I’ve been hearing the hype for the past 10 years, but the dream of making human drivers unnecessary doesn’t appear to be that much closer.
For the past 70 years, commercial nuclear fusion reactors have always been at least 20 to 30 years away. Are fully self-driving vehicles in the same category? Will they always be 5 to 10 years away for the rest of my life? (I’m part of the Goonies and Saved By The Bell Generation.)
Doppler,
I can see that, self driving cars would never make human drivers unnecessary.
There are cars now that can drive themselves. That is done. Yes there is an accident rate. The problem is our legal system. There is no legal framework to allow for the accidents. Which insurer? Which company or owner is at fault? How would either be at fault in an accident?
There is also no way to resolve that. Meaning the car or car company is often to blame. So can you buy a car and over the years as their are crashes the car company pays and pays and pays for the accidents in the fleet? Economically that is impossible. But if you own a car and you have a defective product how are you to blame? If you do not use the driverless feature why buy the car?
The problem is our legal system.
No, I don’t think that’s it. At least not all of it. Self driving cars are still not quite good enough for mass adoption. For some reason the industry picked the hardest target, with the smallest consumer demand, to go after first.
Caterpillar has self-driving 80-ton dump trucks. But those are in areas of mining operations. We have self-driving agricultural equipment. Self driving factory bots. I saw an autonomous robot in the Atlanta airport a few months back (I forget exactly what it did, but it was in the food court).
But self driving cars, in city conditions (not interstate highway)? We’re not quite there yet. Technology, legal or insurance.
I’ve been hearing the hype for the past 10 years, but the dream of making human drivers unnecessary doesn’t appear to be that much closer.
I first heard of voice recognition in 1957, the Shoebox Project, an MIT graduate thesis. When did voice recognition really enter the market? 20 years later? 30 years? Some problems are really hard which evolution took millions of years to figure out.
When will self driving be good enough? When self driving kills fewer people that human drivers? When self driving never, ever, ever, kills anyone?
The Captain
aviation kills people and it is not yet illegal
aviation kills people and it is not yet illegal
Heh, and we still pay human pilots six figures to move cargo instead of using AI. It would seem that automating planes should be easier than cars but that has not happened yet either.
When will self driving be good enough? When self driving kills fewer people that human drivers? When self driving never, ever, ever, kills anyone?
It will probably be permitted in any context where, at the time the self-driving is activated or in the circumstances where the vehicle is operating, the self-driving vehicle is at least as safe as a sober adult driver would be operating that same vehicle under those conditions. In other words, when I (as a theoretical average adult driver) push the button to engage the AV driver, am I increasing the safety of the vehicle.
We’re not anywhere near demonstrating that yet. Tesla’s got lots of data that shows that Autopilot results in fewer crashes than average. But that’s not the same question. Nearly all Tesla’s with autopilot are relatively new (meaning advanced safety devices and fewer mechanical failures that have no bearing on the skill of the driver), they’re very expensive (less likely to be driven by teenagers, who disproportionately cause crashes), and they driver gets to choose when to engage Autopilot (making it likely that it’s disproportionately engaged when driving conditions are suitable for Autopilot, like uncongested highway driving in good weather).
The pilot programs that are being run in a few cities with geofenced AV’s might start to compile data that could demonstrate that.
When will self driving be good enough? When self driving kills fewer people that human drivers? When self driving never, ever, ever, kills anyone?
It will never be 100% safe. That is not only impossible and unreasonable to expect, it is not the stated goal. The stated goal of FSD has always been “much safer than human drivers”.
But I’m still trying to figure out how many human drivers really want FSD in their car. Sure, a small number are thrilled for it. But the masses? I don’t see the demand. But I see a ton of push-back.
But I’m still trying to figure out how many human drivers really want FSD in their car.
Most people didn’t know they wanted a computer in the living room so they could get the weather forecast at the touch of a button. “I watch channel 6 at night, that’s good enough.” Most people didn’t know they wanted a microwave so they could reheat food badly. Most people…
When people realize that the thousands of hours they spend watching the road could be spent watching a tv show or reading a book or playing with the kids in the backseat they’ll go “wow, this is great”. Push a button, get dropped off at work.
Even now that’s a pipe dream because the so-called “self-driving” requires you to keep your hands on the wheel and pay attention to the road, so it’s barely a step forward.
For the past 70 years, commercial nuclear fusion reactors have always been at least 20 to 30 years away. Are fully self-driving vehicles in the same category? Will they always be 5 to 10 years away for the rest of my life?
According to Elon Musk, over the past six years fully autonomous Tesla vehicles have gone from two years in the future to less than one year! So there’s been some progress, right?
