on the initial Austin zone/service supposedly took just 90 days from project-start to 1st rides. I used to live in Austin, & was there (& noticed Cruise's cars operating there) a few weeks ago. The central zone they're covering is one of the more congested & idiosyncratic areas to drive – as opposed to newer/wider/calmer roads further out.
I've driven a bit in Sydney; don't recall it being any harder than SF. I've walked & otherwise travelled lots of the busiest parts of London, Tokyo, & Beijing. There's more of everything, including megawide avenues, giant intersections, & narrow old alleys – but nothing totally different from what's already been mastered among SF's hills, alleys, illogical intersections, dedicated streetcar/bus lanes, unruly traffic, assertive pedestrians & cyclists, and erratic jaywalkers/skaters/derelicts.
Here's an illustrative video of a self-driving Cruise navigating San Francisco's North Beach – a difficult area any time or day – on a busy St Patrick's Day night with streets filled with cars & revelers: https://twitter.com/kvogt/status/1641123102858919953
Can't speak to Delhi, or Kiev, or Moscow, or Havana, or Port-au-Prince. Everyplace has its difficulties. But autonomous cars are working well, right now, in tough places, as the whole stack of sensors, logic, data, & computation keeps getting faster, better, & cheaper. Remaining political & social challenges are likely bigger than technical blocks.
And claims like, "in a fully self-driving scenario it would be difficult to pinpoint who is at fault" are tells this spokesperson is just making-spin-up. The massive logs of sensor data inherently collected by all self-driving tech ensure every accident generates evidence of preceding events, & each car's good or bad decisions, far beyond what any human-driver-only accident offers.
BYD was licensing NVidia's self-driving car technology. So apparently that doesn't work.
Tesla will probably double down on high margin features, like self driving, as I'm not sure it will be able to compete in price with BYD.
Driving data is culture specific. I have rarely, if ever seen this point hammered home.
Simply put, a vehicle trained on US driving data will not manage in many developing nations.
You do not control the variables on the road, driving data is by default contingent.
The only way to make level 5 to work is if data from road side sensors or a mesh constantly updates driving models.
For me this veers wayyy too close to central control.
Factories are perfect on the other hand, because you DO control the variables.
you dont have to worry about kids on e-scooters or potholes or weather problems or drunk drivers etc etc