> "This is one of the hardest problems we have. This is like we are going to Mars," Hitzinger said in a comment. "Maybe it will never happen."
First of all, it seems obvious that we are going to go to Mars, eventually. Maybe not any time soon, but never? Seriously?
But the bigger thing is that there is about 1000 times more economic benefit to self driving cars than of going to Mars, at least in the near term. To think we'd just give up on it seems absurd.
Exactly. Using Mars as a second example is basically just doubling down on the "Elon isn't trustworthy" argument. That should be a giant red flag to readers.
Not reaching Level 5 is not synonymous with "giving up." We could get to 4+, with 90+% of driving being automated, but never actually reach 100% (the definition of Level 5)
>First of all, it seems obvious that we are going to go to Mars, eventually. Maybe not any time soon, but never? Seriously?
Given the questionable economics underlying humans having a presence on Mars and the extreme toll on the human body this will have, foremost by radiation I think it's actually a fair comparison in particular because full autonomy like mars colonisation is constantly being overhyped mostly by a very small group of very affluent individuals who seem to be more inspired by sci-fi than engineering.
I don't mean to be rude, but I have strong feelings about this attitude.
Let's take going to Mars. We can't reliably go to the moon. We can't even go everywhere on the Earth, where we have every possible advantage. Spend a year on the ISS and you'll develop all sorts of health issues. Spending time on Mars isn't likely to be less hazardous. We may send humans to Mars but it is by no means guaranteed. (And I would also ask, what reason do we have to go to Mars? Probes do a better job of exploring, and I doubt we could colonize it.)
Until someone builds a real self-driving car it is just an idea. On today's roads, it is probably not possible to safely implement a fully self driving car. Driving is not only a technical exercise but a social activity that involves communicating your intentions to other humans, and interpreting the intentions of others. That is something that humans do far better than machines. (Edit: To clarify, humans are better at communicating with other humans. Machines do a great job of communicating with each other, but have mixed results communicating with humans.)
If every car was self-driving and the roads were remade from first principles, then sure, that seems feasible. The degenerate case here would be a self-driving train, which seems perfectly reasonable. But the technical challenges are the easy part of that endeavor. Funding such a project, developing the political will to see it through, and organizing the logistics are far more difficult. Consider for instance; what will happen to the legacy vehicles? Will it be illegal to drive them? Will there be a massive government buyback? Who will fund that? Where will the cars go? How will we organize the logistics of moving hundreds of millions of vehicles? How long will that take? What other matters will we need to turn our attention away from to accomplish that task?
A much more likely scenario is that companies continue to come up with partial, ad-hoc solutions, driving gets more automated, ride sharing becomes more popular and car ownership less so, but that humans remain in the loop for the foreseeable future. What happens outside the foreseeable future is something we can't and shouldn't pretend to know with any degree of certainty.
There are a million ways in which we could never go to Mars or build a self-driving car. We could get a better idea for how transit should work, making self driving cars superfluous. We could discover life on Mars, and make the decision that it would be too dangerous for us to visit. The superpowers of the world could go to war with each other, and our infrastructure could be devastated to the point where space travel is impossible. Climate change could drive us to extinction. Something could happen that we cannot predict or imagine, that we have no precedent for, that completely changes our situation and outlook.
Some of these are more likely than others, but the point is that it would be folly to take the future as read. And frankly, a couple of them are more likely than us ever going to Mars.
I think making a self driving car that is safe - they are billed as being safer than humans - is part of the problem definition. I could put a brick on my gas pedal and my car will drive autonomously, but I don't think anyone would be willing to give me credit for solving this problem. Clearly there is a threshold somewhere.
We absolutely have the technology to go to Mars. We could probably have sent a manned mission to Mars in the 70s with the same technology level that took us to the moon.
It’s just that it would cost an incredible sum of money for very little gain and so it hasn’t happened - even the mission to the moon was a huge drain on the budget and the program was canceled once we won the space race.
Going to Mars is like supersonic passenger air travel (Concorde) - something we have the technology to do, but it just doesn’t make sense.
An autonomous, self-driving automobile is something we just can’t do today no matter how much money we spent on it. I think it’s a different question because we are trying to predict if we will ever reach the level of technological sophistication to allow it. (Of course we have a solution for “driving a car from one city to another” and it’s putting a human in the seat, or of “making a self driving vehicle” and it’s putting the vehicle on grade separated tracks like the Morgantown PRT)
I disagree. I'm sure that, if we really wanted to, we could put a person onto Mars. I don't think we have the ability to send someone & retrieve them, or for them to fend for themselves on the surface. So this is a bit like saying we have the technology to explore the inside of a volcano because we can jump into one. It's not really the proposition people are talking about when they say "go to Mars".
But I agree that it's actually a generous comparison to self driving cars. After all, we do occasionally send machines to Mars, and they work quite well.
Read an book on space/aerospace engineering from the 1980s and look at how many things happened. Supersonic was a huge bust. (The Concorde is closer in time to wooden biplanes than to today. Routine supersonic travel still nowhere on the horizon.) Energy technology has been a huge disappointment. When I grew up, we were going to have nuclear fusion, expanding the scope of what society can do. Today we’re back to windmills and we’re told to put on a sweater. AI was of course a huge bust once.
Windmills' technology has improved dramatically, like also nuclear fusion's.
We are not "back" to windmills that give 3 MegaWatts of power because they never existed in the first place, like affordable solar panels with 20-40% efficiency.
Progress is not automatic. It is millions of times harder to create or improve a technology than imagine it. And also takes lots of money.
We have nuclear tech because Manhattan project, because WWII(and because they were scientists coming from Europe that were scare of Hitler). It took a tremendous amount of money and sacrifice to get there.
We have nuclear tech because Manhattan project, because WWII(and because they were scientists coming from Europe that were scare of Hitler). It took a tremendous amount of money and sacrifice to get there.
No, Manhattan project took trivial amount of money by today's standard. It costed less than the rounding error in typically given federal budget figures for past few decades.
Yes, and that's smaller than a rounding error in our current $4.7T budget. Hell, the $100B (today's dollars) Apollo program would also be smaller than the rounding error, since it was spread over a decade. The R&D costs on F-35 were almost triple the whole costs of Manhattan project.
Truly, we are easily able to afford projects like Manhattan and Apollo today, it's just we aren't able to pull them off anymore.
With the caveat: for commercial travel. And even then it’s partly due to noise restrictions. It’s not like the technology for supersonic flight isn’t widespread.
What list? The one where Thomas Watson said there’d never be a world-wide market for more than five computers? Other than that, what else does that list have to stack up against:
Nuclear power, let alone fusion. What happened to “too cheap to meter”?
Supersonic travel: raynier already pointed out what a bust the Concorde was. Nothing else on the horizon now.
Automation giving me a 20 hour work week. Nope, capital holders just skim that efficiency right into their pockets.
I’d go on, but suffice it to say that about the only thing much different than my childhood in the 70s are computers in our pockets. Revolutionary, no doubt, but we are still burning oil for our energy needs, our cars don’t fly, and I still show up and do my 40 hours. And healthcare has gotten worse in the U. S., not better, if you can believe that.
So, yeah, when the head of the autonomous driving division of VW says Level 5 ain’t gonna happen, I don’t immediately jump to doubting him/her and attacking their resume.
ALL of the autonomous driving experts say that level 5 is, at best, 50 years off. Unanimous. They don't know how to do it, see no path to doing it. It's hard! Maybe we'll make real AI and we can enslave our human-level intelligent computers to drive cars for us, at least until they figure out how to rebel. Maybe. But we don't know how today, and not tomorrow or the next day either.
And yet HN threads are all about "here are all these cool things that will happen tomorrow if Elon Jesus delivers".
In the dot com era there was a self driving car startup that started with a simplifying assumption: don’t run the cars at grade.
If you go up or down, the number and kind of obstacles reduces. The location of interactions between the vehicles is reduced, and the interactions with other classes of vehicle are zero, so you can negotiate.
I've read the paragraph several times and maybe it's my ESL, but I don't quite follow.
Is the term "Grade" here used in the "Slope" sense, as in don't run the cars up and down the hill?
And if that's the interpretation, I don't necessarily agree with the next point that obstacles are reduced on slopes/hills... so I probably am not following correctly :-/
If we put a wire in the road or maybe rfid tags every 40 feet, we could easily have self driving cars.
However that would mean that everyone would have access to it.
The idea of self driving cars now is, a winner take all situation where whichever funded effort that succeeds, generates outsized profits from licensing or going public at a high valuation.
A wire in the road can't handle stuff like a pedestrian. Pedestrians normally aren't in highways but sometimes are after break-downs, accidents, construction, etc. Lane-following is close to solved on the highway, it is all that other stuff and more that is an issue. A wire could still help, but you'd still need a complex system or significantly more infrastructure than just a wire (maybe caged barriers over the lane in a way that doesn't cause issues if there is a fire, and more).
Assuming you mean dedicated self-driving lane, what about when a non-self driving car crashes into the self-driving lane? Assuming you mean to have a dedicated lane with barrier (which is already much more than a wire). And what about when a large mining truck tire rim falls off an 18-wheeler and bounces over barrier in a way that any human driver would be able to hit the brakes and be ok?
The point is that (primarily) the road needs to be smart, not the car. There should ideally be a synergy between the car and road, but the road has to be the primary vector to guide the car.
This backfires badly in places with cold winter climates where the offical road markings and edges become obscured due to snow and ice for days and sometimes seasons (road edge creep) at a time.
Human drivers don't follow the official lane markings because they can't be seen. They follow the paths in the snow everyone else has packed down. These paths often diverge from the road markings or any sort of absolute positioning system.
Why? There's very little information that would be better coming from the road. To drive you want to know where you are, what the road surface conditions are like, where the road goes, and what else is on the road.
Road conditions can easily come from anywhere. Weather radar is at least good enough to know when roads might be wet or cold. Making roads smart enough to sense oil spills or even wetness would be incredibly hard.
Knowing where the road goes is certainly far better done by cellular. Connection to each segment of road would be fraught with hard to repair problems. Traffic conditions likewise are far better done from somewhere else, and cars would be much more able to see things on the road etc.
The only argument I can see as at all reasonable is that locating cars is difficult, and doing it with vision is incredibly challenging. You may not be aware how much GPS has improved. With a good view of the sky you can get (somewhat slow) accuracy to about a foot. Realistically that's just as good as you could possibly expect from a roadside device like RFID, bluetooth, or induction. The last inches may be important, but billions of dollars spent burying things in the road will not help.
To elaborate on this, camera based lane keeping systems are readily available and generally work well if the lane markings are in good enough condition and aren't obscured by snow.
A signal embedded in the pavement wouldn't be subject to wear, but it would make adjusting lanes much more difficult (if you've driven in the bay area, you've probably noticed lane lines moved back and forth for construction pretty regularly) and it would actually be worse in the snow/ice --- a consensus lane appears which may not follow the marked alignment, and following the marked alignment would involve driving over accumulated snow and ice.
There are thousands of miles of roads without lane markers, shoulders, signage. Some aren't even paved. Who's going to come along and bury RFID tags or wires?
I don’t think it will be winner takes all. I expect competitors to be relatively close, and may not even be clear who is the winner. One brand may do better in cities, another on highways, etc.
That will affect how outsized the profits will be.
Also, because of politics, I expect there to be separate geographical winners, at least in China and not-China.
Traffic inductive loop counter costs a couple thousands to install. I don't have a source for the detailed cost breakdown, but I can imagine the majority of the cost is labor and digging up/filling the road since the sensor itself should be less than a thousand.
Now imagine we're doing that 1000 times per mile, like what you're suggesting. Even if the device is free (both initial cost and maintenance) it's just too cost inhibited.
I think we shouldn't assume that the reason is some malicious intention behind "However that would mean that everyone would have access to it." It could simply be that the idea is not economical.
> I don't have a source for the detailed cost breakdown, but I can imagine the majority of the cost is labor and digging up/filling the road since the sensor itself should be less than a thousand.
Those are installed with a massive handtruck saw while cutting off a lane of traffic all day. It takes a half dozen people working on-and-off, doing different jobs. It takes jackhammers, high pressure water from a tanker truck, and people to place and wire the sensor to power. It's way more complicated than anything you'd do at scale, because it's infrequent and the goal is to do it thoroughly with non-specialists.
If you were installing millions of tags you'd have a dry drill that could go off the back of a truck and place+fill a tag in ten minutes. If you had a line you'd have a it hanging off the back of a truck and place it continuously, like a street cleaner or edge clearer. For tags there's no reason they'd need more than one person to place and no reason to even put them in the road when they could just go on the edges. Triangulate with directional antennas or something.
That said, I think it's pretty obvious that locating the roads is by far the easiest problem for self driving. If you wanted to make a serious attempt you'd want every car to broadcast a short range location, and to share data over a mesh network. "Knowing where the road is" to precision RFID would give you has been solved for over a decade with GPS and digital maps.
Much more pressing issues are non-obvious sensing like hearing a car around a blind corner or knowing when to be cautious about moving. Knowing when something is coming onto the road or when a vehicle is having a problem. Inter-vehicle communication is just so obviously important to that... it's really frustrating how vaguely it gets talked about. I don't give a shit about teslas coordinating braking so they can form a tailgate train for efficiency, I want cars of all kinds to be warning each other about what they intend to do. I worry that legislation or at least a regulatory body will be the only way to even get people talking about it seriously.
Other than that, cameras watching for intrusion into a road would be easier than solving it from vehicles. It seems patently ridiculous putting cameras to watch every 50' section of road. 1080p+ cameras, simple detection, and mesh wifi can be built in a $30 package... but there are >2.5 million miles of paved roads in the US. 30$ per 50' would cost, bare minimum (and ignoring electricity requirements+labor+the pole to put the cameras on) 8 billion dollars.
Creating a local mesh network for cars to communicate is an idea I've had for a while. Something like an open hardware and software stack/protocol for future autonomous cars. It's an inevitable thing that must exist to realize the full potential of autonomous cars, all a question if I'd be the one to build it :).
Would be curious to know the current work in this area.
If I were to put wires in roads, it would be for wireless charging, so that autonomous taxis and delivery vehicles could keep driving forever (or at least until they require maintenance) without needing to stop to charge.
I can see this happening gradually, beginning with docking stations, followed by parking spots with wireless charges and then downtown roads and highways. With each step reducing the required amount of space destined to parking.
This doesn't work in climates where the road is obscured for days at a time due to snow and ice.
The tags, whatever form they may take, will designate the official lanes. But people in snowy climates don't drive in the official lines because they can't be seen. They drive an emergent set of paths where everyone else drives. It's often, if not most of the time, that these paths human drives take don't follow the actual road markings.
Despite that, he's confident in VW's ability to make a Level 4 autonomous vehicle, saying that the upcoming I.D. Buzz electric van will be the first VW to receive the technology.
He's poo-pooing level 5 autonomous driving and says that they just about have a level 4 autonomous vehicle?
This article makes no sense.
Honestly, the CEO of my company has very little idea of the details of the technology that we produce. If you picked some cutting-edge technology that isn't key to our market-share yet, he'd have even less of a clue.
But it's really not. Level 4 will require the vehicle to take you from door to door without driver intervention in normal driving conditions. Level 4 is mostly what people think of when they think of autonomous driving because you can take a nap, work on your laptop, whatever. You get in your car, tell it where you want to go, and it does everything.
Level 5 basically removes the steering wheel so you never drive. But once you're at level 4, almost all of the hard problems have been solved.
Here's a summary of his entire background in order: Formula 1 engineer for many years; Technical director for a Le Mans Prototype vehicle; Head of VWs self driving division for the past 1 year.