The Model S I got in 2014 was one of the very first built with Tesla’s original autopilot capability. It relied on radar and a MobilEye camera system. It went from doing almost nothing when I got it to being somewhat capable by the time I sold it in 2018. But full autonomy was never in the cards for that vehicle.
At the end of 2017 I bought two cars with Tesla’s “Full Self Driving”, a Model 3 and a Model S. They used radar and Tesla’s camera system. I still have them, and they’ve been regularly updated with Tesla’s latest efforts. Both are in the FSD Beta program, so they’re trying as hard as they can. But while they’re much better than they used to be, and they regularly get me around suburbia without requiring much input from me other than my destination, I don’t think they’re particularly close to being able to let me read a book rather than be in charge. One interesting advance is that they no longer use their radar, just cameras.
So while I keep expecting a step change in capability, I haven’t seen it yet. Each update is two steps forward and one step back (not the step change I want to see).
After 4.7 years I expected better. Now, I’m guessing at least another two years, and there will be some compromises. But it’s been an exciting journey to experience!
-IGU-
When people realize that the thousands of hours they spend watching the road could be spent watching a tv show or reading a book or playing with the kids in the backseat they’ll go “wow, this is great”. Push a button, get dropped off at work.
I think the people most affected by real autonomous driving will be those unable to drive: those who can’t get a license; kids; people on drugs of various sorts. All of a sudden they will be mobile.
-IGU-
…those who can’t get a license; kids; people on drugs of various sorts.…
Drunks to blitzed to drive. Rehab psychiatrist will be out of work.
Much fewer DUI’s. AA meetings will be near empty.
When will self driving be good enough? When self driving kills fewer people that human drivers? When self driving never, ever, ever, kills anyone?
“Killing people” isn’t the main thing. It’s a big thing of course, but it isn’t the thing that makes “full” self driving possible. The MAIN thing that makes full self driving possible is a system that never disengages, never, not in any scenario, familiar or unfamiliar, new or old. And not a single commercially sold car on the road today will ever have that capability. That means that none of the models currently on the road will ever be full self driving, including my Tesla.
Full self driving = I send my kids to school in the car in the morning, and the car will drop them off at school, then drive back home to me so I can use it during the day, and then do the same thing in reverse in the afternoon/evening.
And not a single commercially sold car on the road today will ever have that capability. That means that none of the models currently on the road will ever be full self driving, including my Tesla.
Can you substantiate this assertion? The reason I ask is because self driving cars are made up of two systems, the vehicle itself and the AI hardware and software that replaces the driver. The AI hardware and software are not that hard to update. What is the insurmountable problem?
The Captain
Can you substantiate this assertion? The reason I ask is because self driving cars are made up of two systems, the vehicle itself and the AI hardware and software that replaces the driver. The AI hardware and software are not that hard to update. What is the insurmountable problem?
I’m not Mark, but I share his skepticism that any cars currently on the road will ever be fully self-driving, so I’ll give my answer to that question.
The insurmountable problem is time.
It’s increasingly looking like “solving” the AI problem for Level 5 autonomy is going to be much, much, MUCH harder than recently hoped. IMHO, there’s an excellent chance that there won’t be regulatory approval for turning on Level 5 autonomy in privately owned passenger cars for at least a decade, and a decent chance that it might be 20 years or more before we reach that point.
That’s outside the useful life of a vehicle on the road today. Sure, there will be outliers - you can still see a 1990’s or earlier car being driven out there from time to time - but it’s highly unlikely that anyone who wants Level 5 autonomy is going to pay to have it installed in an old beater.
Albaby
The insurmountable problem is time.
Albaby, your reply is internally consistent.
but it’s highly unlikely that anyone who wants Level 5 autonomy is going to pay to have it installed in an old beater.
Why are people willing to pay $15,000 for Tesla’s FSD? It could be just the enthusiasts. As with all developing technologies one should focus on market adoption/penetration, crossing the chasm, and the “S” curve. Musk is saying that the FSD beta will be available to a much wider audience by the end of the year (I know, beware Elon Time).
The Captain
Why are people willing to pay $15,000 for Tesla’s FSD?
