I Was Proved Fucking Right
Peak "driverless car discourse" was a few years ago. Some subjects don't just bring out disagreement, civil or otherwise, but extreme condescension....A bit like "Iraq war is a bad idea, yo," "driverless cars aren't going to work" brought out the kind of "you just don't know how the world works, silly child" haughty responses. And then once everyone (not everyone, but more people) realized Elon's cars were never going to work as promised, and even Waymo, which is doing it more sensibly, stopped promising the future was already here, people just stopped talking about it.
....The fatal flaw in the concept is even if they work surprisingly well - and even watching scary videos of them not working, they still do work surprisingly well - they absolutely require the thing that they are supposed to free us from. They require 100% driver attention for the moments when they fail, perhaps even more so than normal driving, which after awhile does become somewhat instinctive.
Well, now, hold on there. It's true that we haven't heard a lot from Waymo lately, but that's because they suspended testing during the pandemic. They may have lost a year, but as near as I can tell they're still basically on track. They're just more realistic about it than Elon.
But here's the part I don't get. I'm not trying to start a blog fight—I kind of miss them, but there aren't enough blogs left to risk one—but I find it odd that Duncan keeps talking about the "fatal flaw" being that driverless cars have to work perfectly, more or less. That's true, but far from being a deep secret that he alone has figured out, it's something that every single person who works on driverless technology is keenly aware of.¹ It's precisely the thing they're trying to accomplish. I think it's fair to say that it's turned out to be harder than anyone thought, even after they admitted it was harder than they thought, but 2025 isn't here yet! We still have a few years to go before the current predictions about fully driverless cars have to be shitcanned.
As for "people stopped talking about it," that's wildly the opposite of the truth. Not only are people still talking about it, but every car manufacturer on the globe has jumped on board. Maybe this hasn't produced banner headlines on a monthly basis in the New York Times, but the auto industry is completely abuzz about it. Just last month we were all talking about Apple and Hyundai in negotiations to build a plant to manufacture driverless cars.
Anyway, the general chatter about driverless cars has been muted lately because of the pandemic and also because it's been overshadowed by the boom in electric car hype, but it's still out there. I continue to predict that by 2025 I will be able to tell a car to take me to Duncan's address, and then sit back and take a nap until we get there.²
¹Except maybe Elon, of course.
²Though I suppose I'll still have to wake up once in a while to pump some gas. I don't think anyone has automated that yet.
2017 was kind of peak Self-Driving Car hype, but even after the disappointment they've still been working on it plenty. I figure at some point the auto-industry will go all in, get it about 95% of the way there, and then lobby for changes to signs and such that they can go the rest of the way.
There’s two problems with doing that. The first is that it’s increasingly looking like the only way that autonomous vehicles will be genuinely anything more than a dangerous novelty is if the world and everyone in it are remade to suit the needs of self driving vehicles. Every road will needed to be re-engineered and rebuilt. The way to avoid killing people is either to severely limit their movements or made the wear a transponder of some kind. Hardly seems worth it.
The other problem is who will pay for whatever needs to be done. For the moment, it looks like the plan is to externalize the huge costs of making the world safe for autonomous vehicles. And the same with injuries and deaths. That means you and I will pay through the nose but the autonomous vehicles companies will reap massive rewards. I say, screw that.
Are self-driving cars going to be FULLY INSURED against the damage they may cause to other drivers BY THE MANUFACTURER WHO CLAIMS THEY ARE READY TO GO ON EXISTING STREETS ??
The companies certainly don't want to address this,, and all the prospective users, at least on this thread, don't seem to realize that it's an issue. Is there a publicity-seeking defense attorney out there, I will be your client for a suit against all the companies and all the state legislatures that attempt to put these things on the road without being fully insured against all faults for any problems of any type that their machines may cause.
Apparently one of the things that various players have sought from time to time is immunity from all forms of liability. Individuals who owned the cars would need insurance but it isn’t clear how that would work if everyone else is immune and also has absolutely no incentive to care about safety precautions.
Why don't we make the whole car out of the gas tank?
-- Ford Pinto design team
Large corporations, like most car companies, don't need insurance. They can cover the risk out of their own resources. The correct question is whether or not they are liable for damage caused by their products.
I think all large manufacturers have substantial products liability coverage. Even those that have captive insurance companies layoff with the big reinsurers. There’s very few companies with the resources to self insure
Buying state legislators, congresspersons, and judges is cheaper than buying insurance.
