Is private car ownership about to disappear?

Humans don’t need LIDAR to determine distance and motion because we have a visual cortex and eye that has undergone several hundred million years of evolution to be able to interpret an astonishing amount of visual information and integrate it into other senses. Even then, we are easily fooled by optical illusions or perceptual artifacts into mistaking distance and motion. Contrary to popular belief, perception of distance has almost nothing to do with parallax of binocular vision beyond a few inches and is the result of complex shape recognition and integration in the brain, the extent of which we are still struggling to understand.

I know there are a few people, Elon Musk in particular, who feel that computer vision will soon equal that of humans, but that is a conclusion borne out of blithe ignorance. Computer vision is one of the most difficult practical problems in machine cognition, and even after decades of effort by tens of thousands of researchers working every conceivable approach we still have yet to develop a system that can recognize a known 3D geometric shape at an arbitraty orientation and distance, and particularly if it is partly occluded by another object in the field of view. Recognizing orbitrary patterns, like a road lane with poorly identified boundaries, is something experienced drivers do almost instinctively but that computer vision systems have consistent trouble with.

LIDAR helps with rangefinding but the problem of synthesizing a wolrd model including inferrences of unseen parts of objects, like a motorcycle going around a truck and will appear at the other side at an expected time and speed, are really complex problems that have to be solved before fully autonomous vehicles are feasible for the reasons stated above. And while the costs of LIDAR and computer vision systems will reduce with higher volumes and advances in technology (although I doubt they will be so low as to be inconsequential of,the cost of a vehicle), the point remains that once autonomously piloted vehicles demonstrate order of magnatude improvements in safety and reliability, and offer the ability to offset the direct and indirect costs of ownership through communal ownership, subscription service, or making it available for secondary use when not immediately needed, the benefit to most people of not bearing the expense of a privately owned vehicle that sits idle for most of,the time is pretty evident, especially as other lifestyle costs like food, medical care, and housing continue to rise.


"One thing I realized. It’s been talked about for a while, and most car companies working on autonomy have discussed this “shift”.

Simply put, the economics mean that manufacturers have no reason to sell you autonomous cars for affordable prices at all. Why sell them when they can rent them to you by the mile or the month? And the liability involved means they can’t really sell them without you paying a software upgrade, maintenance, and vehicle replacement monthly fee…basically a lease.

I am aware that Tesla is still talking about this, and selling Model 3s that allegedly have this capacity, but the math just doesn’t check out.

Look at all the middlemen it eliminates :

Car dealers. Factory storage lots. Car Financing banks. Car Insurance companies. Most car advertising - no need to sex up a car so you will make a 40k purchase, you can just enjoy a nice car by the mile. Most of the gasoline sales will disappear : it makes sense for some of these vehicles to be hybrids that do burn some gas but they would only use gasoline doing long runs across big cities and between cities.

Middlemen remaining : mechanics, and a few centrally operated refueling/recharging stations per city. Probably 5-10% of the total gas stations and mechanics that we have now, since the manufacturer pays for repairs, they will design them to be cheaper to service and less likely to break. I don’t think these service/refueling/recharging/cleaning centers will be manufacturer owned because it’s more efficient to have them serve all brands."

I think the OP is underestimating the economic impact of this idea. Leaving aside the social impacts, this would drastically affect thousands of small business owners.

Not sure large manufacturing concerns itself with this issue. But I rather think they will when their customer base continues to erode

Humans could probably benefit from it too, considering human drivers kill >40,000 people in the US every year.

Absolutely. I was simply trying to show that the economics of it - by not paying all this middlemen - make it much more efficient. Which in turn either means large profits for the manufacturer, or in a competitive market, very low prices. Very low prices for autonomous rides make owning your own car an expensive luxury.

People will still do it, but as fewer people buy their own cars and as gas stations and mechanics who repair them and auto parts stores and car dealerships start to close down and disappear, it would be harder and more expensive to keep driving your own manually operated car.

It probably won’t happen as fast as streaming overtook Blockbuster video, but it’ll happen.

Yes, I missed a step. It will become more cost-effective to use a self-driving car on an ad hoc basis. Owning a car costs a lot of money: it has to be taxed and insured, it has to be garaged or have a parking pitch, it has to be serviced and tested annually (at least, here in the UK), and it depreciates. So as the cost of the automated car comes down it becomes cost-effective to not have a car.