I have no idea. Some of it is Tesla enthusiasm, no doubt. Some of it is just some really weird, strange, inexplicable market weirdness - because people aren’t willing to pay anything close to that when buying FSD on the used car market, but some people are still willing to pay $12K to have TSLA turn it on for a used car:
Analysis of used party Tesla sales shows results are all over the map. Some buyers will pay more for FSD cars, and it is reported that they sell for an amount that’s about 25% of the price to add FSD to either a new or used car — under $3,000. It is not unusual to see Teslas on the used market selling for a very similar price with or without FSD. This has led the various companies which provide estimated values for used cars to add modest value for the option. A query of the Kelly Blue Book on a 2020 Tesla Model 3 SR+ with 20K miles shows a difference of $4,300 for the $12,000 option. This is striking when some people are buying used Teslas without the package and then paying Tesla the full $12,000 (soon $15,000) to enable it, yet others are getting lucky and finding a car which already has it for a very modest premium.
https://www.forbes.com/sites/bradtempleton/2022/08/31/tesla-…
But FSD isn’t Level 5 autonomy. It’s Level 2. It’s not clear that they’re getting any closer to even being able to offer Level 3 autonomy using their FSD framework, even as other brands are rolling out Level 3 (and pilot-testing Level 4) systems:
But with this price hike, is it any closer to Level 3 systems currently being rolled out by other automakers? A few days ago Tesla CEO Elon Musk said it was his goal to see FSD in wide release by the end of 2022, once again promising Tesla vehicles would be capable of "self-driving."
But the system, still considered to be Level 2 by industry experts (as is Tesla's Autopilot), does not appear to be on its way to becoming Level 3 or Level 4 with its current suite of hardware and software, which permit users to completely divert their attention from the road for minutes or hours at a time, while the earliest cars with FSD that buyers bought years ago are getting older.
https://www.autoweek.com/news/technology/a41091584/tesla-ful…
Tesla (and Musk) have stopped making promises about what FSD will be able to do, or any assertions that FSD will get beyond Level 2 in any specific time frame. There will be several commercially available Level 3 systems on the market from other automakers in the next two years, apparently at closer to a $5K price for the option:
https://www.autoweek.com/news/technology/a39485558/mercedes-…
…which may concentrate Tesla’s position on this a little bit. As the Forbes article points out, FSD is a nice option play for Tesla - they’ve “pre-sold” a super-expensive software package that doesn’t exist yet, and if they can build (or buy) a product that actually does what they previously claimed, they can get a lot of money. And if they don’t manage to deliver it, they can just refund the money that people paid. They’ve been able to do that for the last 5-6 years or so, but as Level 3 systems come to market, they might have to change that strategy.
In any event, FSD isn’t close to Level 3 yet - let alone Level 5 autonomy. I have no idea why people are paying so much for it, other than some combination of brand enthusiasm and FOMO.
Albaby
Why are people willing to pay $15,000 for Tesla’s FSD?
My guess is that folks who drive a lot are willing to spend for a system that parallel parks and drives well enough to allow the driver to send a text, drink their chai latte, eat sushi, and quickly make a fantasy football trade. Level 2-3 is probably sufficient for that.
My guess is that folks who drive a lot are willing to spend for a system that parallel parks and drives well enough to allow the driver to send a text, drink their chai latte, eat sushi, and quickly make a fantasy football trade. Level 2-3 is probably sufficient for that.
Ummmm…it’s probably worth mentioning that Level 2 is not sufficient for that (other than parallel parking). It is, and will likely remain, illegal to take your hands off the wheel, or your attention off the road, with a Level 2 system. It is dangerous to do so as well. Please don’t do it.
You can do those things with a Level 3 system, but there are no approved Level 3 systems authorized to be used in the U.S. Autopilot and FSD are Level 2 systems, not Level 3 systems.
Albaby
Tesla (and Musk) have stopped making promises about what FSD will be able to do, or any assertions that FSD will get beyond Level 2 in any specific time frame.
Albaby,
Have you read anything that proposes, at some future, to allow people to belong to a CoOp of self-driving vehicles. I saw an article some years ago that proposed ownerless self-driving vehicles. This would apply to urban centers like San Francisco or New York City where self-driving vehicles would pick you up and deliver you to a pre-set destination. You don’t need to pay for parking or have garage space, worry about break-ins when parked, etc… It would also allow for Co-Op members to be collectively moved around on the same trip, conceivably reducing traffic.
Have you read anything that proposes, at some future, to allow people to belong to a CoOp of self-driving vehicles. I saw an article some years ago that proposed ownerless self-driving vehicles.
No, nothing like that. Though I’m not sure what the point of that type of ownership (or lack of ownership) structure would be. Assuming arguendo that you had a fleet of Level 4 autonomous vehicles (self-driving, but within a specific geographic area), it seems much easier to have them owned by someone who then provides service to riders. Either the “robotaxi” model (pay per ride) or a subscription service (pay for unlimited service or a bulk number of rides). Operated either by the municipality (public transit) or private companies (yellow cabs). But a Co-op model doesn’t seem to add anything.
I cannot, under any circumstances, envision anyone offering shared rides with autonomous vehicles, due to safety concerns. Shared transportation spaces can sometimes get dicey even when supervised by a bus driver, or the nearby presence of either a conductor or transit police or airport security at the next stop. Getting into a car with a stranger, and no driver to supervise? I can’t see that happening.
Albaby