Mitch, we've been underinvesting in road and transport systems for DECADES now. This is to the point that there are a lot of roads that are very dangerous to drive on with an aware human driver in the car. I regularly have to drive on badly maintained roads with degraded sign and lane markers that can be hard for human eyes to detect in rain and sleet.
IMHO, we need to put in a lot of investment to get our roads safe(er) for humans to drive on. I don't think it would take all that much more to add tech to roadways that would work with autonomous vehicles and tax breaks to encourage shifts to them.
We bought a Nissan Leaf just under a couple of years ago - they don't make any claim of autonomous driving. But the improvements they have put in greatly help me be safe. Best one is very reliable adaptive cruise control - automatically slows you down to keep a safe distance back from the car ahead - right down to a dead stop. I even had it do a serious brake application on me when an idiot zipped out of an entrance lane, realized the traffic was coming to a halt, and panic braked. So did the Leaf, very well, and a good heartbeat before I started on the brake. Good thing there was no one behind me...
Side sensors intelligently tell me what's coming up behind and either side of me and give me warnings based on how fast they are approaching - the side mirrors give visual warning, and I get a stick shake if I turn on my turn signals. Rear sensors activate when I back up up out of an obscured parking space and there's a pedestrian or car coming up from where I can't see and give visual and audible warnings.
I'm quite happy with the progress - but want a lot more future prep that will be needed just for better current human aid tech to get in place, which could help autonomous cars and could definitely help current driver assist systems.
I agree that the lack of investment in road and transportation is a real problem. Our roads are degrading to the point where in some places it’s like a third world country. But the problem is that the upgraded roads that would benefit autonomous vehicles would typically not be of much valued to human drivers.
https://www.mouser.com/blog/creating-safer-roads-for-autonomous-vehicles
Weirdly untrue. Since humans make the same mistakes. All the time. More of the time.
Yeesh.
They do and they don’t. But the point here is that while improving roads can make them safer, the needs of humans and autonomous vehicles are very different. Consequently, road improvements designed to aid autonomous vehicles will generally not make things safer for humans and might actually make them more dangerous. That was the point of the article I linked to.
Dangerous novelty...
...Based on what, again?
They don’t function safely on their own, they don’t function well on their own, and they’ve killed at least one person, probably more if you count Tesla’s alleged self driving feature.
Thank god none of that applies to any human.
Self driving just means more time for Elon & Grimes to bang.
I see you're not the sharpest tool in the shed.
ilil - The funny thing is, Word Press has such bad formatting I'm not sure which comment your worthless ad-hominem applies to. So I don't even know which side of the argument you're not supporting.
Perhaps you could use your obviously superior intellect to help me to appreciate the keen analytical skills of my betters by explaining where I have gone wrong.
I hope you're right. My driving abilities are declining as I age. Fortunately the distance I want to drive and the number of trips I want to take are also declining. So it ought to be easier to develop a driverless car that meets my needs before I become completely uncapable of driving.
Ditto! I'm hoping my next car will be fully electric powered and with at least a driverless option. Of course, this will give even more horrifying undesired outcomes, like children zipping off in the family auto (running away becomes practical!), or sending the car off to nowhere, or even more disgustingly, people setting the car to just orbit the store or their place of work until they get out, to avoid having to park.
It's always something.
Today we are all Joe Biden ascending a staircase.
Oddly enough, this post and "the death of driverless cars" comes on the same day I saw a news story about Waymo testing driverless taxis in, I believe, Arizona.
I'm not a fan of this concept and agree that they really don't eliminate the need of somebody in the car from paying attention 100% of the time. But, it's pretty clear that "the death of driverless cars" is premature.
You’ll notice, however, that the place chosen is also the place where it’s now established that they can kill people with impunity (in large measure because of political corruption ). If the technology was really ready for testing in the world, they’d be conducting the test. In Los Angles or San Francisco where they’d would at least get a ticket for murdering innocent civilians.
Can you explain??
Yes, if the technology was genuinely ready to be tested in a place with a lot of car, they’d be testing it in such a place. But instead they are returning to a place where their not ready for prime time car killed a pedestrian without then suffering any consequences. The implication is that the company suspects that it will be killing more pedestrians and perhaps even people in other vehicles and it was to be sure nobody will have to go to jail or pay millions in damages when that happens.