He has no software background... he specializes in making gas cars go fast. He has no idea how to make a self driving car.. and honestly, why would we think he would? He hasn't exactly been pioneering vision systems, or anything else related to the tech.
Yeah, because making Formula 1 cars "go fast" involves no software whatsoever, right?
I love how you dismiss "Formula 1 engineer for many years". Maybe if he helped build a payment processor we could take him seriously.
>He has no idea how to make a self driving car.. and honestly, why would we think he would?
Because he's running the autonomous car division of the biggest automaker in the world? I have no idea if he's right or not, but dismissing him as not knowing what he's talking about is hubris. He's trying to build this stuff.
So much ignorance and inaccuracies in one comment.
(a) He was the Head of F1 Development and Advanced Technologies which involves significant software exposure. F1 has a ridiculous amount of software both on car and in the factory to process all of the sensor data and design future cars. I've first hand seen their streaming big data stacks and they are on par with anything you will see at a top tier startup.
(b) He was the Head of Product Design for Apple's Car and we know they were going to have autonomous capabilities as well as hundreds of ML models powering AR, Maps, facial recognition etc. If you think you can get away without deeply understanding software whilst leading a major Apple project then you really don't understand how that company works.
(c) He's the CEO of VW Autonomy. He doesn't even need to be an expert in software. He just needs to be able to listen to his engineers.
If Bezos started a book company, then failed and admitted he's not sure if it's possible to sell books... then I would question whether or not he has the background to sell books... especially if I look around, and other experts in the field don't agree.
My eyes looked at this comment, see that it's totally true, but yet downvotes. Context. But yes, there's an absolute truth to someone that can't think outside the box and deny something is possible vs the people that make it happen.
It's called a sceptic. Almost every great true visionary had them. I'm no Musk fanboy but he's kinda like the modern day epitome of everyone saying no this isn't possible and it happened anyway. And to add insult to injury he started a rocket company kinda simultaneously. There absolutely is truth to scepticism literally blocking the mind. If you will it enough it can happen is a thing. If we all just put our heads down on how complicated that would be, I don't think we'd be very far.
Consider the Osborne effect. And consider that this person has a strong vested interest in making sure potential VW customers does not decide to delay their car purchase decision in the hope of getting a more automated solution next year or the year after.
Maybe he's right, maybe he's not. But either way it'd be dumb of him to claim level 5 autonomous cars were right around the corner.
Agree completely. It would be suicidal (in a business sense) to say that Level 5 was right around the corner right when you're on the verge of rolling out Level 4.
> Apple is in the top tier of self driving cars though.
I mean all we have to support that they're working on self driving cars at all is DMV records, articles, and pictures of their car efforts. We have no idea how far along they are.
This is arguably one of the biggest competitive advantages Tesla has. For $47,000 I can get an EV from Tesla with a better self-driving system[1] than a $100k Audi, BMW, Mercedes, or Lexus. I haven’t pulled the trigger on a Tesla yet, but I can’t imagine paying $40,000+ on a new car that isn’t a Tesla right now. For many people Tesla has made EV and self-driving table-stakes and nobody else is delivering on that.
So, considering the accidents, maybe Tesla is "better" at self-driving in the same way Intel was better at speculative execution before they had to patch it.
The only company that knows what they are doing is Waymo (which actually launched self-driving taxi service in Phoenix), Tesla and others are kind of "fake it until you make it" crowd.
Waymo has 600 cars in a very limited/ well known area and most of their cars still rely on safety drivers. If Waymo is around in 20 years their service might get to the small-ish city I live in.
The big thing I would want self-driving capabilities for is long cross country driving. A thing Waymo doesn’t offer even to people in it’s service area.
Tesla has hundreds of thousands of cars which range over the entire country and has driven millions of miles doing exactly the sort of self-driving I most care about.
The two barely overlap in terms of offerings so I’m not sure what the point of comparing them even is.
>>I can’t imagine paying $40,000+ on a new car that isn’t a Tesla right now
It's personal & subjective, and nobody will persuade anybody in a thread, but FWIW I absolutely positively can.
- I'm unlikely to utilize self-driving any time soon
- The stories of their firmware terrify me. And I don't want a DAW and computer games running on my ECU :O
- More pragmatically though, the Tesla UI paradigm is completely foreign to my way of driving/thinking.
I'm looking for a "HOTAS" type UI, where I can do anything I want without taking my eyes and focus off the road. A UI that's one giant screen, that may change position of buttons from minor firmware to another, is basically as scary and alienating concept as I can imagine.
I get that I am in a minority nowadays - a lot of manufacturer's are replacing switches, buttons and knobs with a touchscreen and deep menus, but not thankfully all just yet :|
Tesla's increasing have voice commands for most functions. What you said was true before but isn't now. Honestly this feature, being able to push updates to the entire fleet when ready is a significant killer feature.
Again, at the risk of sounding like a grouchy old man (which I suppose I am :P), when I'm chatting with my wife and kids, I have limited intetion of interrupting that conversation to turn on the lights or seat warmer or wipers or change a song :-/. And even when alone - the amount of time and subconscious effort to hit a button, vs converse with the computer...
Honestly - these shouldn't be "either or". Sure, have a screen and voice for those who prefer - but leave a button or six for us ol' timers :-)
Everything you need at least once week while driving in a car should be doable without taking your eyes from the road for more than a second. Radio, AC, window heating and such fall under this category.
And you have exactly the same amount of buttons on a screen anyway, they're just not physical.
Next, Previous, Pause, Play, Mute - I personally want them to be physical controls. On steering wheel ideally, on the dashboard otherwise. I use them multiple times a drive, and I don't want to take my eyes off the road to do them.
Same with seat heat, lights, wipers - anything I may want to do while driving, I want a button.
Setting up the exact shade of my dashboard light - that can be buried in a menu :D
Basically... when you say "Crazy how many buttons my normal car has" :
- You say that as a bad thing
- I see that as a brilliant thing... IFF done well:
Of course, physical buttons/levers/knobs can still be done well, or poorly.
Having many identical buttons in a confusing layout is just as bad as touchscreen - I have to look at them to use them.
Having buttons in a good, intuitive layout; especially buttons which are distinct from each other, as opposed to row of 6 buttons all the same, is brilliant. Even better if it's a distinct combination of buttons, knobs, switches, levers, etc - anything to help haptic feedback and intuitive access. Sometimes I think people who are against buttons may simply never had a car with good physical UI:/
(simple thing - my old 2004 WRX has a next / previous knob-like-thing, rather than two identical buttons next to each other [1]. It felt ridiculous when I first saw it - but then I realized its quality of purpose vs sexiness - I never ever ever have to think or be distracted even a millisecond to know exactly how to skip a song :). Compare to cars which have several identical square buttons for next, previous, pause, play; or temp up, temp down, fan up, fan down, A/C -- that's just horrible UI by clueless people for customers who don't know / haven't experienced better :-/ ]
> I'm looking for a "HOTAS" type UI, where I can do anything I want without taking my eyes and focus off the road.
I agree, the cockpit design on the Tesla is not my choice and may actually be a deal breaker. But since there is a 30 day no commitment period I’m willing to give it a shot and see how it works in actual use.
But I feel a lot of it is:
Sexiness of how it looks
vs
Practicality of usage
The problem is, sexiness is immediately obvious & attractive. Practicality takes a long time and active observation to notice and appreciate. I fear that touchscreens with bad UI will win; though I hope eventually there may be some backlash from consumers - or at least thought and compromise from manufacturers. :-/
Is Tesla following accepted industry best practices for designing their algorithms? If not, Tesla is exposing itself to a tremendous amount of liability when their autonomous system fails(and, as software engineers, we know it will fail). I applaud them for pushing the technological envelope and driving expectations. I'm not criticizing them for the damage those failures will cause, because it may be offset by reductions in human caused damage. I'm saying there is established legal precedent that will cost them dearly if they're not properly developing and testing their code(ISO26262, MISRA, etc). See the Toyota accelerator debacle for background.
>I can’t imagine paying $40,000+ on a new car that isn’t a Tesla right now
Aside from the fact that the Tesla is not available where I live, if I was only going to own one car, it would not be a Tesla. A Tesla (or any other EV, really) does about 80% of what I need a car for, so it would be perfect to buy as a second car for me, to be used most of the time and have the other car be a backup for when I really need to.
If I had to settle for only one car, it would probably have to be something like a Chevy Volt, which I am quite sad got discontinued. Even then, that would get me about 95% of the way there.
I wouldn’t want a loved one driving a Tesla with autopilot enabled. Some of the people who have died so far have been tech enthusiasts who knew the limitations of autopilot, yet they still grew complacent and it cost them their lives.
If my non techy family see the Tesla performing what they deem to be self driving then they will quickly trust it 100%.
Level 5 means no human intervention, under any circumstance. Ice, snow, night, fog, unmarked roads, tunnels, mega-urban, isolated rural, parking lots/structures, etc, etc. I think he's correct. It's hard to think of any technology that works without human intervention whatsoever, no matter the conditions.
We've had autopilot for decades, we're certainly not flying planes without pilots. Or even getting them from runway to gate without humans. Nor is anyone claiming that pilot-less planes are coming soon.
Planes are different in that there is a large ratio of passengers to "drivers". There isn't the same economic incentive to eliminate pilots when the cost of flying a plane is still going to be quite a few dollars a minute.
The thing is, there are some conditions where people probably just shouldn't be driving. Like we shouldn't be driving in a heavy snow storm, at least not for most trips where you can just wait it out. Planes will wait out storms, even with human pilots. I don't see a problem with "level 4.9" cars that occasionally say "I'm going to wait out the storm" or that will avoid certain routes (such as "crazy mixing bowl" intersections where you can instead just take a side street)
How much do pilots actually pilot vs how much do they just assume responsibility? I.e. I as a passenger feel much safer knowing that there's someone actually responsible to prevent me from dying (because if the plane crashes, the pilot dies with me), not just some C_O suits pushing PR through legal trying to gaslight the public and avoid any culpability (cough Boeing cough).
That's exactly the point, even if pilots use autopilot most of the time (level 4 autonomy).
Guess when they take control? When something bad is happening (and there are literally a million things that can go wrong and the autopilot could not fix, ever).
Guess why MCAS did not result in more crashes (the system has been known to malfunction a couple of times before the crashes)? Righty because there were pilots that could manually control the plane well enough to land safely.
There will always be a facility to manually override the decisions of the autopilot. It’s likely to be a legal requirement so that in an emergency the vehicles can be directed to do things they normally wouldn’t (like drive on the wrong side of the road). But you don’t need a steering wheel to do that.
I imagine there’ll also be manual modes to get the car into trucks and ships, or even just out of the factory.
This is definitely not consistent with statements Alex has made in the past. Seems like more of an off the cuff remark a German engineer would make while confident that the fully realized result is right around the corner.
Opinion: level 5 autonomy would be cool, but it's not where most of the value of self-driving cars is.
Most of the value in being able to travel by auto without attention is in trips longer than 15 minutes. Commutes, tourism, vacations. Shipping. Most of which is certainly highway driving.
If it's possible to automate highway driving under most conditions (and safely transition either to a stop or human-piloting when those conditions aren't met), then at least 80% of the value is there.
Wrestling with the harder edges of the problem is still the right thing to do for tolerance reasons, but I hope we don't have to see last-mile problems solved before we start reaping the benefits.
This 80% solution already exists in the form of trains and airplanes.
Self-driving cars that only go 80% of the way would compete against those forms of transportation which have established business models, networks and in the case of trains are subsidized in a way that self-driving cars likely won't ever be.
Additionally the key advantage of cars is flexibility, not having to figure out how to get to or from the train station or airport. In your scenario self-driving cars would lack this advantage.
Now granted this may still sound like an appealing proposition in countries with poor public transportation infrastructure but that substantially reduces the size of the overall market. Sure this might be attractive for the US but what about Europe? Asia?
Now of course there is an obvious answer to this problem: Keep the steering wheel and drive the rest of the way yourself. That just means you'll have a lot of drivers who won't gain experience at the current rate though. Not sure I would like to share the road with those people.
Yes, the tremendous capital influx into AV could probably support advances in trains and rail infrastructure. Yet, we will chase after this idea — it receives copious media attention because it is technically challenging and is “sci-fi” idealism that is hard to mitigate and feels like a good idea for society. Also, the investments seem like chasing after what everyone else is chasing after (FOMO); an automaker would find it difficult to “pivot” from vehicle production to advancing rail (competition) so capital will be stuck in that industry.
It seems to me that the problems that trains and rail infrastructure is facing are: bureaucracy, politics and an increasing inability to execute on infrastructure projects on time/budget by governments.
To the extent cost is a a factor, it's unnecessarily high due to these aforementioned issues[1]. So I don't think money invested into self-driving cars hurts trains and rail infrastructure.
Lots [1][2]. These have seen most deployment in closed networks (e.g. subways, or remote mining railways [3]), but experiments on "normal" interconnected networks are ongoing.
I think self-parking (as in “go find a space, possibly auto-paying”, not auto-parallel-parking) could also be a game-changer, not only for convenience and cost, but even zoning and city planning.
I have to disagree. Most of the value is in taxi services, where you get dropped off / picked up exactly where you want, there is no need to dedicate so much land and structures to parking, there is no need to have cars sit idle for 95% of their existence, and no need to search for parking spots and walk long distances.
Of course, I live in a city, where parking is a major pain. But the other things still hold. It it far more efficient of resources to have cars be used most fo the day, rather than sitting idle.
I actually think autonomous taxi services will be niche.
Think about Uber cars today. Their entire value proposition centres around them breaking parking laws in order to pick you up. Now the liability for that behaviour is with the driver. In an autonomous car the liability would be Uber which is why they won't do it.
An Uber that needs to look for parking in order to pick you up is never going to work in most busy cities. So I suspect you will see people opt for the driver based taxi service instead. Or maybe autonomous taxis will be reserved for emerging markets e.g. Indonesia which are more relaxed about parking.
I don't doubt that things take time to adjust. But when it makes a lot more economic sense to take a cab compared to owning a car and driving it (after all, the main reasons cabs aren't all that economical today is the requirement of paying a driver while you just sit there), there will be less cars parked on the road and therefore it will be easier to have places where the cars can pull over and pick up / drop off.
Anyway, "people are willing to break the law but companies aren't" seems a really weird reason to stick with the status quo over something far more efficient.
I wonder though if autonomous taxi's would simply highlight the parking issues and increase demand for better parking solutions (like no parking, just autonomous taxi lanes).
Are parking laws different for Taxis than for Ubers? Say in New York City? Lots of Taxis there so if Ubers are working under the same law, seems like an already solved problem. UPS have millions? of parking violations per year, yet it is still in operation.
Well, no company is going to be able to put in writing "Our plan is to have cars park illegally." Any system that gets built will be constrained by actually having to be law-abiding, which Uber drivers are not constrained by.
So the government is going to allow individuals to do it, but not the self driving cars? I don't see it working that way.
It's just like self driving cars will go a bit over the speed limit if that is the flow of traffic. If they don't, then the laws will be adjusted. It would be silly to indefinitely hold back a huge industry because humans get to exploit legal loopholes but machines don't.
I think Uber sells the cars to John Doe who registers it in their name then let’s their car take Uber rides. Uber is off the hook for both the illegal parking and the capex.
Think about Uber cars today. Their entire value proposition centres around them breaking parking laws in order to pick you up.