Well, I’ll get right on installing laser-equipped cybernetic implants in infants. I don’t see how anything could possibly go wrong with this plan.


Interesting comparison, seeing that the telephone model used to involve leasing a phone from Our Lordships at AT&T (no one was permitted to own their own phone). Now, private ownership of phones is taken for granted.

There are many people who like renting their music in perpertuity rather than owning albums. Lots of folks prefer leasing cars to owning them.

The thing to remember about cars is their basic appeal to freedom - getting out on the road and going where you want. For many, owning your own vehicle is part of that.

I think at least one major problem with this prognostication is that it’s a purely utilitarian analysis that, even if right, overlooks the personal, subjective, and deeply ingrained cultural aspects of owning a car, particularly one that’s been customized to one’s tastes and becomes an extension of one’s personality. I’ve no doubt that autonomous self-driving cars will transform the ride-for-hire businesses like taxis and Uber, but I’m doubtful that it will do very much to personal car ownership. Car-sharing services like Zipcar already exist, touting potential savings over ownership, but they haven’t displaced car buying in any measurable way, and though they may become more convenient with autonomous cars, I don’t think they will in future.

Sure there are hardcore urbanites who don’t own cars and rent them when they want to go on a trip, but it’s mostly because it’s not practical for them to own and they’d use them very rarely, which is a niche market that doesn’t apply to most of us. I think most of us will always want to own their own transportation for personal reasons and indeed many people consider their cars one of their most iconic possessions. In fact just last week a friend of mine was in town to pick up his new Jaguar F-type, which, Lord knows, he doesn’t “need” in any utilitarian sense, particularly since he owns two other cars. Personally I’m not at all a car nut, but even I can appreciate pride of ownership. Similar principles apply to boats and country cottages, which many people choose to own despite the availability of rentals and the far superior economics and practicalities of renting.

It’s interesting to note how many people currently choose to make car ownership an expensive luxury. Indeed, the number of marques that prosper by catering to this market is impressive.

Yes - but for as long as people want to do something, there will be people supplying the goods & services they need to do it.

Look at the example of horses, whose impending demise was obvious to many when the automobile arrived on the scene ~100 years ago. Yet here in the 21st century, the number of horses - and the money people spend on them - continues to increase.
In short, there are good reasons to expect the patterns of car ownership to change. But the idea that private car ownership is “about to disappear” is dubious.

Autonomous driving doesn’t have to be better than humans at their best for it to be a massive improvement. It doesn’t get drunk, or sleepy, or distracted, or sluggish, or angry, or impatient, or inattentive. Most accidents involve at least one person doing something stupid and easily preventable. Avoiding these will be the primary benefit of autonomous cars; “superhuman” level driving may come some day but it’s not needed to reduce the death count by a huge factor.

That’s… not exactly true. Just try covering one eye while driving and see how awkward it feels (actually, I don’t recommend this). What is true is that with practice, humans are able to easily compensate and use parallax-from-motion (both head and body). I’m not aware of any great difference in accident rate between one- and two-eyed individuals.

You speak as if you think computer vision is a separate problem from LIDAR processing. It’s not. You have all the same problems, and then some, because you have a new sensor to integrate.

It’s probably better to think of LIDAR as like a new spectral band than as some wholly different kind of sensor. You still get an image out of it; it’s just that the pixels represent depth instead of color.

To use that imagery, you still need the same kind of computer vision processing. It doesn’t give you any kind of object recognition or even discrimination automatically. You still have to deal with noise and other errors (particularly in inclement weather). It’s also low-res and low-framerate compared to cameras.

For me, it’s tough to see exactly which benefits LIDAR provides once robust computer vision is available. It’s not that this is an easy problem: it’s not, by any means. It’s that LIDAR needs it too, but if you have it, then cameras are just as good. LIDAR only helps with the earliest and easiest layers of the processing stack, and not with the hard stuff like object discrimination.

It may end up, of course, that all AVs get LIDAR just because it’s cheap and easy to throw it on. And in the short term, it does make some things easier, and if nothing else you get mapping for free (an advantage for Waymo, etc.). But again, it doesn’t solve the hard problems of computer vision; depth estimation is comparatively trivial.