Governor Ducey issued an executive order to allow them to operate. You can now summon a driverless taxi: https://www.fox10phoenix.com/news/no-driver-needed-waymo-rolls-out-fully-autonomous-vehicles-in-chandler-tempe-mesa-areas
Again, is there any requirement that the manufacturers are FULLY COVERED against whatever damage they may cause for any reason ?
How does that mean they escape liability for accidents?
Doesn’t explain how they’re able to operate without liability.
Why are you busy lying?
Why are you comparing the company that removed safety features and used an impossible testing regime with a company that has yet to suffer an injury in millions of miles driven?
Kevin, I think your footnote #1 is precisely what Atrios is banging on about. That is to say, Elon has done the driverless car community a huge disservice, with his loose talk and bullshit. And sure, Atrios goes too far in tarring Waymo. But really, his be`te noire has been Elon. That's what gets his fingers a-tappin'. And rightly so.
Atrios has certainly been right about those who keep saying driverless vehicles will be ready real soon now, which includes Kevin, because the arrival keeps being put off. He has not been proved right if he said that they will never arrive, because the future isn't here yet.
But Kevin's claim to be right is especially bad because they aren't here now and the actual tests show that they are not close to being ready. Apparently Teslas have a habit of running into parked vehicles - that is a pretty simple task compared to what a real driverless vehicle would have to do. Kevin will be proved right if and when there actually are driverless vehicles.
Kevin expects to be credible, but he sometimes makes unjustified claims like this. Of course he is a blogger and not a scientist, but this is just simple logic. If you predict something will happen, you are not right until it happens, not when other people claim it will happen.
Atrios has never claimed that they will never arrive, just that it is a much harder problem than Kevin thinks it is, and far beyond what Musk claims that Tesla can do right now.
Atrios literally has claimed that. In the article Kevin quotes, he literally says that.
https://www.google.com/search?q=site%3Aeschatonblog.com+never+driverless
No he doesn't, he says that *Elon Musk's* cars will never be able to do what he claims. If you look at a Waymo car or Cruise self driving car (there are plenty on the streets of SF being tested) they both have tons of gadgets on the roof to try to figure out the world around them. Teslas ... do not. In fact, he even gives props to Waymo for being closer.
I disagree with the notion that driverless cars have to be perfect. We don't hold human-driven cars to that standard. All driverless cars have to do is be better than human-driven cars.
Everybody agrees with this but despite what Kevin says, they aren't anywhere close to that.
Apparently, Waymo is at around the level of 18-19 year old drivers.
Have you seen what it costs in insure 18-19 year old drivers?
At least a driverless car won't decide that driving is too much & end up making TikTok videos in their bedroom instead of going anywhere.
Driverless cars: NOT Claudia Conway.
But human and software imperfections are very different. The Uber car that killed the cyclist in Phoenix wasn’t instrumented or programmed to monitor traffic in adjacent lanes, so it wouldn’t swerve to avoid the collision. An inattentive human driver might have also hit the cyclist, but that would have no implication for what any other driver might have done. But it is clear that in every case, for every Uber-equipped car, if an obstacle appeared in its path too late to brake, the Uber car would hit it.
You'd think automating a gas pump would actually be easier than the rest of it.
Odd that Kevin thinks that he will still have to get gas. Electric cars may be here before self-driving ones.
Precisely. One of the curious things about autonomous-vehicle developments is that, after the relatively simple people-movers and trams, attention went right to the most difficult problem: automobiles. I expect we’ll actually go through a little sobering-up period, and the first implementation will be over-the-road trucking.
The easier autonomous vehicle solutions are off-road. For instance, mines are starting to use autonomous haul trucks; all they have to do is shuttle back and forth along a defined route a couple of miles long moving loads of rock from the pit to the crusher. I expect farm vehicles can be fully automated fairly easily too, since they operate in a defined space with well marked boundaries and no other traffic. This would be a huge boon to farmers, who tend to spend much of their time in the driver's seat of a vehicle of some sort.
Yes. But the sales volumes would be much smaller than for personal autos, and I think it was the visions of huge profits that led to the pursuit of the difficult before mastering the simpler applications.
Maybe Kevin has no plans to be in Oregon.
We don't pump our gas; we pump our smug sense of self regard for our vegan organic fairtrade artisanal moonpies & bespoke RC Cola.
Yeah, I’ll take Atrios and the points. If you’re driving through Irvine with no hands every kid is going to step off the curb and slam on your breaks for fun. Any tech that be stymied by bored 14 year old boys is insufficiently robust.