I take Uber and Lyft in NYC multiple times per week. No one ever “parks”, they just pull up, you jump in, and off you go. Takes less than a minute in most cases, doesn’t block traffic, and I suspect it virtually never results in a ticket. Maybe if they sit there for 10 minutes or more, but that’s exceedingly rare. And even then, they’re not going to get a ticket outside a few ultra-congested areas of manhattan that make up a tiny fraction of this city. Even in most places in manhattan, you could sit in a running car in a no parking zone for hours without getting a ticket. I had a moving truck up on the curb on the Upper West Side in a no parking zone for hours and the NYPD rolled past several times without a glance.
Also, this is exactly how taxis all work too, btw, which seems to undermine your entire point.
Umm, that is is point though - with autonomy the cars presumably won't break the law and stop where it's technically illegal for 30s to pick you up. They will presumably need to find somewhere legal to park to do so.
Anecdotally, I've often had Ubers pick me up in specifically marked and designated passenger loading zones when and where that's been appropriate and necessary. And in principle it's really no different from picking up a friend if you're giving them a ride somewhere.
There are explicit laws in NYC to allow this behaviour which is understandable given the power of the taxi lobby and history of the city.
Similar laws simply aren't present in most other cities. Stopping in the middle of the road to pick up passengers is illegal and drivers do it since the police are never there. Hence my point that Uber would never program this illegality into their self driving algorithm.
Autonomous cars won't have the same algorithm as selfish humans to position themselves.
A human will go to one of the busy places and will probably drive back to the original point of pickup.
While a computer will be notified ( 1 car only parked on the busy place) and immediately replaced by a new available/nearby car. Making the process more efficient, more remote places will be handled faster. Busy places more efficient.
Not all cabs will need to earn money, since the biggest expense will be the car driving and not the human driver ( waiting for work).
And as always, if income > fine. They will just pay the fine, some companies even negotiate it because of the sheer amount.
> And as always, if income > fine. They will just pay the fine
Up until the point they are banned.
Everyone, especially lawmakers, are pretty sick of tech companies breaking the law in order to be "disruptive". As we've seen recently with all of the e-scooter bans.
You won't get banned for paying too much parking tickets.
Scooter bans were mostly because it was disruptive having a staple of scooters in the middle of the pedestrian road and they were consistently breaking the rules.
Ps.
We do similar stuff for cities at our company. So I don't need an externzl source for this currently.
This isn't much a problem in terms of self-driving, but solutions to this aren't hard; eg. someone could use the app before they enter a self-driving taxi to report that the car has stains, trash, food, etc. left in it. The company behind it could then queue it to be cleaned.
Well, considering that these won't allow anonymous usage this can be mitigated by kicking out "bad" users and big fines. Also cars can visit the cleaning zone frequently.
For "free floating car sharing" (DriveNow/ShareNow, Yandex.Drive, ...) this seems to work quite well.
If the taxi is soiled then the next passenger indicates this via the app. The taxi is sent for cleaning and the previous passenger is charged for the cleaning. Also, there will be cameras inside in case it has to be proven who made the car dirty.
If people have to pay the cleaning bill they will be careful not to make a mess.
> If the taxi is soiled then the next passenger indicates this via the app
So riders blame the mess on the previous passenger. Then the taxi company installs cameras. It’s all gross (and yes, I’m aware many taxis have cameras already).
The margins on taxi services are so incredibly thin I find it difficult to believe there is any value in autonomous taxis. Especially when you take into account all the edge cases that will have to be accounted for, and the traffic that will be generated. Shipping and trucking are a massive industry, and I think automating highway driving has a much much better cost/benefit. Not to mention being safer.
> It it far more efficient of resources to have cars be used most fo the day, rather than sitting idle.
This is better and more easily addressed by car sharing services like car2go than autonomous cars. The more I use them the more I wish for a world where every car was shareable: you need to go somewhere? you simply get into the nearest parked car and drive it to where you want to go. And, unlike autonomy, there are no technical barriers to this, only the will to do it and the economic model to sustain it. Sadly these services seem to be in full retreat now but I will miss them and still hope they'll rise again soon.
> This is better and more easily addressed by car sharing services like car2go than autonomous cars.
This is not the conclusion I would draw from car2go failing out of North America. They still had the problem that you are required to find a legal parking spot and there's no guarantee that there will be a car available anywhere near you. Taxis and private vehicles solve the latter and former respectively but they all suffer from the inefficiencies inherent to the system design.
In the end the availability comes down to fleet size. car2go worked best here when they blanketed Seattle with Smart Cars, the fleet peaked around 750 and they were easy to find. After they switched to Mercedes exclusively the fleet was smaller and it became harder (though not impossible) to find them close by. My guess is that 2,500 cars would be enough to make them feel ubiquitous. A small fleet of autonomous cars won't mean it's likely one will be available to pick you up, and nobody has figured out how to deploy & maintain them at scale yet (I think Waymo in Phoenix is only running around a hundred vehicles).
The catch is the cost of parking: they weren’t hard to find here in DC but if you’re seeing lots of available cars nearby that means they’re paying a fair amount for all of that unused capacity. The Uber/Lyft model shifts that cost to the drivers so it looks cheaper as long as you don’t factor in congestion.
BYO. Which is borderline OK for boosters (the Mifold [1] weighs under 2 pounds and folds to fit in a handbag), but totally impractical for a baby/child seat.
Well you still need to park them somewhere, but yeah you could have less cars. And they don't work if you can't drive (or are drunk). And it is more complicated if you need one with a car seat or other special accommodations.
A car also functions as a usually-secure place to store a good chunk of stuff, rather than taking it with you. This is more relevant if you have children (carseats) or take multiple-destination trips.
This is such an under-mentioned point. Cars are fantastic for logistics and storage. Though I imagine if we can automate people-moving, we can figure out logistics bots as well.
There's a fantastic amount of value in cities and urban areas. Self-driving cars can work as a massive network to reduce or eliminate congestion and give people back hundreds of millions of hours spent stuck in traffic.
They can also, if Level 5 becomes a thing, reduce traffic collisions. And if they all end up electric, through route optimization they can significantly reduce energy consumption requirements for vehicle usage.
How would they "reduce or eliminate congestion"? Congestion is inherently caused by too many people trying to go to or through the same place at the same time. That's not possible without reducing the amount of people trying to travel during a given time period in a given place.
Given the same number of cars on the road, a self-driving fleet would eliminate many mechanical failures and many human failures, and be able to adapt to sub-optimal infrastructure at a network-wide level.
Infrastructure would also be easier or more efficient to improve, because you'd have removed much of the human variability that makes identifying choke points difficult.
The problem is that in many cases, many people want to get off at the same freeway exits, and the local network doesn't have the road capacity to match.
AI doesn't really have the power to change that, and might actually be worse depending on how it reacts to pedestrians and cyclists.
By offering laminar rather than turbulent flow. Take a two lane highway where traffic moves at 100 kph. It contracts to one lane at 100kph. Humans will pull up close, stall in one lane, do a flaky and awful zipper merge. Latency will be higher than necessary and throughout lower.
Self driving cars can plan the whole exchange, negotiate (or be directed by an intersection controller) to decelerate, align, merge, and maintain high throughout at low latency. ATC does this for planes already; we have the coordination technology. It’s just about getting drivers to listen to precise instructions without deviation. For everywhere I know of, that’s going to take automation.
Reading up on standing waves taught me a lot; you might like adding that wrinkle to your model.
There are other actors on the road, namely pedestrians and cyclists. Unless we're going to remove the right of citizens to the street at all, they will be present and will need to be accommodated for.
My point is that unless you get rid of them completely, traffic flow is not going to increase. You can't have free-flowing traffic everywhere and also have pedestrians safely crossing the street, and the lower bound on a traffic light cycle today is the amount of time it takes an old person to cross the street from curb to curb.
That neglects limited access roads and wide boulevards. In either case, even if pedestrians aren’t banned, they are rare enough to be an exception rather than a rule. For example, the old guy trying to make their way across 8-lane Chang-an Road in Beijing...well, there is a reason Chang an road forces pedestrians onto bridges.
By and large, today the congestion problem is not really on the massive roads themselves, but where they interface with low-capacity local networks. Most destinations are not located on a limited access road by definition, and the busy exit into city center with lots of pedestrians is not going to get any less congested, AVs or not.
In modern Western urban planning, large car-capacity streets aren't considered a good thing to be running through urban areas anyways. Most of the large cities (Paris, Berlin, New York) are taking steps to reduce road widths to more equally distribute the road between pedestrians, cyclists and transit users. As it turns out, building pedestrian overpasses at a convenient enough spacing for pedestrians is extremely expensive.
It's hard for me to guess how much something like this would reduce delays caused by traffic compared to the increased number of cars driving due to the fact that some driverless cars would be out with no passengers heading to pick people up.
Self-driving cars cannot plan an entire exchange in a dense urban areas because there will always be actors like pedestrians, cyclists, and stray plastic bags that do not participate in planning and act adversarially to their shared model. Optimizations will be highly local, spatially and temporally, and I suspect they will end up looking a lot like humans trying to coordinate on the same problem. And even if AVs can technically plan and execute faster, their actions will need to be artificially slowed to be legible to humans. Likewise with their raw speed — cars, self-driving or not, are already moving too fast in urban areas. Reaction times might improve but braking distances will not.
So if AVs can't increase the throughput of city streets, I'm skeptical that they can increase the throughput of off-ramps which are bounded by city streets, or urban freeway segments which are bounded by the off-ramps. And even imagining that significant (2x?) throughput is achieved, it's not going to meet the induced demand ceiling; there would be the same amount of congestion, only with more cars.
L5 is dead on arrival as congestion-mitigation technology and I hope at least some of the billions earmarked to be spent on researching and deploying it are redirected towards better walking, cycling, and transit amenities instead.
Self-driving car networks have the benefit of not being bound by the same finite capacity of a network of humans trying to manage real-world traffic.
Neither would they have to operate at artificially handicapped speeds, as no human operator would really be able to keep up regardless with the 10,000 or 100,000 cars running around.
Part of reacting and mitigating acute issues is being able to reroute traffic early enough, or slow cars down early enough, to minimize disruption (instead of a traffic slinky blocking everyone up).
There's no reason so far here that self-driving vehicles can't improve traffic and overall throughput, and it's a bold claim to declare L5 dead on arrival.
In a dense urban area, the speed limit is generally not dictated by road design but by the simple physics interaction between a person made of meat and bones, and a multi-ton metal vehicle.
Follow distances in urban areas with low speeds are already really small. I don't think AVs would make them meaningfully tighter.
> Part of reacting and mitigating acute issues is being able to reroute traffic early enough, or slow cars down early enough, to minimize disruption (instead of a traffic slinky blocking everyone up).
This was supposed to be what Waze and Google Maps were set to achieve, and they did not, but they did piss off a lot of people when cut-through drivers started jamming up their formerly quiet residential neighborhoods.
When roads belong exclusively to self-driving cars, they can increase speed, reduce gaps between cars and eliminate traffic wave effect.
Imagine highway where everyone drives 80mph with 5ft of distance between cars. No one suddenly changes lanes, no one suddenly slows down for no reason.
Even if there is no reaction time, 5' of space at 50mph is silly. You still need room to maneuver if whatever caused the vehicle ahead to throw on its breaks comes back or halts the vehicle ahead faster than your breaks can stop you.
It’s basically a virtual tandem semi. The presumption is that you are delegating command and control to the lead vehicle, and there is bidirectional communication between the lead and follow vehicles.
Well, why don't we go a step further then and physically connect the vehicles? Maybe we can also make them longer and fit more passengers? And we could optimize the rolling friction, replace rubber on tarmac with something better like steel on steel, and use some sort of tracks to provide the sideways forces?
You jest, but I'd love it if we could automatically group cars into physically connected trains for long distance travel, and have them automatically split off when they approach their destination. Maybe there's some way to do that with rails, carrying peoples' cars, but not sure if the efficiency gain is worth the increased complexity of including another set of wheels/motors/etc, and making the automated separation/joining work with all of that.
Lol that is going to lead to the mother of all pile-ups not if but when something just so slightly goes wrong. Imagine somebody dropping shit from an overpass or really anything. Better build a giant cage over all the highways!
That's because you're thinking of human drivers whom are so slow to react that it is recommended they keep a whopping 2 or 3 seconds [0] worth of trailing distance to respond to changing road conditions like the person in front of you slamming on their breaks. The hypothetical clustering of self driving cars requires them to react at much faster times so they don't need to keep as much space between them.
Really looking forward to one day being able to sit im a self-driving car that goes at 150+ km/h towards a crowded intersection with other cars crossing left and right at equal speeds, and then just pass the intersection through a tiny gap in the wall of crossing cars. Because all those cars talk to each other and can precisely co-ordinate to create the nessesary gaps at just the right moment.
Local streets will always have pedestrians and cyclists. The main limit on road capacity is generally where highways interface with the local network, and the local network itself.
When I drive to work there is an empty transit lane for cars with 2+ people which seems to imply that most rush hour cars have just the driver in. If everyone is ubering in robot cars then they’ll probably share. It’s not that weird as people share public transport and choosing between $10 to get to work and $3 is a no brainer.
They might even Uber to the train station then Uber from the destination train station to work. Lots of ways of cutting traffic!
Values change. No one would take a taxi where the driver is just some unlicensed taxi driver. Back in the 90's that would be consider akin to hitchhiking in danger! Now we have Uber.
Also I drive because it's faster, and because once you go from A->B getting to B->C is probably easier. Not for privacy. Most people just need to get to work or school or their kids to school for the bulk of their car use. Kids to school might work by sharing with other parents you know.
Cars are different than a lot of goods though, because in and of themselves they are a wealth signal (at least in the US).
Most Americans don't really need an SUV or a pickup; a minivan or a sedan are more practical. Yet most Americans are buying them, because it makes them feel better about what they're driving and they want to show off. You may only take your car to the grocery store, but if you meet Becky from PTA in the parking lot and she has a really nice car that she owns, that's still a big deal.
Unless the ownership cost of a car were to spike significantly, this part of American culture seems hard to change, particularly when automakers spend lots of money making sure that Americans see premium gas-guzzlers as status symbols.
Self-driving cars will of course increase congestion because it will be possible just to chill out and do something else while car is driving itself. In Silicon Valley normal story will be “I worked in the car for 3 hours while commuting to the office”.
Work also needs to be decoupled from inner city offices (so you don't have to go into the city/office/etc) and local and regional legislation needs to be enacted in concert to lower overall personally operated vehicle density.
> If it's possible to automate highway driving under most conditions (and safely transition either to a stop or human-piloting when those conditions aren't met), then at least 80% of the value is there.
Given that planes, which rely on a class of specifically trained professionals with minimum hour requirements, still have issues with this from time to time, I don't think this is a realistic thing for cars.
Driving standards, at least in the US, are very low, and raising them is a hard problem.
> Given that planes, which rely on a class of specifically trained professionals with minimum hour requirements, still have issues with this from time to time
Most of this issues come from the fact that planes can't just stop in the middle of the road and wait for help, they have to land somewhere and do it fast enough not to be out of fuel.
Cars don't have to drive themselves to nearest service center when they encounter any hardware malfunction, they can just stop and unload passengers. Self-driving company would use another car to drive you to your destination (not even necessary self-driving car).
And yet air travel is far safer than road travel and they embraced autopilot long ago.
The problem with complete automation is how accidents are perceived. Probably a good thing considering how low the bar is right now for road safety. Being killed by a robot's mistake is perceived far worse than by human error.
I think a middle ground like autopilot on the highway could significantly improve safety while still accommodating perception issues.