All of this is true; however, the issue of liability is going to require a high degree of reliability by autonomous piloted systems. With a human driver, you can blame the driver for accident and resulting damages unless there is some mechanical failure; with an autonomous piloted system, there is no clear liability, and insurers are going to be reluctant to cover such systems until they are at least as good as an alert human driver at preventing avoidable accidents. Your other points about such systems not suffering from fatigue, distraction, intoxication, et cetera, are on point, and in addition, it is relatively trivial to equip such a system with 360 degree situational awareness, avoiding many of the common accidents that occur while backing up, changing lanes, or turning through an intersection. Once such systems on on par with an alert human driver in terms of interpretation of hazards, they will already be far superior to a human driver on average.

LIDAR provides active rangefinding instead of interpolating from a “flat” image, and at a high enough resolution, can at least provide the shape of a frontal aspect. It is qualitiatively different than interpreting video imagery from a camera, which is a very difficult problem that researchers have struggled with for several decades with only very modest gains. At some point, presumably, there will be a series of breakthroughs in which visual processing will improve to the point that it is comparable to human vision interpretation, but I don’t think we can really predict when that will happen. The multiple completely avoidable accidents with Tesla’s Autopilot is evidence that visual recognition of road hazards and lane indicators is not robust, particularly when those indicators are not well marked or are confusing, as many of them are. And I don’t note this specifically to bash Tesla beyond their fielding a manifestly immature capability; nobody else really has a better capability without resorting to LIDAR and other conventional collision avoidance systems.


Those won’t actually stop, though. All of those costs would, under the OP’s concept (as well as every related concept I’ve seen) simply be transferred to the renters in some fashion, whether implicitly or openly. And most or all of them will not benefit from economies of scale in the way you think; in fact, there are ample reasons to expect some level of diseconomies of scale as it requires higher inventoyies to satisfy customer demand.

Thre short version is that while the OP is assuming an extremely specific set of social changes that force a wide array of options into an extremely narrow possibility space. In fact, the modern ecnomy shows the exact oppositte: given more choices, consumers almost inevitably show greater divergence over time as they seek more convenient, or more cost-effective, solutions.

LIDAR just provides a low-quality depth map. All of the hard problems of visual processing are still there.

There is first the problem of just grouping samples together. Primitive machine vision systems used techniques like edge detection–trying to find the outline of an object by using contrast or color to trace along the edge. This is woefully inadequate for probably obvious reasons. But depth processing has nearly the same issue. Primitive systems found an edge by looking for discontinuities in the depth. But again, this runs into problems almost immediately.

Supposing you’ve solved that somehow, now you have to figure out what those samples represent. Is it a car, a person, a tree? One can perhaps imagine matching a picture against a large database of cars at various angles to see if any are a close match. But this isn’t sustainable; you run across an unusual-looking car and the system fails. Again, LIDAR has the same problem. You could try to match your depth image again the geometry of vehicles at different angles, but it just has the same problem with the fixed database.

There’s a lot more obviously, but for basically every problem that is still unsolved in vision, it’s also unsolved for depth. And so the current method de jure of solving these problems is the same: convolutional neural nets (CNNs).

CNNs have made incredible progress, doing a decent job on both problems I mentioned (and vastly outpacing previous non-machine-learning work), but there’s a lot more work to be done. And as yet unproven if there’s some giant additional step that needs to be taken.

It is true that LIDAR is not a panacea—the processing system still has to integrate data fast enough to form a dynamic model of the local world, and be able to distinguish between discrete moving and static objects as well as perceptual artifacts, although LIDAR operating in the near-infrared range have little in the way of anomolous phenomena—but LIDAR doesn’t just give a “low-quality depth map”; it provides a dynamic vector field of the local world, and at automotive control ranges it provides a resolution of ~5 cm or better, which is certainly enough to distinguish objects larger than a baseball even if it cannot provide a detailed “image” of a human face. More importantly, it doesn’t require perceiving identifiable objects in order to establish a potential hazard far beyond that of radar-based collision avoidance systems and can operate in complete darkness to perceive objects not emitting or reflecting headlights or ambient light, or if blinded by sunlight or high beams. This is independent of the problem of actually identifying an object.

The claim that visual-only sensing is “good enough” is belied by the failures of vision-based current systems such as those used by Tesla or Uber, which admittedly are still in their infancy but demonstrate the immaturity of machine vision approaches that require the system to positively identify a potential hazard through pattern or object recognition alone rather than active sensing of a potential hazard regardless of whether it is positively identified. Tesla, Uber, and others pursuing this approach are doing so not because it is the best technical approach but because they want to be soonest to market at a cost point that makes LIDAR-based systems prohibitive. There is, however, a massive disconnect with research into machine vision, which indicates that we are still far from building systems that can process video information into a coherent and accurate model of the world that is anywhere close to what a human brain can do, and the ambitions to field a visual-only autonomous piloted vehicle which can result in injury and death to the occupants and bystanders should it fail.