This seems very unlikely to me.
I know I read that in one of the cities where self-driving cars are being tested, human drivers quickly realized how easily intimidated they are, and started bullying them at 4-way stops, and the like.
This is the actual prequel to the Terminator film series.
How are they going to know?
...And how would that be bad? Won't they get bored of playing in traffic?
Yeah, I’ll take Atrios and the points. If you’re driving through Irvine with no hands every kid is going to step off the curb and slam on your brakes for fun. Any tech that be stymied by bored 14 year old boys is insufficiently robust.
ummmm.....
pump gas????
Really????
I hate getting out of my car to pump gas. I miss the days when there were still gas attendants. Are there still gas attendants in NJ? Years ago it was illegal to pump your own gas in NJ.
Yes, they still pump gas there. Even at places like Costco. I go to NJ periodically to visit my son and his family. It’s always kind of a shock. And I always think about how my 94 year old mom who still drives but has trouble with gas pumps would like it there.
Years ago, I was on a business trip. While driving on the Jersey Turnpike I pulled into a service island, got out of my rental car and started pumping gas and a service attendant came running up and shouted "What do you think you're doing!" I'm surprised to hear that they still have service attendants.
When I got to Oregon & asked another transplant (from New Jersey) if her homestate was also still requiring full service refills at has stations, she lied & told me the Garden State had phased it out.
Damn.
^^^And I always think about how my 94 year old mom who still drives but has trouble with gas pumps would like it there.^^^
Wow. 94 and still drives AND pumps her own gas? Your mom sounds amazing.
She actually was a Rosie the Rivieter during WW2.
I think it's safe to say there is a < 1% chance that Kevin will be able to make this trip in 2025 as described. Unless Kevin is planning on traveling with someone else sitting in the driver's seat.
There's the tea.
Here is KD in 2013 predicting that "genuine self-driving cars will be available within a decade and that they’ll be big game changers". I guess he has 2 years yet, but that is extremely unlikely to be an accurate prediction.
https://www.motherjones.com/kevin-drum/2013/01/driverless-cars-will-change-our-lives-soon/
For a more complete trip down memory lane, which includes the different page stylings in use by MoJo at the time:
http://web.archive.org/web/20130128023019/https://www.motherjones.com/kevin-drum/2013/01/driverless-cars-will-change-our-lives-soon/
To be fair (but only a little), in that particular post all Kevin wrote was "soon." He didn't say (that I could see) ten years or a decade etc.
Going to take the over on 2025. Think we are a decade or two away
I’m wondering if my 7 year old grandson will have to learn how to drive. I suppose his kids won’t even know what a steering wheel is in the same way the kids in that video didn’t know how to operate a dial phone.
Given there will still be times when the "auto drive" feature will be flummoxed and need help, yes, your 7 year old grandson will still need to learn how to drive. That is, if he is going go be in a driverless car rather than a car with a driver other than himself.
Even aircraft autopilots still wish to disengage from time to time...
Aircraft autopilots aren't fully autonomous, either. You're not supposed to leave the seat.
I used to drive a car that had a choke, and a manual crank for the windows, and cigarette lighters, ash trays, white wall tires, a chrome bumper, manual door locks, a floor switch to turn on the bright lights, just an AM radio, no seat belts, and a hand crank to turn the engine. Not sure why I had the last one, the car had an electric starter. My grandfather used to start his car with a manual crank.
There are too many edge cases such as roads with no striping, snow covered roads, construction areas with unclear lanes. No way to program for that. How about fine maneuvers such as parking, going into a car wash and such where you want to position the car somewhere just where you want it.
I can see self-driving on interstates which are well marked. About ten minutes before your exit the car will try to get your attention. If you are sleeping it will pull over somewhere till you wake up.
There was a local case where a guy fell asleep in his Tesla going down the interstate. He was speeding and the CHP tried to pull him over. Sirens and horns had no effect. Finally, they did a traffic break then one car got in front of him and gradually slowed to a stop which brought his car to a stop. Alcohol may have been involved.
"No way"
https://www.youtube.com/watch?v=DIBObV-_42I
Tesla already does that. They're fine tuning it; it doesn't have weather adaption activated yet, either. But it can follow a snow-covered road.