>The problem with complete automation is how accidents are perceived
I don't think so. Airplane autopilot automates away the safe, boring parts of air travel, much like highway driving, so in that sense I agree with you. Truly autonomous landing is still a big deal, and planes that the benefit of ATC, and I don't think there are any planes that taxi or takeoff autonomously. Furthermore, accidents involving autopilot are always at the edge, when control changes from computer to human unexpectedly or when the human has to take over the autopilot in less than ideal conditions. I don't think there's much perception in the public eye of "robot pilots killing people"
I think once autonomous driving accidents go through the courts a few times, the same will happen to cars. After another decade of development and legislation, people will die not because the robot made a mistake per-se, but because it encountered a situation it can't cope with and the human wasn't ready to take over in an emergency, or because the robot tried to cope with the emergency situation the wrong way and the human didn't notice in time and took over too late (if at all). This may still sound like the computer making a mistake, but the subtle difference is that the mistake is made in an extraordinary situation, as opposed to the current accidents that happened in regular driving conditions. I think this subtle distinction will be enough for the PR people to spin the blame away from the manufacturer.
Call my cynical, but I can also see successful lobbying efforts in the future by corporations to reduce liability in class-4 and lower autonomous car accidents involving unusual situations, citing the fact that the driver should have been paying attention and taken over.
That aside, road safety is better than ever these days, with a lot of what would have been fatal accidents being merely property damage due to modern safety systems, and highway driving is probably the safest kind of driving out there, thus it is not clear to me (and I have no data either way, I don't think a study has ever been done) that current offerings (Tesla et al.) are any safer compared to similarly priced modern luxury cars per mile driven, and I'm not sure that will change in the future. The only reason I see for adopting highway autopilot is not safety but comfort. You might argue that comfort contributes to safety and enables longer driving times, but driving for 8 to 12 hours is pretty safe regardless, and I don't think I could do more than that even if I was a passenger.
This is a bit of a controversial opinion, but I don't think we will see a meaningful reduction in fatal accidents or serious injuries due to self-driving cars for a long time, but I do think we will see a reduction in accidents involving only property damage or minor injuries from low to medium speed accidents, and I see a sharp decrease in parking-lot scrapes and dings in the near future. Low to medium speed streets and tight maneuvers is where I see most "bad" human drivers have the most trouble and already things like parallel parking assist are a godsend to these people.
The first airplane that received CAT IIIc autoland certification was the Lockheed L-1011 Tristar back in 1970. It's far from a big deal 50 years later.
I had at least one autonomous landing due to heavy fog. Was announced by the captain, he said something like: "Because of the weather conditions, we'll use automatic landing for safety reasons". Landing was quite flawless, despite the fact that I couldn't see the ground until we were a couple of meters close. But the landing approach seemed to take longer, maybe there is a trade-off.
Travel by plane is orders of magnitude safer than by car. Your odds of dying in a traffic accident are close to 1:100. Your odds of dying in an air accident are closer to 1:10,000.
If anything, I think your example underlines just how valuable and realistic this is for cars.
Full autonomy was always, IMHO, a silly pipe dream. But that doesn't mean that there isn't tremendous value -- in terms of safety, time, money and environment -- to partial autonomy.
> If anything, I think your example underlines just how valuable and realistic this is for cars.
I interpret the stats the other way. Cars are not planes and driving involves constant encounters with potential collisions and which would make level 5 too difficult technically and socially rendering them too costly to be valuable.
Maybe if cars could fly, they could sort themselves at different heights so that they would never hit each other. And also communicate with others to make sure there are no collision courses. Tall buildings could be a problem. But it should be way easier problem to solve then city street driving. Because it would be a new tech we could require all flying cars to be autonomous and only allow manual control in case of emergency.
Both from a comfort and safety perspective, automating long highway drives is a big win. It’s still disappointing to urbanites in particular who thought they’d never need to own a car, drive anywhere, or maybe even learn to drive in the first place.
Onboarding cars onto trains in a driver friendly way should also be considered. This way you don't exclude non self driving cars, and still have good coverage (taking into account that railway systems aren't far behind highway coverage in many countries).
My opinion along these lines lends itself to depiction of self driving cars in the series Tek War. Upon entering the limit access highway or interstate the car drives on its own for the duration until your exit.
that to me is something automakers can easily cooperate with governments to get done and done well. they already mark HOV and Express lanes and with federal legislated lane markings it would make automating those drives pretty simple. Then as the number of cars increase the number of free drive lanes is decreased until the vanish.
using metro Atlanta as an example express lanes have their own entrance and exit points so they are already separated. One set for I75 is wholly elevated two lane bridge work nearly the full length of its run.
Tesla Model 3 owner, I think even Musk is walking back what he "implies" as self driving. As into self driving with human oversight
That is a bit American centric. The value of self driving cars for society will be the ability to optimize existing road infrastructure. Cars can travel more closely together, can cooperate to work out bottlenecks, and so on. Think about the traffic problems Beijing is having, and that’s the huge opportunity for self driving tech.
People who can't, or shouldn't, drive themselves at all are totally unaccounted for in this logic. Full autonomy would be hugely freeing to the elderly, for example. Presumably it would enable travel for people too young to get a driver's license as well. And I have to assume driving patterns for capable drivers would change too, in ways that are likely hard to predict or appreciate right now.
Cars with no person inside them at all (vehicle drones?) would unlock the biggest productivity gain in a century. Revolutionize anything that can be delivered to your house (food, goods, groceries) and make sharing items feasible (need a power tool? Just have a drone drop it off for an hour then pick it up after you are done). I think the biggest advance of self driving is having cars with no people in them at all
You cant have other people on the road then. If I was a protestor or bored teen, I would find it pretty fun to block self driving cars' cameras and sensors, stranding them and creating chaos. Imagine a line of self driving cars just blocking a busy street. It wouldn't be hard to do, but it would take a lot of work to recover them to the point it wouldnt be profitable for the companies running them.
It would be fun, then the police would show up and arrest the people who did it using camera footage from the automated cars. Most people are good people, and the ones who aren't already fuck with the road network now causing all kinds of chaos.. throwing bricks from bridges, ripping the copper out of traffic signals, but it's relatively rare because most people aren't trying to ruin everything at great risk to their freedom. I don't really understand why automated cars would be a huge change of affairs.
cuz its just funny. If you saw little robots all over the place run by facebook or $mega_corp, wouldnt you just strand them for fun? Cameras are easily defeated by a hoodie and a mask.
Bricks from bridges could kill people and just damages regular people's property. It's just different.
What if I told you I could offer you a 99% automated intercity solution? That's right, a solution where 99% of the people involved have no obligation whatsoever to engage with piloting the vehicle... EVER! Where the vehicle just sort of seems to... drive itself? Is this some futuristic sci-fi fantasy? Well hold onto your hat, because the future is NOW!
Riding trains is great, getting to the train station(s) in a busy city, worrying about the schedule and being in the right spot for the right train, etc, is not at all fun.
But otherwise agreed a ton of this stuff could be solved with modern trains.
But appeasing all of the special interest groups in between has made infrastructure development impossibly expensive or takes a decade. So we keep driving cars even long distances between cities.
This is still a solved problem in many European cities. Train stations are often in the city centre, or on the edge of it. They almost always connect with either metro or bus networks, typically with frequent services.
Not in Germany, may be in big cities, but any other region is plagued by disruptions and way less frequent services paired together with horrible local public transport.
The former government owned national railway basically owns every piece of important track and doesn’t give a damn about customers. The prices are constantly rising while the service declines. Can’t count how much time I’ve wasted on several train stations, waiting for up to two hours for late trains, hell, even having to get a hotel multiple times.
Cities like Tokyo have solved most of that for now decades. Def depends on whether the government is willing to make the investment (heavy) into the right amount and level of infrastructure.
When you know that Los Angeles used to have over a 1000 miles of rails before the 60's and after pressure from car and tire makers ripped it all away, you understand where all our traffic problems come from :)
> Riding trains is great, getting to the train station(s) in a busy city, worrying about the schedule and being in the right spot for the right train, etc, is not at all fun.
I'd say that, in the majority of cases, it's either no more difficult or outright easier than getting to an airport, as airports tend to be located on the outskirts of cities (requiring connecting train/bus service or private cars), while most rail hubs tend to be in city centers.
Trains kind of suck though. I used them a lot in the UK because I didn't have a car. The UK has an OK train network.
The problem with trains is that they are very, very inconvenient for most journeys. Sunday evening? forget it. After 11pm? Nope. Journey that runs perpendicular to the local line to London? Lol, enjoy a 5 hour journey to go 100 miles. Want to take cargo like a new washing machine? Lol. Want to go somewhere that's not near a station by bus? Enjoy Journey times that are 4, 5 or 10 times slower than a car.
Trains are good for busy commuter lines and nothing else.
But to stab at that anyway, it wouldn't be silly if he did. People driving impaired for lack of other good and affordable options is an emergent reality, no matter how morally hazardous it is.
I don't drink, but it is very, very inconvenient if your activities are constrained by the train timetable. Much better to have a car available and be able to travel when you want to.
Of course sometimes you dont want to travel late or at weekends
1) trains go basically everywhere frequently in Japan, it’s very much a functioning rail network, they exist and kick ass.
2) a train doesn’t need to replace 100% of personal car trips to be useful. I’m sure you’re not transporting dishwashers or taking trips after 11 pm every day. Covering 90% of use cases is sufficient, and the other 10% can be handled by either renting a car or taking an Uber.
3) we can and should make busses faster than cars during times of congestion by giving them their own lanes.
I guess my wider point was that today, 55% of the world’s population lives in urban areas, a proportion that is expected to increase to 68% by 2050.[1]
Urban commuter trains will have little effect on that ~45% of people who won't live near them even if they are built.
On the flip side, though, I guess, that doesn't mean we shouldn't build more train lines.
The biggest fault of the entire Japanese system is the stoppage at night. 24 hour service is not the solution and is one of the reasons the services there work very well. But I don't understand why they don't have a replacement system. In Philadelphia the (smaller) system is completely replaced by buses. In other countries the trains run 24 hours but only Friday nights and weekends nights which could increase the use case of 90% of riders to 95% of riders.
The Japanese rail network is awesome but it does not stop everywhere and it’s quite expensive. Japan is huge and taxis get expensive real quick.
Consider Singapore, which is tiny, has a kick ass public transport system (trains and busses) and where it costs 90-100k SGD to own a beater Corolla: people still own cars.
Some people own cars. Car ownership rate is around 11% according to [1], which doesn't look that high.
That said, Singapore public transportation (at least SMRT trains [2]) is freaking strict about food and drinks. I could not even have a sip of water without triggering a warning message. So anyone who can't control their thirst or hunger (like, you know, basically any kid) isn't really welcome there.
Nowadays I live in a city with a tram system, trains to other larger cities and bike lanes.
If you are going from one city centre to another, during the day, the train is the best choice.
If you are moving within a bike-friendly city, bike is best (or Uber, but it is banned here). If you are somewhere in an urban sprawl, it's car or Uber.
I definitely find trains useful, but I think that they are consistently overrated by people for emotional and irrational reasons. I personally use trains for maybe 1% of my journeys and trams for 0% despite them being extremely prominent in my city.
Look, from the point of view of a passenger, the best transport system goes:
- where you want,
- when you want,
- carrying what you want,
- as fast
- and as safely as possible,
- at the lowest cost possible.
Trains are weak on multiple axes here: they do not go where you want to go and they do not go when you want to go, and you cannot carry large amounts on them. They are also quite slow.
The only reason that anyone uses trains is because in dense urban areas like London, cars are even slower, and for intercity trips it is both slow and hard to park when you get to the destination city. There are perhaps some exceptions to this but they don't really exist in the West. Really fast intercity trains are a thing in China and Japan, but the UK has no trains that take you from A to B at an average speed greater than a fast car once you take waiting and stopping time into account. As far as I am aware the same applies to the US as well.
Trains do not cover 90% of use cases, they cover maybe 10%. In my opinion, trains are a technology that should be completely scrapped in favor of driverless minibus and driverless car networks. Modern electric & driverless vehicles with internet connectivity and global transport optimization like Uber would beat trains on every single axis.
No, trains in the UK are not "OK", they are crap compared to most of Western Europe. Fewer trains and more cancellations than any other country I've ever used trains in, stupid connections and pricing structure, all because of privatization. Look to Switzerland for the best train network in Europe. Reliability is top notch, there's only one company to deal with, and prices are reasonable for the income level.
Reliability of trains in the UK really varies by route. When I was studying and when I first started working I extensively used mainline routes and had no issues.
The last train I could take on a Friday night was 10pm, but given it was a 3 hour journey that didn't seem unreasonable. Each year there was usually one incident that caused delays, but other than they worked well. If I took a fast train the journey time was around the same as driving - quicker in fact given on the journey I would stop for a break - and ticket prices were about the same as driving (for 1 person).
On the other hand I was working in London in 2018, and at weekends often flying from LTN. In the end I added an hour buffer (to a 45 minute journey) because the reliability of Thameslink was so bad.
They're not as good as the best, but they're a lot better than the worst countries for trains. The train network is also fairly thorough in its coverage of the country. I think it's a good representative example (not best, not worst).
Albany is a middlin' sized city. Taking the train to NYC is like, still a nontrivial event. Also passenger rail has to sidestep freight on this corridor. Megabus is a bit better experience overall but also kinda meh.
On top of that, I personally (and know many agree) HATE the stress of keeping track of times, scheduling around infrequent departures, logistics to/from stations, etc.
> Bonus: the vehicle can be resized at will, depending on the number of passengers.
Yeahno. "please wait 15-30 minutes as we shuffle cars around the rail yard" is not exactly "at will"
Waiting 30 minutes while a new carriage is attached still beats sitting in a traffic jam. At least for me.
I gladly pick rail over car for good lines, because it's stress free. Get to the station in time (with public transportation it's quite deterministic, subways and trams don't get into traffic jams, buses sometimes do).
Cars of course win when it comes to flexibility (late night getting to a small town address).
I quite like trains. I hope someday the US gets back to having better rail service.
Highway-autonomous vehicles would give you the benefits of rail service plus arbitrary departure times, more direct routes, on-demand route changes, and you're traveling in a vehicle that you can use for first and last mile and intra-city travel.
The problem in the US is we aren’t going to start getting European level train infrastructure anytime soon. Also it’s arguably not a worthwhile goal, we have so much space, and it’s an uphill battle to change US culture to be one where trains are going to be a good fit for our land use and urban planning.
I think surmounting those challenges is useful in itself, not just as a way to enable trains. Using self driving cars to avoid doing that would be a real shame.
I don’t mind driving 200 times one hour commuting a year. I want to a) not own a car but be picked up by one I can share with others b) be driven home after having a drink. Both of those use cases are only working if there is a 100% (or close to) self driving solution. Driving 95% of the way or 95% of the trips is nice, but I wouldn’t pay Tesla autopilot money for it.
The value in the 95/95 autopilot is because it will make commercial drivers be able to never need breaks (they sleep or rest until the autopilot is in trouble, and they can even remote control cars). And of course because a lot of people seem to hate driving and would pay to avoid it.
The 95% thing can also be very different depending on how it's split up. 5% of miles have a random "you need to take over" vs. the most rural 5% of roads need manual control make a world of difference.
A car that can only drive within major cities in the UK and on motorways would be entirely fine for the vast majority of people. Exceptional journies I can always rent a car that suits my needs for the one week in several years I'm in the Highlands.
> Driving 95% of the way or 95% of the trips is nice
People will abuse that. They will get high/drunk/whatever and when that “5%” rolls around (quite unexpectedly mind you), they’ll run into a bus full of nuns. Always assume people will be judgement impaired because “the car can drive itself”....