And as scr4 notes, even human drivers fail at this all too often, in part because of inattentive or reckless driving but also because of confusion of the image scene, being blinded by bright lights, or just misjudging distances and speeds. Making an visual only autonomous piloted vehicle would require a perception system that is actually superior to human vision, and we just aren’t anywhere near that capability right now or in the near future.


I don’t have time to read the whole thread, but while “ordering” a self-driving car like you would an Uber would work just fine in cities it would definitely NOT work in suburbs or the country. If you live in a place like north-eastern Vermont and the nearest house is too far away to see you NEED a car in the driveway in case of an emergency. Waiting over an hour for the nearest self-driving car to show up may not be an option.

I think that there will be a shift towards not owning a car for many people, but it certainly would not make private car ownership disappear for a long, long time, if ever.

It depends entirely on the price points, of course. Owning a car is expensive. There is obviously the purchase price, then the maintenance costs, along with the gasoline which is the only real variable cost.

And this is for something that is sitting there idle 90%+ of the time.

If the cost of me travelling to and from work would be less in a self driving autonomous car, then I’d get rid of my car and switch, no problem. I don’t mind driving, but I’d rather not for the most part, and I certainly wouldn’t mind saving a bunch of money by not having to own my own car.

There will absolutely be those who want to own their own car, especially since you’d have to own a car to have any chance at having manual controls (they aren’t going to put manual controls in transport vehicles), but I do think that many will opt into saving the money, space, and hassle of personal ca ownership. The question is, if you own your own self driving car, are you going to have it sit in your driveway or parking spot 90% of the tme, or will you sent it out to make you money.

As is, the transit system nearly anywhere sucks. If I want to take a bus down to the city, I have to travel at least 3 miles to the nearest stop, and it only has 3 routes in the morning, one in the afternoon, and 2 in the evening. And that is not in the direction that I work in, if I want to take a bus to where I work, well, I just can’t, there isn’t one. I need to have a car in order to get to work.

With a different model based on electric AVs, I can have a car come pick me up in front of my house, or maybe for just a little cheaper, pick me up at the end of my road. If I don’t mind sharing, and I book in advance, it would not be hard to create very efficient routes, and much more adaptable than bus routes, as it would be much easier to send a van rather than a car, or send another car, if the number of passengers on a route increases, than it is to try to assign a new bus to it.

As far as maintenance, I think the OP goes a bit overboard, but the cars would be better maintained. You would need fewer fueling/recharging stations, as right now, mst gas pumps are empty most of the time, that could be allocated more efficiently. At the stations, there would be some basic testing and diagnostic systems. While it is recharging, it would not be hard to check the responsiveness of sensors or other minor calibrations. The car would also have fairly extensive on board diagnostics, and would be able to run itself into the shop anytime it needed preventative maintenance, rather than what many of us do, drive it till it dies then get it towed to a shop.

With cheap and prevalent AV’s, I do think that that standard for having a driver’s license should be much higher. We set it at a pretty low standard today because having a car is essential to survival. Unfortunately, being a bad driver is not conducive to survival. Just eliminating the worst 10-20% of drivers would probably decrease the crash and fatality rate by more than 50%. Only allowing those who are currently in the top 25 to 10% of drivers would pretty close to eliminate it.

Give it 100 years, and maybe private car ownership will be at an end, but only as far as basic transportation. People would still, even then, want their private cars for recreational activities, even if they are limited in the areas they may drive manually.

Agree. Military, police, fire, and small rural homesteads would manual drive vehicles.

As a side note, the same technology that makes autonomous vehicles possible would eliminate a lot of farming jobs.

At some point, I expect only self-driving cars will be allowed on public roads. Vehicles for those special needs may have manual-driving capabilities, but not be 100% manual.

And I think cars can be more efficient and cheaper if all cars were self-driving, because there would be less need for passive safety.

When your car is able to drive itself, you actually own a taxi that is dedicated to you and your family. A taxi that is sitting idle for hours and hours every day.

A fleet of self driving cars that can be used 75% of the time instead of 10% would be significantly cheaper to run. High asset utilization is the name of the game.