I watched about half the video. There were many places where he couldn't use autopilot because the car couldn't find the lanes. Even when it did work, I think it was dicey. When he was on that narrow road it was following the tracks from earlier cars. If a car had come in the other direction it would have been interesting to see what happened. Even for people this is hard. If it had been new fallen snow with no tracks from other vehicles it would have failed. The driver pointed out several times where he couldn't engage the autopilot because it couldn't see the lanes under the snow.
I'm not trying to complain about Tesla's performance, just saying that there are really hard situations that the software can't handle.
"I’m wondering if my 7 year old grandson will have to learn how to drive."
He'll probably never learn to write. Cursive, that is. Joined-up writing. He'll use a fingerprint or crypto device or something to sign documents -- which will not be printed out. The term "fine print" will make sense only to doddering geezers.
Except that Atrios *was* right; it is a much harder problem than people made it out to be and there is no reason at all to believe that self driving cars will be as good as a bad human driver by 2025. To give an example, I do a regular 60 mile freeway drive and very, very often, I am consciously thinking about something else, a math problem or whatever, and realize that I don't remember a single thing that happened over the last 20 miles or so. And yet, my car didn't run into anything else. That's where self driving cars are right now. They are as almost as good as a driver that isn't really paying attention under ideal conditions. If I allowed the same thing to happen while driving in the city, there would be dead bodies all over the place.
Is this like Ray Ratto's Piles of Grandparents?
Off topic, but if you really want to read about Elon's (and now Richard Branson'ss) vaporware, here is a gushing article about what it will be like to ride in a hyperloop car based on a single successful test in a 1 mile long tube in the middle of the desert. Where we are going, we don't need right of ways, concrete, steel, or any of those expensive things, we just need to imagine it.
forgot the link: https://www.sfgate.com/local/article/SXSW-panel-reveals-what-the-Virgin-Hyperloop-from-16038809.php?IPID=SFGate-HP-CP-Spotlight
"Where we are going, we don't need right of ways, concrete, steel, or any of those expensive things, we just need to imagine it."
Astral projection of course!-)
Heaven's gate.
To paraphrase Homer J. Simpson, "Do!"
I’m sure that 10 years from now when the hyperloop between Red Bluff and Redding is nearly complete you will look back on your skepticism with amusement.
Other than the lack of riders, that's a nearly perfect location for hyperloop as they really could run most of it down the I5 corridor. I'm to lazy to work out the cost per ride if all 100K residents ride it every day and they meet the optimistic goal of $100M/mile in construction costs. Here is the best write up I have seen so far about the test. FWIW, I've read elsewhere that the reason Virgin is so interested in funding this is that it helps to discourage investment in HSR projects that actually work right now thus keeping distance travelers bound to airplanes. That makes some sense as a few hundred million is a good investment to protect $5B in annual revenue.
https://defector.com/virgin-hyperloop-has-invented-the-worlds-crappiest-high-speed-rail/
I want the flying cars & aerial interstate highways of Steven Spielberg's Minority Report.
(Allegedly, that was a futurist interpretation of Washington, D.C., 2051. Only 30 more years.)
Considering 80% of Atrios posts are nonsensical non-sequiturs just meant to drive people to the comments (and generate ad views) I wouldn't worry too much about his theories on much of anything. Though, oddly enough, he is correct on self-driving cars.
All of our assumptions about driverless cars have so far been predicated on entirely passive roadways. Autonomous vehicles have to detect (hard) and predict (nearly impossible) other vehicles and their actions.
But what if vehicles weren't acting in relation to each other, but in response to a routing system not unlike packet routing on the internet? Google is probably already close to being able to provide such a network even without any kind of sensors built into the road system. Add some additional detection capability to roads and you're in a whole new world.
Back in 2003 or so as part of the Halo 2 hype there was a story set in the era (2550s) where the teen girl told her dad to "let the road drive" and that has always stuck with me as the most probable method. Cars that inerface with and are guides by a central road network that feeds them the positioning data.
I've thought that as well, for the last five percent of those edge-case assists. Note also that cities would in all probability have that infrastructure in place long before the (much) more rural outlying areas. TL;DR: More inequality!
Well, let's talk about specific programs, not "driverless cars" in general. Let's talk about Waymo.
47 contact events in 6.1 million miles driven. I can't seem to find data. Is that good or bad? Bring citations.
My personal experience - I lived in Mountain View and saw the Waymo cars all the time, was that they were a bit annoyingly cautious, but otherwise not a problem.