Where the 95% ends can only be a decision the car makes. There can’t be a button you push when you think “come on autopilot you can do it, It’s just a short jump and you’ll fit through that narrow passage in the run-up easily!”. Autonomous cars will give up and stop when they require manual intervention.
But, this is also why I think the next car manufacturer race will be about who can solve this for their customers. That is, who solves the problem of making 95% autonomous cars able to drive in 99.9% if all situations without manual intervention by the driver, where the provides automatic remote control from massive rooms full of people steering drunk car owners out of situations the autopilot couldn’t. The better the autonomy, the fewer people needed to remote override.
Most of the value in being able to travel by auto without attention is in trips longer than 15 minutes. Commutes, tourism, vacations. Shipping. Most of which is certainly highway driving.
A 90% autonomous trip where I have to pay attention 100% of the time gets me no real value. It's just a slightly nicer, maybe, ride.
It seems unlikely you commutes would completely autonomous even people commute on freeways.
I am late to the discussion. I dont understand why we cant have new Cities that only allows self driving cars. Full AV can react to other drivers, but they can surely act upon themselves.
This way, you can set out home with cars and you only operate the last part going to Cities for work. And when you leave you drive out to high way and then every full else will be automatics. It also partly solve housing problems and people can now live further away from work.
Surely this is a solution that is simply enough for the short term. What am I missing here?
To a certain extent I agree. I had already written at length in another post about this topic, but my opinion is that what we think of as a "car" right now will never be fully autonomous.
I see the future diverging into two paths, fully autonomous commercial vehicles, like taxis, delivery vehicles, semi-trucks, etc. that work within urban areas or other designated, mapped and specially prepared areas. This will possibly involve a centralized system of control and communication, something like an ATC but for cars. These will be owned by corporations and only used by people. It makes no sense for a person to buy one of these fully autonomous vehicles, although I imagine some people would pay extra to have priority access so they always have one available.
The other side of the coin will be privately owned cars that have autonomous capability, or autonomous cars with override. These will be able to go anywhere the driver wants, including unmapped villages, small towns, off road tracks, etc. I suspect these will be the domain of enthusiasts, people who really need them for work (ranchers and farmers, for example) and people who choose to live away from urban centers. They will be more expensive than cars now, but the need for these vehicles will never go away. Even if we get true AI capable of driving anywhere with only the sensors aboard the vehicle, there will still be the need for a human to override it, even if that involves just authorizing a risky maneuver or putting the AI into "unsafe driving" mode.
I think the movie I, Robot (with Will Smith) got the future of autonomous cars surprisingly right, autonomous inside cities and on highways, and using the manual override comes with penalties (higher insurance, being at fault in an accident, etc.).
On a personal note, and this may sound bad, but I would never buy a car which I cannot use to break the law. Even if I never plan to do it, being able to speed, jump the curb, intentionally crash into a wall (or another car) or even run over a person (for example, in self defense) may be at some point required or the least bad of many bad options. In this case any consequences should fall on me, but I don't think a thing I own should be designed to prevent me from breaking the law or doing something stupid if I really want or need to, although providing warnings or an optional safe-mode is fine. I suspect many people feel the same way, even if they don't put it in such an extreme way. This can be seen by the fact that a lot of cars, particularly those focused on performance or off-roading, come with switches to turn traction control off, and if they don't, it will get mentioned as a negative in any review done by publications focused on those audiences.
The technology is not the only problem. The lawyers that are defending every dead body, and there will be dead bodies, that will sink self-driving cars. Even if everyone is 95% safer with self-driving cars, those that are killed by a self-driving car (in combo with a public that is easily swayed with non-objective arguments) will be hard to dismiss.
Most drivers on the road are effectively being subsidized by bankruptcy protection, because most cannot possibly cover the liability they are exposing themselves to by driving. This "subsidy" is far less valuable to a self driving car manufacturer than individual drivers.
I mean, sure, you can come up with some scenario where the liability exceeded the insurance coverage, but I haven't heard of many of those. Anyway, it comes from somewhere. If bankruptcy protects drivers, it also exposes them to the risk that they will suffer damage that isn't compensated.
Regardless, expecting a legal loophole to preserve the status quo indefinitely seems quite unrealistic and inherently unstable. If that actually holds up something that could massively benefit society (both economically and in saving lives), we simply legislate liability limits.
Liability you can incur while driving is almost arbitrarily high. Individual drivers rely on the existence of bankruptcy protection to cover these rare scenarios, or simply don't think about or plan for this at all.
> but I haven't heard of many of those.
How many do you think it takes to put a self driving car manufacturer out of business?
I'm not saying the status quo is a great situation, or that this is a good or bad argument for or against self driving cars. Only that it's a description of the current situation, and why legal issues might be a much bigger problem for self driving car manufacturers than individual drivers.
Because the deaths will be network level effects that a person would be helpless to mitigate through behavior. There's no sense of being a careful or responsible driver with a self-driving car; its all whether or not the algorithm or software is correct.
It's a different type of potential error and a much more scary one. I can mitigate human drivers as a pedestrian by taking care walking, but I cannot mitigate an AI that thinks my shirt makes me look invisible due to its learning being deceived by a pattern.
Because a jury will hand out a huge multi-million dollar award against a self-driving car company when its car crashes causing a death, while when a human causes a death it is usually just an "accident" or a much smaller settlement based on the person's insurance coverage.
Because the autonomous car is a systematic issue that could affect every single car on the road using that manufacturer's software, and the "driver" of the car has no way to stop or mitigate the risk, or cause risk. He puts his life in the company's hands each time he drives the vehicle.
I'm not sure why people even embrace self-driving cars. By now we know that centralization, lack of maintenance, and fragmentation in software are serious risks, as well as how software can increase the attack vector on people as well as provide benefit. I'm not sure if the risks are worth the modest efficiency increase, and this isn't even getting into existential risks like external parties being able to control when you drive, or attacks on the networks or technology.
Even if the death toll 1/100th ... the few dead will be paraded as examples. There will be outrage. There will be lawsuits. You are correct, it is not logical, but logic does not always prevail.
That works both ways, though. Once self-driving is proven, the higher accident rates from human-driving will become a greater liability. "Why were you driving yourself at night instead of engaging autopilot?!?"
You are correct; the public won’t accept the technology until it’s at least as safe statistically as air travel (and even then, there’ll be pushback in response to specific inevitable tragedies).
However, I think the profit incentives of the trucking industry will manage to carve out some regulatory exceptions; something like “freight trucks can self-drive between 11a and 5a on these specific Nebraska highways, with warning signs on both roads and vehicles”. This sort of lobbying will be the thin end of the wedge for both iterating the tech, and normalizing its acceptance.
We could say that now about automobiles. If I were to get into an accident and then have lawyers for the other vehicle, plus everybody stuck in traffic behind me also sending their lawyers after me, I’d never drive.
Insurance solved that problem, both by eliminating the possibility of a catastrophic financial loss, and by creating a buffer between me and all those lawyers.
I predict that insurance will solve the lawyer problem for self-driving cars as well. At some point, it will cost me $5,000 a year to drive my own car, and $500 to let it do all the driving.
And on top of all that, if my car drives itself into an accident, the lawyers will talk to my insurance company, not to me. I see the insurance companies as the enabler for this tech. And they will want to enable it, it will put them in control of the market.
Indeed we can which is why some cities on the planet have gone carless and more and more opposition is mounting in the face of traffic deaths, pollution and so so on. Even in the US, maybe the most car dominated country in the world, there is a political revival of talks about high speed rail and alternative forms of transportation.
The other important difference being that the step from having no cars at all to having cars was one of the largest leaps in mobility in human history. Self-driving cars are nice, but not that much of a leap, and they have much more ambiguous implications when it comes to the job-market. They will face significantly larger hurdles with significantly less payoff in sight.
Pontevedra in Spain for example, Venice and Oslo are also to large degrees car-free in the city proper. Yes cars are banned from the city, here's an article about the consequences.
The world is not the USA. The legal system that exist in the US is different in other places of the world.
With a self driving car you know exactly what happened in an accident as video and telemetry is recorded. This is a tremendous advantage over having to reconstruct it without this data.
On the contrary I expect legislation forcing every car to include telemetry like the Chinese are forcing every car to be connected.
This accident's data is evidence, not opinion, not a belief, not a prejudice.
The usefulness of this has already been proven with airplanes.
How happy will it be when camera data in the age of hyper realistic cgi films and “foolproof” telemetry will protect us from killing robot car makers at court. :)
Agreed, if a self driving car hits me and breaks my neck, who do I sue to pay my bills and care? The driver? The car manufacturer? Or the self driving software company? Right now with a driver its clear.
To the contrary he should be admired for his candour.
It's very easy for execs to talk up utopian smack, it's harder the other way around.
I for one agree: L5 may not be viable for many decades at least, not with the tech we have. It may never happen because conditions will change to the point wherein it won't make sense.
What is viable, that nobody really talks about are 'controlled areas' - for example, highways built for long-haul trucks with 0 human drivers. This is a great place for AI because the conditions are set for all the crazy unknowns (esp. humans) to be controlled.
Special roadways within cities, where there are no pedestrians, AI drivers only, that can communicate with 'the grid' etc - could not only mean full L5 'driverless' but no stop-signs either - traffic could be designed to flow much, much more efficiently using 'today's tech'. Cars won't have steering wheels, they can be called 'on demand' and 'parking' will be very different, more like 'temp storage' and won't involve humans. All viable.
This is a point that always seems to be missing. All of our roads are/were constructed at a specific size with specific engineering requirements and specific materials. If we have a major change in the technology of the cars, it seems to be perfectly reasonable that we could have a new type of road to accommodate that. For example, before interstate highways it probably would have sounded strange to imagine high-speed roads with limited entry and exit. And before that, may have sounded strange to have automated stoplights and traffic signals.
>A leader that fires people for challenging relevant mainstream opinions is not any kind of leader I want to work for or follow.
There we go...most of the skepticism about self-driving cars appears to be driven by ideology and less by the actual evidence. "Mainstream opinions"...it is not an opinion, there are literally thousands of these cars being tested/piloted and improved upon constantly.
Might it take a little longer to get to nirvana? probably, but it borders on delusion at this point for anyone to say it will never happen.
>There we go...most of the skepticism about self-driving cars appears to be driven by ideology and less by the actual evidence. "Mainstream opinions"...it is not an opinion, there are literally thousands of these cars being tested/piloted and improved upon constantly.
You can't provide evidence of something that hasn't happened, and level 5 autonomous vehicles haven't happened. All you can provide is evidence of things we have already achieved, and use that to support a strong opinion that we will be able to achieve more things.
Anyways, back to my original point, firing someone for having an informed counter-mainstream opinion is the sign of a weak and petty leader.
Volkswagen doesn't develop self-driving tech. They use Mobileye's solution like everyone expect Tesla, Waymo and GM/Cruise. They aren't really in the position to make that argument.
Level 5 means that the car can drive autonomously > 95% of the time, and it will absolutely be possible. Level 4/5 distinction doesn't really make any sense once the autonomy gets beyond certain percentage. It doesn't mean that it's level 4 until it hits 100% (which is impossible).
Level 5 car is designed to drive in all conditions, but there are always statistically unlikely corner cases or situations that require high-level decision making, which the car can't handle by itself. A single driver may never hit such case, and for them the experience is full self-driving.
I stand corrected. Still, VW is using MobilEye's tech in their ride-sharing service [0], which probably means that their own tech is not very close to the competitors.
I think he is the one and only sane C-level person in the industry, after hearing this.
People claiming most of AI hype barely know what the "AI" people mention is.
And people who go as far as drawing rosy pictures of human like general AI being your personal chauffeur are past ridiculous.
The entire idea of human-like general AI for practical applications is like trying to make people using horses for transport in 21st century, by trying to make a horse than if better than a car.
I don’t think it’s impossible if certain roads/highways or lanes become marked as autonomous only. Just as pedestrians and bicycles are banned from using high-speed roads, manually-operated cars could be banned from using certain roads/lanes, removing a lot of variables that manually-operated cars bring.
How is the ability to leave at your own time, have your own dedicated cabin, and ability to stop nearly anywhere, even in the same ballpark as a Tram...?
I wish governments would consider the death and injury toll of manually driving, and fastlane (pun intended) infrastructure for self driving cars, such as specific roads and lanes.
Isn't it a little silly to call someone "only sane C-level person" after one comment?
Also, while claiming these innovations are only a few years way may be insane, saying they will never happen feels probably just as non-sense. History hasn't been on the side of such absolute claims.
It will take time, but Waymo already has working 100% self-driving cars out in the wild. Yes it's in a small area, with perfect weather, and with technicians ready to jump in, but it's still a good start and easily paves the way for it to extend over the next decade or so.
I do feel like that. From my past few years dealing with automotive clients, I haven't heard of a single company that was not completely obsessed with it.
All and every piece of electronics we did in the automotive sector had some lame "AI platform" attached somewhere, for no apparent reason.
The most blatant "AI washing" I saw was an "AI power steering" which was a plain PID, inferior on all fronts to purpose made systems already on the market.
Second to it was an "AI airbag," which was basically an OpenCV hello world that will trigger the airbag ahead of time. Terribly unreliable, and would've probably killed more people than saved.
Yes, it's a hype based buzzword. Just like "cloud" could be found everywhere a few years ago by people trying to profit from said hype. And just like "cloud", usage will start to become somewhat more realistic eventually. But I doubt the industry will look exactly like it did before... just like many of us didn't switch back to local bare metal servers.
I totally understand mistrust. Less so if there are somewhat capable alpha products already available with huge economic incentives behind made by somewhat reputable companies. It might not happen in the current hype cycle, then it will be a future one. But I don't ever see self driving cars quietly being deployed. That's just a way to well known game changer hardly anyone will want to sit out.
To me those are two very different statement, at least for my interpretation of the first statement. If the VW exec was referring to very hard weather conditions, then fair enough, but I doubt that's what he meant.
The Waymo CEO was very clearly referring to extreme conditions like storms, and honestly, most humans can't and shouldn't drive in said conditions, so it's not that crazy to say Waymo may never be able to either.
> saying they will never happen feels probably just as non-sense. History hasn't been on the side of such absolute claims.
I wish there was a collection of stubborn dismissals of things that did come to pass and became bigger and better than originally imagined, so we could point and laugh at those cynical sticks in the muds.
Steve Ballmer's dismissal of the iPhone being a popular relatively recent case.
I still remember that quote I discovered through Civilization, apparently made by some popular leader (Napoleon?) about steam engines: "You would make a ship of iron and without sails move by lighting a bonfire under its deck?? Absurd!"
You don't understand. We just need more data. Once we have enough data the AI will be able to model all possible scenarios and make a better decision than a human driver ever could! /s
Agreed. I have running bet with fellow at work who owns a Tesla about when we will have self driving cars. He keeps telling me the AI revolution is just around the corner. Still waiting.
He talks about difference between level 4 and level 5 (4 - attainable, 5 - may never happen). This is a strange proposition in my book: level 4 is already good enough to bring a lot of value (taxis in a big city are level 4 afaiu), level 5 - ideal but who cares?
It's actually debatable that you ever need level 5, because once you reach level 4 it's probably only a matter of time until most road infrastructure is converted for L4 self-driving vehicules (once you can statistically prove a self-driving road kills 1000x less than one with human drivers).
Like we never wondered about self-driving trains, essentially. It just became a fact once we could do remotely supervized L4, and that novelty was short-lived too.
taxis in a big city isnt neccessarily level 4. True, theres no person in the drivers seat, but there are still things that restrict the operation of a Level 4 system, while 5 is supposed to go anywhere, anytime. Snow, Rain, Fog, even night could all impede Level 4 systems depending on the implementation's restrictions.
You are wrong. And the "serious autonomous teams" are also wrong in not using AI (which is mostly true AFAIK, as long as we don't call basic image recognition to already be "AI").
But that's just because the AI we currently have - basically idiot-savants made of silicon - is mostly useless to them, since the kind of AI they would need for actual, 100%, no-compromise level 5 self driving capabilities hasn't been invented yet, nor does anyone have a promising idea as to how to invent one.
I am pretty sure that the only way to actually reach this goal requires a full-blown AGI. Because what we call "driving" and what is often misunderstood as a rather mundane and repetitive mechanical operation of machinery governed by a few simple ground rules is actually an insanely complex task that requires several higher-level capabilities of the human intelligence: besides pattern recognition, quick evaluations of situations with an unbounded number of variables and application of knowledge and concepts to an unlimited number of ever-new traffic situations it requires foresight, communication (of the difficult nonverbal kind) and sometimes even carefully dosed brazenness - to know when rules should better be ignored or bent instead of followed.
Or perhaps, VW exec hopes full self-driving cars may never happen, because it would imply a fleet approach where transportation would be a service instead of cars being a prized possession. That, of course, would mean far less cars. So don't expect a traditional car manufacturer to lead the way.
This is incidentally one reason for Tesla's huge market value. The company actually has a plan to transition from individual ownership to fleet, so when this happens it will be prepared to deal with a new manufacturing reality.
Just try to imagine VW without all those ads to sell a positive self image because you drive a sexy cool car they make.
>Or perhaps, VW exec hopes full self-driving cars may never happen, because it would imply a fleet approach where transportation would be a service instead of cars being a prized possession.
Cuts both ways. Tesla execs have the same reason to overhype autonomous driving, because their market value depends on being perceived as a hypermodern tech company.
So far reality has proven VW right. Elon himself had to walk back on grandiose claims about full autonomy and robotaxi fleets, or advertising cars with slogans like "the driver is only there for legal reasons, the car drives itself" a bunch of years ago.
Full autonomy on a human level requires human levels of intelligence and 'common sense', it's a ridiculously hard problem that requires several leaps in AI and plenty of other fields.
> O r perhaps, VW exec hopes full self-driving cars may never happen, because it would imply a fleet approach where transportation would be a service instead of cars being a prized possession.
Why on earth would anyone use this over public transit? That would be a hugely expensive way to travel.
Public transportation where I live is disgusting and often dangerous. Buses and trains don't pick me up and drop me off in front of my home. And they run on fixed schedules.
Due to these advantages, I usually take Uber/Lyft instead of BART or Muni. Autonomous vehicles would reduce the cost of such services even more.
Same reason people use Uber and Lyft. Same thing but no driver. Uber pool is already down to maybe 3x public transport cost, without the driver and with EVs the cost could be quite close.
After Elon Musk so helpfully pointed out that once cars are self-driving there is absolutely no reason for companies to sell them to us, I have to say I hope it never happens. Why would a company sell me a car for $30,000 when it can add it to a self-driving taxi fleet and generate $300,000 in revenue over the lifetime of the car?
If it works it could be so much better in many ways. It's like a packet switching network instead of circuit switching. And there's no reason larger vehicles couldn't be part of this "packet-switched" vehicle network and act as a virtual public transport system.
I was at an ML conference in 2017 and the keynote speaker asked the audience how many years they thought it would be until ADAS lvl 5 was on the market. The keynote had proudly proclaimed 5 years, and then audibly scoffed when about 90% of the hands went up for "20 or more years". There's such a disconnect between real engineers and cheerleaders.
I have a friend who a mechanical engineering professor doing robotic research. He says that one of the reasons that robotics has not advanced more is because it has primarily been the domain of computer scientists who don’t operate in the real world. Every robot works great in a simulation. Same could be said for cars.
Though there's the Kurzweil argument that people's intuition doesn't work well with exponential growth. The engineers are probably thinking these systems work terribly and have for decades and so will for decades more but meanwhile GPU performance is doubling every year or two and many people are trying to take advantage of that.
I've been following this stuff casually since about 1980 when I totally failed to write an AI program and since then processor performance has been steadily increasing, first with clock speed and now more with multi cores. Hans Moravec, a robot guy did a reasonable estimate that to get equivalent hardware to the brain you'd need about 100 teraflops and just recently the first 100 teraflop GPU has come out (https://www.nvidia.com/en-us/data-center/v100/) which is kinda historic in a way, maybe a billion times faster than a cheap computer in 1980. Of course this will keep going for a while so the situation will flip from we can't do much because processors are much slower than the brain to the other way around.
(update NIDIA may have fudged the numbers a bit but anyway.)
923 comments
[ 0.64 ms ] story [ 199 ms ] thread> "This is one of the hardest problems we have. This is like we are going to Mars," Hitzinger said in a comment. "Maybe it will never happen."
First of all, it seems obvious that we are going to go to Mars, eventually. Maybe not any time soon, but never? Seriously?
But the bigger thing is that there is about 1000 times more economic benefit to self driving cars than of going to Mars, at least in the near term. To think we'd just give up on it seems absurd.
Given the questionable economics underlying humans having a presence on Mars and the extreme toll on the human body this will have, foremost by radiation I think it's actually a fair comparison in particular because full autonomy like mars colonisation is constantly being overhyped mostly by a very small group of very affluent individuals who seem to be more inspired by sci-fi than engineering.
Let's take going to Mars. We can't reliably go to the moon. We can't even go everywhere on the Earth, where we have every possible advantage. Spend a year on the ISS and you'll develop all sorts of health issues. Spending time on Mars isn't likely to be less hazardous. We may send humans to Mars but it is by no means guaranteed. (And I would also ask, what reason do we have to go to Mars? Probes do a better job of exploring, and I doubt we could colonize it.)
Until someone builds a real self-driving car it is just an idea. On today's roads, it is probably not possible to safely implement a fully self driving car. Driving is not only a technical exercise but a social activity that involves communicating your intentions to other humans, and interpreting the intentions of others. That is something that humans do far better than machines. (Edit: To clarify, humans are better at communicating with other humans. Machines do a great job of communicating with each other, but have mixed results communicating with humans.)
If every car was self-driving and the roads were remade from first principles, then sure, that seems feasible. The degenerate case here would be a self-driving train, which seems perfectly reasonable. But the technical challenges are the easy part of that endeavor. Funding such a project, developing the political will to see it through, and organizing the logistics are far more difficult. Consider for instance; what will happen to the legacy vehicles? Will it be illegal to drive them? Will there be a massive government buyback? Who will fund that? Where will the cars go? How will we organize the logistics of moving hundreds of millions of vehicles? How long will that take? What other matters will we need to turn our attention away from to accomplish that task?
A much more likely scenario is that companies continue to come up with partial, ad-hoc solutions, driving gets more automated, ride sharing becomes more popular and car ownership less so, but that humans remain in the loop for the foreseeable future. What happens outside the foreseeable future is something we can't and shouldn't pretend to know with any degree of certainty.
There are a million ways in which we could never go to Mars or build a self-driving car. We could get a better idea for how transit should work, making self driving cars superfluous. We could discover life on Mars, and make the decision that it would be too dangerous for us to visit. The superpowers of the world could go to war with each other, and our infrastructure could be devastated to the point where space travel is impossible. Climate change could drive us to extinction. Something could happen that we cannot predict or imagine, that we have no precedent for, that completely changes our situation and outlook.
Some of these are more likely than others, but the point is that it would be folly to take the future as read. And frankly, a couple of them are more likely than us ever going to Mars.
I think it's more than an idea when we have self driving cars, they just don't meet our standards for safety yet so they aren't legal on public roads.
I mean, we had airplane flight in 1903. Good chance you'd die if you went up in one for the first decade. But we did have powered flight.
It’s just that it would cost an incredible sum of money for very little gain and so it hasn’t happened - even the mission to the moon was a huge drain on the budget and the program was canceled once we won the space race.
Going to Mars is like supersonic passenger air travel (Concorde) - something we have the technology to do, but it just doesn’t make sense.
An autonomous, self-driving automobile is something we just can’t do today no matter how much money we spent on it. I think it’s a different question because we are trying to predict if we will ever reach the level of technological sophistication to allow it. (Of course we have a solution for “driving a car from one city to another” and it’s putting a human in the seat, or of “making a self driving vehicle” and it’s putting the vehicle on grade separated tracks like the Morgantown PRT)
But I agree that it's actually a generous comparison to self driving cars. After all, we do occasionally send machines to Mars, and they work quite well.
What I mean is it's a fascinating but daunting task.
We are not "back" to windmills that give 3 MegaWatts of power because they never existed in the first place, like affordable solar panels with 20-40% efficiency.
Progress is not automatic. It is millions of times harder to create or improve a technology than imagine it. And also takes lots of money.
We have nuclear tech because Manhattan project, because WWII(and because they were scientists coming from Europe that were scare of Hitler). It took a tremendous amount of money and sacrifice to get there.
No, Manhattan project took trivial amount of money by today's standard. It costed less than the rounding error in typically given federal budget figures for past few decades.
That’s about the same as NASA’s annual budget.
Truly, we are easily able to afford projects like Manhattan and Apollo today, it's just we aren't able to pull them off anymore.
But of that 4.7T budget we only spend 600 million of that on fusion research.
With the caveat: for commercial travel. And even then it’s partly due to noise restrictions. It’s not like the technology for supersonic flight isn’t widespread.
Nuclear power, let alone fusion. What happened to “too cheap to meter”?
Supersonic travel: raynier already pointed out what a bust the Concorde was. Nothing else on the horizon now.
Automation giving me a 20 hour work week. Nope, capital holders just skim that efficiency right into their pockets.
I’d go on, but suffice it to say that about the only thing much different than my childhood in the 70s are computers in our pockets. Revolutionary, no doubt, but we are still burning oil for our energy needs, our cars don’t fly, and I still show up and do my 40 hours. And healthcare has gotten worse in the U. S., not better, if you can believe that.
So, yeah, when the head of the autonomous driving division of VW says Level 5 ain’t gonna happen, I don’t immediately jump to doubting him/her and attacking their resume.
And yet HN threads are all about "here are all these cool things that will happen tomorrow if Elon Jesus delivers".
He ain't going to deliver, people.
[1]https://boomsupersonic.com/
Health care is worse? More expensive, yes, but anybody getting cancer and wanting to live is much better off now than in the 1970's.
If you go up or down, the number and kind of obstacles reduces. The location of interactions between the vehicles is reduced, and the interactions with other classes of vehicle are zero, so you can negotiate.
Solve a simpler problem, if you can.
Is the term "Grade" here used in the "Slope" sense, as in don't run the cars up and down the hill?
And if that's the interpretation, I don't necessarily agree with the next point that obstacles are reduced on slopes/hills... so I probably am not following correctly :-/
Thanks!
However that would mean that everyone would have access to it.
The idea of self driving cars now is, a winner take all situation where whichever funded effort that succeeds, generates outsized profits from licensing or going public at a high valuation.
Human drivers don't follow the official lane markings because they can't be seen. They follow the paths in the snow everyone else has packed down. These paths often diverge from the road markings or any sort of absolute positioning system.
Road conditions can easily come from anywhere. Weather radar is at least good enough to know when roads might be wet or cold. Making roads smart enough to sense oil spills or even wetness would be incredibly hard.
Knowing where the road goes is certainly far better done by cellular. Connection to each segment of road would be fraught with hard to repair problems. Traffic conditions likewise are far better done from somewhere else, and cars would be much more able to see things on the road etc.
The only argument I can see as at all reasonable is that locating cars is difficult, and doing it with vision is incredibly challenging. You may not be aware how much GPS has improved. With a good view of the sky you can get (somewhat slow) accuracy to about a foot. Realistically that's just as good as you could possibly expect from a roadside device like RFID, bluetooth, or induction. The last inches may be important, but billions of dollars spent burying things in the road will not help.
A signal embedded in the pavement wouldn't be subject to wear, but it would make adjusting lanes much more difficult (if you've driven in the bay area, you've probably noticed lane lines moved back and forth for construction pretty regularly) and it would actually be worse in the snow/ice --- a consensus lane appears which may not follow the marked alignment, and following the marked alignment would involve driving over accumulated snow and ice.
That will affect how outsized the profits will be.
Also, because of politics, I expect there to be separate geographical winners, at least in China and not-China.
Now imagine we're doing that 1000 times per mile, like what you're suggesting. Even if the device is free (both initial cost and maintenance) it's just too cost inhibited.
I think we shouldn't assume that the reason is some malicious intention behind "However that would mean that everyone would have access to it." It could simply be that the idea is not economical.
Those are installed with a massive handtruck saw while cutting off a lane of traffic all day. It takes a half dozen people working on-and-off, doing different jobs. It takes jackhammers, high pressure water from a tanker truck, and people to place and wire the sensor to power. It's way more complicated than anything you'd do at scale, because it's infrequent and the goal is to do it thoroughly with non-specialists.
If you were installing millions of tags you'd have a dry drill that could go off the back of a truck and place+fill a tag in ten minutes. If you had a line you'd have a it hanging off the back of a truck and place it continuously, like a street cleaner or edge clearer. For tags there's no reason they'd need more than one person to place and no reason to even put them in the road when they could just go on the edges. Triangulate with directional antennas or something.
That said, I think it's pretty obvious that locating the roads is by far the easiest problem for self driving. If you wanted to make a serious attempt you'd want every car to broadcast a short range location, and to share data over a mesh network. "Knowing where the road is" to precision RFID would give you has been solved for over a decade with GPS and digital maps.
Much more pressing issues are non-obvious sensing like hearing a car around a blind corner or knowing when to be cautious about moving. Knowing when something is coming onto the road or when a vehicle is having a problem. Inter-vehicle communication is just so obviously important to that... it's really frustrating how vaguely it gets talked about. I don't give a shit about teslas coordinating braking so they can form a tailgate train for efficiency, I want cars of all kinds to be warning each other about what they intend to do. I worry that legislation or at least a regulatory body will be the only way to even get people talking about it seriously.
Other than that, cameras watching for intrusion into a road would be easier than solving it from vehicles. It seems patently ridiculous putting cameras to watch every 50' section of road. 1080p+ cameras, simple detection, and mesh wifi can be built in a $30 package... but there are >2.5 million miles of paved roads in the US. 30$ per 50' would cost, bare minimum (and ignoring electricity requirements+labor+the pole to put the cameras on) 8 billion dollars.
Would be curious to know the current work in this area.
I can see this happening gradually, beginning with docking stations, followed by parking spots with wireless charges and then downtown roads and highways. With each step reducing the required amount of space destined to parking.
The tags, whatever form they may take, will designate the official lanes. But people in snowy climates don't drive in the official lines because they can't be seen. They drive an emergent set of paths where everyone else drives. It's often, if not most of the time, that these paths human drives take don't follow the actual road markings.
He's poo-pooing level 5 autonomous driving and says that they just about have a level 4 autonomous vehicle?
This article makes no sense.
Honestly, the CEO of my company has very little idea of the details of the technology that we produce. If you picked some cutting-edge technology that isn't key to our market-share yet, he'd have even less of a clue.
Nobody's who's contemplating a level 4 is holding out their purchase for a level 5.
Level 5 basically removes the steering wheel so you never drive. But once you're at level 4, almost all of the hard problems have been solved.
https://www.sae.org/news/2019/01/sae-updates-j3016-automated...