Nothing against Wayno exactly, but I take their reported errors with skepticism. They are self reported and they set their own bar for what triggers a driver to take over. If I was testing the software I'd want to set the sensitivity high so that I'd find more edge cases that could be evaluated and fixed.
California DMV requires those doing testing in CA to obtain a license and to submit annual data on disengagements: https://bityl.co/63Zp
What you're citing comes from Waymo's own annual safety reporting, specifically, "Waymo Simulated Driving Behavior in Reconstructed Fatal Crashes within
an Autonomous Vehicle Operating Domain": https://waymo.com/safety/
Yeah, they seem to manage as many 'WTF' moments as I do, as a human driver.
Yeah, the fun fact is that the biggest problem seems to be human drivers rear-ending the overly cautious waymo vehicles.
https://aaafoundation.org/rates-motor-vehicle-crashes-injuries-deaths-relation-driver-age-united-states-2014-2015/
Waymo is reporting ~770 crashes per 100 million miles driven. This is roughly comparable to the accident rate for 18-19 year olds.
But when do we get our jet packs ?
It doesn't seem like anyone here has a Tesla with the FSD package. I have had one since July of 2019, so I have seen all the progress since then.
Its absolutely astonishing. I was going to go on about all the features it now has but to keep it short I would just point out that it is really no comparison between Waymo and Tesla.
Waymo is working on geo-fenced streets, its really more like an expanded trolly system conceptually. I don't know how the system could even scale up to the entire real world. And for taxis it may not have to scale up. Second, each Waymo car is so expensive its not even like they would be for sale.
Teslas are available now, anyone can get one. And although the features of self-driving are rolled out to the fleet on an "action by action basis" once the action, or feature, is rolled out it works in all situations it is designed for.
The expansion of features now includes stopping at all stop signs and traffic lights, and it even includes a signal when the light changes from red to green.
Driving is only reacting to stimuli. The hard part has been the technology of using cameras to identify objects humans easily classify. Actually controlling the car is the easy part.
Oh, so if they haven’t yet introduced the ‘pedestrians steps into the street without looking’ module, you just make sure that your trip doesn’t include that event, then?
"Pedestrian steps into the street without looking" is likely to get the pedestrian killed, whether the car that hits them is human controlled or computer controlled. Computers have faster reaction times than human drivers, and would hit the brakes sooner if a pedestrian suddenly appeared in front of them. In other words, autonomous cars are less likely to kill pedestrians than human controlled cars are,
Jaywalking will be a capital offense in the near future.
Waymo is Dredd.
Jaywalking already is a capital offence, for a few pedestrians.
Driving is not ‘only reacting to stimuli’. It is also situational awareness, among other things. The Uber car that killed the cyclist in Phoenix didn’t swerve because it wasn’t programmed to monitor traffic in adjacent lanes. That is something I did constantly when commuting weekly on the Garden State Parkway before I retired. Swerving into the empty right lane would have saved a life.
That's the thing about testing; it uncovers flaws. Presumably, the driverless cars now being tested are programmed to monitor their surroundings.
That woman died.
Meaning Uber & Jim Boeheim have at least one thing in common.
Waymo can scale up, because every one of their cars is literally the same car, and remembers roads it has driven on.
All systems are geofenced. Even Tesla's.
Remember the PCH at Rat Creek? Well, forget it.
"Though I suppose I'll still have to wake up once in a while to pump some gas. I don't think anyone has automated that yet."
Kevin, you're going to spray gasoline all over your electric, driverless car? Shame on you! It seems crazy to me to ever manufacture an internal combustion engine driverless car. That's mixing the past with the future in a silly way.
Very steampunk.
The thing with Musk's approach, however, is that he's made it available to consumers right now, flaws and all. Assuming the driverless hardware doesn't change -- lidar/radar/other sensors -- don't change much, the vehicle built in 2015 won't be at a major disadvantage to one built in 2025; all that's needed is software updates.
You're going to want to cite that.
Oh, right, you're wrong.
'City Streets' is being run by 2000 beta-testers selected and watched by the developers.
That is literally not 'made available to consumers right now'.
I continue to believe Kevin is quite correct about the eventual arrival of truly* driver-less cars.
I continue to believe he's absurdly optimistic about the timeline of their arrival.
2025? You kidding me? But perhaps in the 2040s? Sure. I think that's pretty plausible.
(And yes, SOME/MUCH of the delay will be not because of technological shortcomings but due to regulatory, political and commercial issues -- including public reticence...Other than that, Mrs. Lincoln?).