He has no software background... he specializes in making gas cars go fast. He has no idea how to make a self driving car.. and honestly, why would we think he would? He hasn't exactly been pioneering vision systems, or anything else related to the tech.
Yeah, because making Formula 1 cars "go fast" involves no software whatsoever, right?
I love how you dismiss "Formula 1 engineer for many years". Maybe if he helped build a payment processor we could take him seriously.
>He has no idea how to make a self driving car.. and honestly, why would we think he would?
Because he's running the autonomous car division of the biggest automaker in the world? I have no idea if he's right or not, but dismissing him as not knowing what he's talking about is hubris. He's trying to build this stuff.
(a) He was the Head of F1 Development and Advanced Technologies which involves significant software exposure. F1 has a ridiculous amount of software both on car and in the factory to process all of the sensor data and design future cars. I've first hand seen their streaming big data stacks and they are on par with anything you will see at a top tier startup.
(b) He was the Head of Product Design for Apple's Car and we know they were going to have autonomous capabilities as well as hundreds of ML models powering AR, Maps, facial recognition etc. If you think you can get away without deeply understanding software whilst leading a major Apple project then you really don't understand how that company works.
(c) He's the CEO of VW Autonomy. He doesn't even need to be an expert in software. He just needs to be able to listen to his engineers.
https://www.linkedin.com/in/alexander-hitzinger-2163a035
It's called a sceptic. Almost every great true visionary had them. I'm no Musk fanboy but he's kinda like the modern day epitome of everyone saying no this isn't possible and it happened anyway. And to add insult to injury he started a rocket company kinda simultaneously. There absolutely is truth to scepticism literally blocking the mind. If you will it enough it can happen is a thing. If we all just put our heads down on how complicated that would be, I don't think we'd be very far.
Maybe he's right, maybe he's not. But either way it'd be dumb of him to claim level 5 autonomous cars were right around the corner.
Agree completely. It would be suicidal (in a business sense) to say that Level 5 was right around the corner right when you're on the verge of rolling out Level 4.
This just reflects VWs ideas to accomplish the goal... ie: they have none.
Let me know when Google says it's not possible.
And since he recently worked there I am sure he knows quite a bit about where the industry is at.
Citations needed. I consider Waymo, Cruise and couple of other companies in the top tier.
Disc: Googler
I mean all we have to support that they're working on self driving cars at all is DMV records, articles, and pictures of their car efforts. We have no idea how far along they are.
https://www.cnet.com/news/alphabet-google-waymo-ceo-john-kra...
[1] Yes, I know about the accidents
The big thing I would want self-driving capabilities for is long cross country driving. A thing Waymo doesn’t offer even to people in it’s service area.
Tesla has hundreds of thousands of cars which range over the entire country and has driven millions of miles doing exactly the sort of self-driving I most care about.
The two barely overlap in terms of offerings so I’m not sure what the point of comparing them even is.
It's personal & subjective, and nobody will persuade anybody in a thread, but FWIW I absolutely positively can.
- I'm unlikely to utilize self-driving any time soon
- The stories of their firmware terrify me. And I don't want a DAW and computer games running on my ECU :O
- More pragmatically though, the Tesla UI paradigm is completely foreign to my way of driving/thinking.
I'm looking for a "HOTAS" type UI, where I can do anything I want without taking my eyes and focus off the road. A UI that's one giant screen, that may change position of buttons from minor firmware to another, is basically as scary and alienating concept as I can imagine.
I get that I am in a minority nowadays - a lot of manufacturer's are replacing switches, buttons and knobs with a touchscreen and deep menus, but not thankfully all just yet :|
Honestly - these shouldn't be "either or". Sure, have a screen and voice for those who prefer - but leave a button or six for us ol' timers :-)
Radio and other such features do not need dedicated buttons. Crazy how many buttons my normal car has.
Next, Previous, Pause, Play, Mute - I personally want them to be physical controls. On steering wheel ideally, on the dashboard otherwise. I use them multiple times a drive, and I don't want to take my eyes off the road to do them.
Same with seat heat, lights, wipers - anything I may want to do while driving, I want a button.
Setting up the exact shade of my dashboard light - that can be buried in a menu :D
Basically... when you say "Crazy how many buttons my normal car has" :
- You say that as a bad thing
- I see that as a brilliant thing... IFF done well:
Of course, physical buttons/levers/knobs can still be done well, or poorly.
Having many identical buttons in a confusing layout is just as bad as touchscreen - I have to look at them to use them.
Having buttons in a good, intuitive layout; especially buttons which are distinct from each other, as opposed to row of 6 buttons all the same, is brilliant. Even better if it's a distinct combination of buttons, knobs, switches, levers, etc - anything to help haptic feedback and intuitive access. Sometimes I think people who are against buttons may simply never had a car with good physical UI:/
(simple thing - my old 2004 WRX has a next / previous knob-like-thing, rather than two identical buttons next to each other [1]. It felt ridiculous when I first saw it - but then I realized its quality of purpose vs sexiness - I never ever ever have to think or be distracted even a millisecond to know exactly how to skip a song :). Compare to cars which have several identical square buttons for next, previous, pause, play; or temp up, temp down, fan up, fan down, A/C -- that's just horrible UI by clueless people for customers who don't know / haven't experienced better :-/ ]
1: Bottom right of the stereo: https://images.crutchfieldonline.com/ImageBank/v200311131204...
I wasn’t trying to paint a broad picture because as you suggest, it’s subjective. But there is some evidence that I’m not alone.
https://www.wheelsjoint.com/tesla-model-3-is-wreaking-havoc-...
> I'm looking for a "HOTAS" type UI, where I can do anything I want without taking my eyes and focus off the road.
I agree, the cockpit design on the Tesla is not my choice and may actually be a deal breaker. But since there is a 30 day no commitment period I’m willing to give it a shot and see how it works in actual use.
Oh, absolutely - as I said, I'm in a minority.
But I feel a lot of it is: Sexiness of how it looks vs Practicality of usage
The problem is, sexiness is immediately obvious & attractive. Practicality takes a long time and active observation to notice and appreciate. I fear that touchscreens with bad UI will win; though I hope eventually there may be some backlash from consumers - or at least thought and compromise from manufacturers. :-/
Aside from the fact that the Tesla is not available where I live, if I was only going to own one car, it would not be a Tesla. A Tesla (or any other EV, really) does about 80% of what I need a car for, so it would be perfect to buy as a second car for me, to be used most of the time and have the other car be a backup for when I really need to.
If I had to settle for only one car, it would probably have to be something like a Chevy Volt, which I am quite sad got discontinued. Even then, that would get me about 95% of the way there.
If my non techy family see the Tesla performing what they deem to be self driving then they will quickly trust it 100%.
We've had autopilot for decades, we're certainly not flying planes without pilots. Or even getting them from runway to gate without humans. Nor is anyone claiming that pilot-less planes are coming soon.
The thing is, there are some conditions where people probably just shouldn't be driving. Like we shouldn't be driving in a heavy snow storm, at least not for most trips where you can just wait it out. Planes will wait out storms, even with human pilots. I don't see a problem with "level 4.9" cars that occasionally say "I'm going to wait out the storm" or that will avoid certain routes (such as "crazy mixing bowl" intersections where you can instead just take a side street)
Guess when they take control? When something bad is happening (and there are literally a million things that can go wrong and the autopilot could not fix, ever).
Guess why MCAS did not result in more crashes (the system has been known to malfunction a couple of times before the crashes)? Righty because there were pilots that could manually control the plane well enough to land safely.
I imagine there’ll also be manual modes to get the car into trucks and ships, or even just out of the factory.
This is definitely not consistent with statements Alex has made in the past. Seems like more of an off the cuff remark a German engineer would make while confident that the fully realized result is right around the corner.
Most of the value in being able to travel by auto without attention is in trips longer than 15 minutes. Commutes, tourism, vacations. Shipping. Most of which is certainly highway driving.
If it's possible to automate highway driving under most conditions (and safely transition either to a stop or human-piloting when those conditions aren't met), then at least 80% of the value is there.
Wrestling with the harder edges of the problem is still the right thing to do for tolerance reasons, but I hope we don't have to see last-mile problems solved before we start reaping the benefits.
Self-driving cars that only go 80% of the way would compete against those forms of transportation which have established business models, networks and in the case of trains are subsidized in a way that self-driving cars likely won't ever be.
Additionally the key advantage of cars is flexibility, not having to figure out how to get to or from the train station or airport. In your scenario self-driving cars would lack this advantage.
Now granted this may still sound like an appealing proposition in countries with poor public transportation infrastructure but that substantially reduces the size of the overall market. Sure this might be attractive for the US but what about Europe? Asia?
Now of course there is an obvious answer to this problem: Keep the steering wheel and drive the rest of the way yourself. That just means you'll have a lot of drivers who won't gain experience at the current rate though. Not sure I would like to share the road with those people.
To the extent cost is a a factor, it's unnecessarily high due to these aforementioned issues[1]. So I don't think money invested into self-driving cars hurts trains and rail infrastructure.
[1]: https://www.vox.com/policy-and-politics/2017/1/1/14112776/ne...
In the US, trains are for freight. It would cost too much to create a new passenger rail network compared to automating existing intercity roads.
Now, increasing the throughput of rail using self-driving trains could make a bundle in the US.
[1] https://en.wikipedia.org/wiki/Automatic_train_operation [2] https://en.wikipedia.org/wiki/List_of_automated_train_system... [3] https://www.railwaygazette.com/australasia/rio-tinto-complet...
Of course, I live in a city, where parking is a major pain. But the other things still hold. It it far more efficient of resources to have cars be used most fo the day, rather than sitting idle.
Think about Uber cars today. Their entire value proposition centres around them breaking parking laws in order to pick you up. Now the liability for that behaviour is with the driver. In an autonomous car the liability would be Uber which is why they won't do it.
An Uber that needs to look for parking in order to pick you up is never going to work in most busy cities. So I suspect you will see people opt for the driver based taxi service instead. Or maybe autonomous taxis will be reserved for emerging markets e.g. Indonesia which are more relaxed about parking.
Anyway, "people are willing to break the law but companies aren't" seems a really weird reason to stick with the status quo over something far more efficient.
Drivers own the liability for illegal parking. Autonomous cars would make the developer liable.
No government is going to tolerate systemic and wide spread violation of the law.
It's just like self driving cars will go a bit over the speed limit if that is the flow of traffic. If they don't, then the laws will be adjusted. It would be silly to indefinitely hold back a huge industry because humans get to exploit legal loopholes but machines don't.
I take Uber and Lyft in NYC multiple times per week. No one ever “parks”, they just pull up, you jump in, and off you go. Takes less than a minute in most cases, doesn’t block traffic, and I suspect it virtually never results in a ticket. Maybe if they sit there for 10 minutes or more, but that’s exceedingly rare. And even then, they’re not going to get a ticket outside a few ultra-congested areas of manhattan that make up a tiny fraction of this city. Even in most places in manhattan, you could sit in a running car in a no parking zone for hours without getting a ticket. I had a moving truck up on the curb on the Upper West Side in a no parking zone for hours and the NYPD rolled past several times without a glance.
Also, this is exactly how taxis all work too, btw, which seems to undermine your entire point.
The vast majority of pickups in nyc at least are not illegal under any current law.
Similar laws simply aren't present in most other cities. Stopping in the middle of the road to pick up passengers is illegal and drivers do it since the police are never there. Hence my point that Uber would never program this illegality into their self driving algorithm.
Autonomous cars won't have the same algorithm as selfish humans to position themselves.
A human will go to one of the busy places and will probably drive back to the original point of pickup.
While a computer will be notified ( 1 car only parked on the busy place) and immediately replaced by a new available/nearby car. Making the process more efficient, more remote places will be handled faster. Busy places more efficient.
Not all cabs will need to earn money, since the biggest expense will be the car driving and not the human driver ( waiting for work).
And as always, if income > fine. They will just pay the fine, some companies even negotiate it because of the sheer amount.
Up until the point they are banned.
Everyone, especially lawmakers, are pretty sick of tech companies breaking the law in order to be "disruptive". As we've seen recently with all of the e-scooter bans.
Scooter bans were mostly because it was disruptive having a staple of scooters in the middle of the pedestrian road and they were consistently breaking the rules.
Ps. We do similar stuff for cities at our company. So I don't need an externzl source for this currently.
I’m imagining how gross a taxi would be that doesn’t have a driver to provide passengers a low level of behaviour moderation.
For "free floating car sharing" (DriveNow/ShareNow, Yandex.Drive, ...) this seems to work quite well.
If people have to pay the cleaning bill they will be careful not to make a mess.
So riders blame the mess on the previous passenger. Then the taxi company installs cameras. It’s all gross (and yes, I’m aware many taxis have cameras already).
Self driving taxi will be a new possible way to get from A to B, and people will chose their own trade offs.
Probably human driven taxis will become the luxury eventually.
And often that is the hardest thing to get rid of.
I've talked to a number of uber drivers. The cameras recording the passengers are a much greater inhibitor of bad behavior than the driver.
This is better and more easily addressed by car sharing services like car2go than autonomous cars. The more I use them the more I wish for a world where every car was shareable: you need to go somewhere? you simply get into the nearest parked car and drive it to where you want to go. And, unlike autonomy, there are no technical barriers to this, only the will to do it and the economic model to sustain it. Sadly these services seem to be in full retreat now but I will miss them and still hope they'll rise again soon.
This is not the conclusion I would draw from car2go failing out of North America. They still had the problem that you are required to find a legal parking spot and there's no guarantee that there will be a car available anywhere near you. Taxis and private vehicles solve the latter and former respectively but they all suffer from the inefficiencies inherent to the system design.
[1] https://mifold.com/products/mifold-the-grab-and-go-booster
Studies have shown that the growth in Uber/Lyft has increased traffic congestion. This system autonomously will be even worse.
https://edition.cnn.com/2019/08/22/us/traffic-commute-gridlo...
They can also, if Level 5 becomes a thing, reduce traffic collisions. And if they all end up electric, through route optimization they can significantly reduce energy consumption requirements for vehicle usage.
https://www.geotab.com/blog/traffic-congestion/
Given the same number of cars on the road, a self-driving fleet would eliminate many mechanical failures and many human failures, and be able to adapt to sub-optimal infrastructure at a network-wide level.
Infrastructure would also be easier or more efficient to improve, because you'd have removed much of the human variability that makes identifying choke points difficult.
AI doesn't really have the power to change that, and might actually be worse depending on how it reacts to pedestrians and cyclists.
Self driving cars can plan the whole exchange, negotiate (or be directed by an intersection controller) to decelerate, align, merge, and maintain high throughout at low latency. ATC does this for planes already; we have the coordination technology. It’s just about getting drivers to listen to precise instructions without deviation. For everywhere I know of, that’s going to take automation.
Reading up on standing waves taught me a lot; you might like adding that wrinkle to your model.
That's only going to work on roads that are AI-only.
I don’t see an AI requirement to be a huge hurdle for those cities.
In modern Western urban planning, large car-capacity streets aren't considered a good thing to be running through urban areas anyways. Most of the large cities (Paris, Berlin, New York) are taking steps to reduce road widths to more equally distribute the road between pedestrians, cyclists and transit users. As it turns out, building pedestrian overpasses at a convenient enough spacing for pedestrians is extremely expensive.
So if AVs can't increase the throughput of city streets, I'm skeptical that they can increase the throughput of off-ramps which are bounded by city streets, or urban freeway segments which are bounded by the off-ramps. And even imagining that significant (2x?) throughput is achieved, it's not going to meet the induced demand ceiling; there would be the same amount of congestion, only with more cars.