* "Truly" means just what it says: not driverless single route shuttles from a hotel. Not driverless goods transports from a factory to a rail depot. Not semi-driverless cars that allow you to plug in a route, but still require you to monitor the auto-pilot. Not partial driverless cars that are good at slow speeds, but not on the Interstate, or are fine in perfect conditions but not when it's raining. The technical refinements and regulatory/marketplace developments to truly replace humans behind the wheel of all trucks and cars is still a couple of decades (give or take) away. Quite possibly longer. It's coming. But not that quickly.
A robo-driver that can do everything a human can will have to be as "smart" as a human. The people who think that all that is needed is the right software fail to recognize that the hardware just isn't up to it yet. I don't know how long it will be before it's possible to build an autonomous car with human level intelligence, but I suspect that it will probably take at least another decade.
Why would it need to be as smart as a human?
Because the difficult cases require that level of intelligence; heavy snow, dirt roads, missing lines (where I live, it snows a lot and by the time spring comes, the snow plows have scraped a lot of the line markings off the road), driveways, etc.
Well, no. I've argued (with you among others) that is is harder than anyone thinks since the beginning. And I'll argue right now that it is STILL harder than anyone thinks, including you.
Driverless cars use machine learning and machine learning is incredibly powerful but also with pretty severe limits. It is basically memory, and if you think that every problem can be reduced to simply remembering then sure, it looks easy. Just around the corner, in fact.
But stating it that way makes it obvious what the problem is -- the solution to every problem is NOT just remembering. Some things are novel situations and need to be figured out. Machine learning won't do that for you. It is a compact way of remembering what it has been trained on. The further a situation is from what it has been trained on, the less reliable its ability to solve the problem.
So if the end goal is "I can sleep on my way to work," you won't ever get there.
No, machine learning is hashing. Novel situations that hash to the right bucket are handled just fine. Novel situations that hash to the wrong bucket don't work so well.
I wish there was a way I could block commenters like you.
Wow. What a curmudgeonly bunch of comments here. Sheesh. The driverless technology in every car is navigation. Most cars sold new today, have all sorts of driverless tech running parts of the driver experience. Adaptive cruise control, ABS, blind-spot alerts, lane assist, self-parking. The path to fully autonomous driverless is incremental additions to the driving experience until there is no reason to actually drive. And every billboard that tells people, not to text and drive is an advertisement for driverless cars.
You're right about the path of incrementalism. But not sure why you're calling us curmudgeons. Most of us accept that Kevin's vision is coming. Eventually. The main dispute seems to the speed of its arrival. I say 2040s at the earliest. You differ?
Driverless cars don't have to work perfectly. They just have to work better than humans. When humans drive, it requires 100% attention from the human, and sometimes humans don't provide that level of attention. When driverless cars drive, humans can provide less attention than they currently do for the same level of risk.
You should read Atrio's actual blog post. His argument is that humans will have tokeep 100% attention and be ready to take over because there is no telling when a situation will arise that the car can't handle. The attention of a human driver can actually wander when the danger level is low. This is his argument not mine, but robot cars will not really be successful until *no* human attention is required.
From time to time, situations arise that human drivers can't handle. Those situations are called "accidents" or "crashed". As long as robo-drivers have a lower accident rate than human drivers, they will be good enough.
***sigh******
This is a tough one for me
On the one hand I like the technology and find it interesting.
So my thoughts on infrastructure bills submitted to Congress should include a HUGE investment in GPS, 5G wireless internet and road improvement. All 3 have to have redundant systems to make sure should one go down, a back up system kicks in. I would try this on an interstate that isn't traveled as heavily as say I-95 or I-40.
But this now leads me to another subject. Just how badly do we want government in our lives? They will of course have access to where we are at any moment in time and does the government own and operate the 5G systems used for this ? Or does this rely on private companies and drivers supplying their own 5G systems?
I'm old and I guess I will NEVER understand the younger generation wanting, no DEMANDING that everyone know where we are at every given second of the day and be in constant contact with everyone.
Whatever happened to "just" going for a ride with the family? "Hey lets take this road and see where it takes us" Let's take a trip to nowhere. No now we have to have point A and point B and let some "system" give us the route, set our speed, get us their in the most efficient means possible and monitor each and every foot we travel along the way.
Get out of my back pocket - get out of my car. Let me drive. I don't text/talk while driving - I don't drink and drive and I like more than anything else to have control over what I am doing - especially if it involves a 2500 pound weapon traveling at 75mph.
I am NOT relying on a system, a company, or a government to protect me under those circumstances. Thats MY JOB.
The technical term for "let's see where this road takes us" is 'kidnapping', in my experience.
Sorry for your experience Scent
I my world it meant lets get lost together or lets discover whats at the end of this road,
Meta, but I seriously enjoy the back and forth on this topic between Atrios and Drum. It's silly, so it's entertaining at least. So far, I'd agree with most posters here that Atrios is right. He'll continue to be "right" until driverless cars are deployed. But...
It occurs to me that it's as likely there will be other, better, means of individual, "driverless" transport that come along in the same time frame, so this whole debate may be moot. If we had more sophisticated mass transit, a bus system with stops at every residence that requested it, I'd probably use that rather than a car. That's how it seems to work, sort of, and has, in large cities with subway systems. Most folks that live in them don't even own cars.
And why isn't the topic of teletransport part of the discussion, yet?
^^^^a bus system with stops at every residence that requested it^^^
Well, that's driverless technology. No way such a system works with human drivers. I do think one of the longer term implications of driverless technology is end of individual ownership in favor of subscription models and robust, smart driverless public transport systems.
I have said this before but it bears repeating: Driverless cars are a technology with minimal benefit to society; they don't use less energy than present day cars; they are not going to be 100% safe. The project is entirely a vanity project for types like Elon.
The tragedy is that huge resources have been used to go after this project at a time when we have an urgent need to solve a slew of problems in the energy sector. It is utterly irresponsible.
The difficulty is that we still do not understand how the brain works. We know well how a neuron works but not how they cooperate to produce perception, memory, reflexes and what we call "thinking". In computer lingo: We have no idea of the "code" of our brain. Cars and roads (!) have evolved over more than 100 years to be safely controlled by the brain of the driver. To have a driverless vehicle operating along human-driver cars requires a computer to simulate the human brain. Does anybody think that this is easy to achieve with no baseline knowledge?
BTW Kevin: A driverless car that can't pump gas? First of all: Would anybody buy that? Sloppy thinking.
Second: Driverless vehicles will appeal to the same crowd as electric cars. They will be electric. The fact that Elon is keen on driverless is no coincidence.
Why would a driverless car be just as inefficient as a human? A driverless car would plan out the route, taking advantage of real-time incoming info about road conditions, congestion, accidents and the like. It would optimize velocity and acceleration to maximize the number of green lights and minimize waiting at red lights. No jackrabbit starts, veering from lane to lane, routine heavy braking. Just like any hyper-miler would do, only with the added advantage of, as I noted, real-time traffic and congestion updates.
Minimal utility to society? What about getting the half-blind geezers (like me) off the road, while still letting them engage in commerce? What about the handicapped, and giving them greater autonomy?
Sure, there's a very real concern about everybody having to have their own expensive rolling transportation device. But autonomous vehicles could be shared more conveniently than POVs. I mean actual SHARING, not "pay an inflated rate for service" like Uber, Lyft, etc. Among neighbors. Co-workers. Members of the same union. Kids going to school.
Okay, the cities will no doubt see the most usage. Fine. You want to enjoy the "rural farmer lifestyle", it'll cost ya. Don't let 1% of the population drive (heh) the way the other 75% lives.
I've left out safety concerns. It's a big problem. Perhaps limited access roadways, like the Interstates will be part of the solution.
I understand the argument that autonomous vehicles make sharing easier. I just don't see where a private owner has any particular incentive to do so. the incentives are different for Uber or Lyft, who save driver pay.
^^^they don't use less energy than present day cars; they are not going to be 100% safe.^^^
Nothing's 100% safe. But about 17,000 Americans die in car crashes every year. What if the eventual arrival of truly robust, perfected driverless technology lowers that number to 60?
You can reply: "It will never be that good!" And fair play -- you're entitled to your opinion. But that's a different argument (namely, the position that driverless technology can't be sufficiently refined or perfected). But what if we CAN perfect it and refine it some day, in this manner? Wouldn't a massive increase in safety AND the ability to forever be freed from the drudgery of traffic be a gigantic improvement? Also, and this is big: truly robust driverless technology will be a game-changer in that it solves the "last mile" problem wrt public transport.
I think Kevin's too optimistic in his time line projections. But I hope he's correct that this stuff is feasible. It'll be wonderful. I hope I live to see some of it.