L5 is dead on arrival as congestion-mitigation technology and I hope at least some of the billions earmarked to be spent on researching and deploying it are redirected towards better walking, cycling, and transit amenities instead.
Neither would they have to operate at artificially handicapped speeds, as no human operator would really be able to keep up regardless with the 10,000 or 100,000 cars running around.
Part of reacting and mitigating acute issues is being able to reroute traffic early enough, or slow cars down early enough, to minimize disruption (instead of a traffic slinky blocking everyone up).
There's no reason so far here that self-driving vehicles can't improve traffic and overall throughput, and it's a bold claim to declare L5 dead on arrival.
Follow distances in urban areas with low speeds are already really small. I don't think AVs would make them meaningfully tighter.
> Part of reacting and mitigating acute issues is being able to reroute traffic early enough, or slow cars down early enough, to minimize disruption (instead of a traffic slinky blocking everyone up).
This was supposed to be what Waze and Google Maps were set to achieve, and they did not, but they did piss off a lot of people when cut-through drivers started jamming up their formerly quiet residential neighborhoods.
Imagine highway where everyone drives 80mph with 5ft of distance between cars. No one suddenly changes lanes, no one suddenly slows down for no reason.
Here is illustration from Waymo testing facility: https://youtu.be/hKfEivMfDPU
[0]: https://en.wikipedia.org/wiki/Two-second_rule
- selfishness eg stopping or parking in no parking zones, driving aggressively
- error eg causing accidents
- ignorance eg not knowing the traffic flow in an intersection, driving slowly to find your bearings.
Also, if we have fewer cars, but higher utilisation of vehicles then we will require less parking.
They might even Uber to the train station then Uber from the destination train station to work. Lots of ways of cutting traffic!
The people who are driving, or at least a significant portion of them, pick driving over public transport because of perceived privacy and control.
I don't think uptake of shared mobility is going to be so popular where that becomes the norm.
Also I drive because it's faster, and because once you go from A->B getting to B->C is probably easier. Not for privacy. Most people just need to get to work or school or their kids to school for the bulk of their car use. Kids to school might work by sharing with other parents you know.
Most Americans don't really need an SUV or a pickup; a minivan or a sedan are more practical. Yet most Americans are buying them, because it makes them feel better about what they're driving and they want to show off. You may only take your car to the grocery store, but if you meet Becky from PTA in the parking lot and she has a really nice car that she owns, that's still a big deal.
Unless the ownership cost of a car were to spike significantly, this part of American culture seems hard to change, particularly when automakers spend lots of money making sure that Americans see premium gas-guzzlers as status symbols.
Half of the work on "self"-driving is actually traffic dispatch.
Given that planes, which rely on a class of specifically trained professionals with minimum hour requirements, still have issues with this from time to time, I don't think this is a realistic thing for cars.
Driving standards, at least in the US, are very low, and raising them is a hard problem.
Most of this issues come from the fact that planes can't just stop in the middle of the road and wait for help, they have to land somewhere and do it fast enough not to be out of fuel.
Cars don't have to drive themselves to nearest service center when they encounter any hardware malfunction, they can just stop and unload passengers. Self-driving company would use another car to drive you to your destination (not even necessary self-driving car).
The problem with complete automation is how accidents are perceived. Probably a good thing considering how low the bar is right now for road safety. Being killed by a robot's mistake is perceived far worse than by human error.
I think a middle ground like autopilot on the highway could significantly improve safety while still accommodating perception issues.
I don't think so. Airplane autopilot automates away the safe, boring parts of air travel, much like highway driving, so in that sense I agree with you. Truly autonomous landing is still a big deal, and planes that the benefit of ATC, and I don't think there are any planes that taxi or takeoff autonomously. Furthermore, accidents involving autopilot are always at the edge, when control changes from computer to human unexpectedly or when the human has to take over the autopilot in less than ideal conditions. I don't think there's much perception in the public eye of "robot pilots killing people"
I think once autonomous driving accidents go through the courts a few times, the same will happen to cars. After another decade of development and legislation, people will die not because the robot made a mistake per-se, but because it encountered a situation it can't cope with and the human wasn't ready to take over in an emergency, or because the robot tried to cope with the emergency situation the wrong way and the human didn't notice in time and took over too late (if at all). This may still sound like the computer making a mistake, but the subtle difference is that the mistake is made in an extraordinary situation, as opposed to the current accidents that happened in regular driving conditions. I think this subtle distinction will be enough for the PR people to spin the blame away from the manufacturer.
Call my cynical, but I can also see successful lobbying efforts in the future by corporations to reduce liability in class-4 and lower autonomous car accidents involving unusual situations, citing the fact that the driver should have been paying attention and taken over.
That aside, road safety is better than ever these days, with a lot of what would have been fatal accidents being merely property damage due to modern safety systems, and highway driving is probably the safest kind of driving out there, thus it is not clear to me (and I have no data either way, I don't think a study has ever been done) that current offerings (Tesla et al.) are any safer compared to similarly priced modern luxury cars per mile driven, and I'm not sure that will change in the future. The only reason I see for adopting highway autopilot is not safety but comfort. You might argue that comfort contributes to safety and enables longer driving times, but driving for 8 to 12 hours is pretty safe regardless, and I don't think I could do more than that even if I was a passenger.
This is a bit of a controversial opinion, but I don't think we will see a meaningful reduction in fatal accidents or serious injuries due to self-driving cars for a long time, but I do think we will see a reduction in accidents involving only property damage or minor injuries from low to medium speed accidents, and I see a sharp decrease in parking-lot scrapes and dings in the near future. Low to medium speed streets and tight maneuvers is where I see most "bad" human drivers have the most trouble and already things like parallel parking assist are a godsend to these people.
If anything, I think your example underlines just how valuable and realistic this is for cars.
Full autonomy was always, IMHO, a silly pipe dream. But that doesn't mean that there isn't tremendous value -- in terms of safety, time, money and environment -- to partial autonomy.
I interpret the stats the other way. Cars are not planes and driving involves constant encounters with potential collisions and which would make level 5 too difficult technically and socially rendering them too costly to be valuable.
But one needs to plan and book a travel in advance. There are also capacity limits.
The way, the channel, the [1] https://en.wikipedia.org/wiki/Erlang_(unit)
all there to calculate whodunit in advance, this is no Voodoo, just calm glance at what is there, to not get stuck, that would suck, and be unfair.
that to me is something automakers can easily cooperate with governments to get done and done well. they already mark HOV and Express lanes and with federal legislated lane markings it would make automating those drives pretty simple. Then as the number of cars increase the number of free drive lanes is decreased until the vanish.
using metro Atlanta as an example express lanes have their own entrance and exit points so they are already separated. One set for I75 is wholly elevated two lane bridge work nearly the full length of its run.
Tesla Model 3 owner, I think even Musk is walking back what he "implies" as self driving. As into self driving with human oversight
It would be the end of owned cars, huge reduction of car production, millions of jobs in the transportation business amortized, to name just a few.
Bricks from bridges could kill people and just damages regular people's property. It's just different.
Behold! https://c1.staticflickr.com/8/7177/6894934663_0619c8bea3.jpg
Bonus: the vehicle can be resized at will, depending on the number of passengers.
But otherwise agreed a ton of this stuff could be solved with modern trains.
But appeasing all of the special interest groups in between has made infrastructure development impossibly expensive or takes a decade. So we keep driving cars even long distances between cities.
The former government owned national railway basically owns every piece of important track and doesn’t give a damn about customers. The prices are constantly rising while the service declines. Can’t count how much time I’ve wasted on several train stations, waiting for up to two hours for late trains, hell, even having to get a hotel multiple times.
https://en.wikipedia.org/wiki/The_Roads_Must_Roll
I'd say that, in the majority of cases, it's either no more difficult or outright easier than getting to an airport, as airports tend to be located on the outskirts of cities (requiring connecting train/bus service or private cars), while most rail hubs tend to be in city centers.
(compact towns & bicycles)
The problem with trains is that they are very, very inconvenient for most journeys. Sunday evening? forget it. After 11pm? Nope. Journey that runs perpendicular to the local line to London? Lol, enjoy a 5 hour journey to go 100 miles. Want to take cargo like a new washing machine? Lol. Want to go somewhere that's not near a station by bus? Enjoy Journey times that are 4, 5 or 10 times slower than a car.
Trains are good for busy commuter lines and nothing else.
You're blaming trains because you want to have a boozy night out? Get a hotel or start and leave earlier like everyone else.
This is just a silly comment.
But to stab at that anyway, it wouldn't be silly if he did. People driving impaired for lack of other good and affordable options is an emergent reality, no matter how morally hazardous it is.
Of course sometimes you dont want to travel late or at weekends
2) a train doesn’t need to replace 100% of personal car trips to be useful. I’m sure you’re not transporting dishwashers or taking trips after 11 pm every day. Covering 90% of use cases is sufficient, and the other 10% can be handled by either renting a car or taking an Uber.
3) we can and should make busses faster than cars during times of congestion by giving them their own lanes.
There are no commuter trains Tasmania.
Urban commuter trains will have little effect on that ~45% of people who won't live near them even if they are built.
On the flip side, though, I guess, that doesn't mean we shouldn't build more train lines.
1. https://www.un.org/development/desa/en/news/population/2018-...
Consider Singapore, which is tiny, has a kick ass public transport system (trains and busses) and where it costs 90-100k SGD to own a beater Corolla: people still own cars.
That said, Singapore public transportation (at least SMRT trains [2]) is freaking strict about food and drinks. I could not even have a sip of water without triggering a warning message. So anyone who can't control their thirst or hunger (like, you know, basically any kid) isn't really welcome there.
[1] https://www.budgetdirect.com.sg/car-insurance/research/car-o...
[2] https://www.smrttrains.com.sg/Journey-with-Us/Travel-Informa...
If you are going from one city centre to another, during the day, the train is the best choice.
If you are moving within a bike-friendly city, bike is best (or Uber, but it is banned here). If you are somewhere in an urban sprawl, it's car or Uber.
I definitely find trains useful, but I think that they are consistently overrated by people for emotional and irrational reasons. I personally use trains for maybe 1% of my journeys and trams for 0% despite them being extremely prominent in my city.
- where you want,
- when you want,
- carrying what you want,
- as fast
- and as safely as possible,
- at the lowest cost possible.
Trains are weak on multiple axes here: they do not go where you want to go and they do not go when you want to go, and you cannot carry large amounts on them. They are also quite slow.
The only reason that anyone uses trains is because in dense urban areas like London, cars are even slower, and for intercity trips it is both slow and hard to park when you get to the destination city. There are perhaps some exceptions to this but they don't really exist in the West. Really fast intercity trains are a thing in China and Japan, but the UK has no trains that take you from A to B at an average speed greater than a fast car once you take waiting and stopping time into account. As far as I am aware the same applies to the US as well.
Trains do not cover 90% of use cases, they cover maybe 10%. In my opinion, trains are a technology that should be completely scrapped in favor of driverless minibus and driverless car networks. Modern electric & driverless vehicles with internet connectivity and global transport optimization like Uber would beat trains on every single axis.
The last train I could take on a Friday night was 10pm, but given it was a 3 hour journey that didn't seem unreasonable. Each year there was usually one incident that caused delays, but other than they worked well. If I took a fast train the journey time was around the same as driving - quicker in fact given on the journey I would stop for a break - and ticket prices were about the same as driving (for 1 person).
On the other hand I was working in London in 2018, and at weekends often flying from LTN. In the end I added an hour buffer (to a 45 minute journey) because the reliability of Thameslink was so bad.
On top of that, I personally (and know many agree) HATE the stress of keeping track of times, scheduling around infrequent departures, logistics to/from stations, etc.
> Bonus: the vehicle can be resized at will, depending on the number of passengers.
Yeahno. "please wait 15-30 minutes as we shuffle cars around the rail yard" is not exactly "at will"
I gladly pick rail over car for good lines, because it's stress free. Get to the station in time (with public transportation it's quite deterministic, subways and trams don't get into traffic jams, buses sometimes do).
Cars of course win when it comes to flexibility (late night getting to a small town address).
Highway-autonomous vehicles would give you the benefits of rail service plus arbitrary departure times, more direct routes, on-demand route changes, and you're traveling in a vehicle that you can use for first and last mile and intra-city travel.
The value in the 95/95 autopilot is because it will make commercial drivers be able to never need breaks (they sleep or rest until the autopilot is in trouble, and they can even remote control cars). And of course because a lot of people seem to hate driving and would pay to avoid it.
I don’t - I hate owning a car.
A car that can only drive within major cities in the UK and on motorways would be entirely fine for the vast majority of people. Exceptional journies I can always rent a car that suits my needs for the one week in several years I'm in the Highlands.
People will abuse that. They will get high/drunk/whatever and when that “5%” rolls around (quite unexpectedly mind you), they’ll run into a bus full of nuns. Always assume people will be judgement impaired because “the car can drive itself”....
But, this is also why I think the next car manufacturer race will be about who can solve this for their customers. That is, who solves the problem of making 95% autonomous cars able to drive in 99.9% if all situations without manual intervention by the driver, where the provides automatic remote control from massive rooms full of people steering drunk car owners out of situations the autopilot couldn’t. The better the autonomy, the fewer people needed to remote override.
A 90% autonomous trip where I have to pay attention 100% of the time gets me no real value. It's just a slightly nicer, maybe, ride.
It seems unlikely you commutes would completely autonomous even people commute on freeways.
Spoken like someone who does not drive regularly in snow.
This way, you can set out home with cars and you only operate the last part going to Cities for work. And when you leave you drive out to high way and then every full else will be automatics. It also partly solve housing problems and people can now live further away from work.
Surely this is a solution that is simply enough for the short term. What am I missing here?
I see the future diverging into two paths, fully autonomous commercial vehicles, like taxis, delivery vehicles, semi-trucks, etc. that work within urban areas or other designated, mapped and specially prepared areas. This will possibly involve a centralized system of control and communication, something like an ATC but for cars. These will be owned by corporations and only used by people. It makes no sense for a person to buy one of these fully autonomous vehicles, although I imagine some people would pay extra to have priority access so they always have one available.
The other side of the coin will be privately owned cars that have autonomous capability, or autonomous cars with override. These will be able to go anywhere the driver wants, including unmapped villages, small towns, off road tracks, etc. I suspect these will be the domain of enthusiasts, people who really need them for work (ranchers and farmers, for example) and people who choose to live away from urban centers. They will be more expensive than cars now, but the need for these vehicles will never go away. Even if we get true AI capable of driving anywhere with only the sensors aboard the vehicle, there will still be the need for a human to override it, even if that involves just authorizing a risky maneuver or putting the AI into "unsafe driving" mode.
I think the movie I, Robot (with Will Smith) got the future of autonomous cars surprisingly right, autonomous inside cities and on highways, and using the manual override comes with penalties (higher insurance, being at fault in an accident, etc.).
On a personal note, and this may sound bad, but I would never buy a car which I cannot use to break the law. Even if I never plan to do it, being able to speed, jump the curb, intentionally crash into a wall (or another car) or even run over a person (for example, in self defense) may be at some point required or the least bad of many bad options. In this case any consequences should fall on me, but I don't think a thing I own should be designed to prevent me from breaking the law or doing something stupid if I really want or need to, although providing warnings or an optional safe-mode is fine. I suspect many people feel the same way, even if they don't put it in such an extreme way. This can be seen by the fact that a lot of cars, particularly those focused on performance or off-roading, come with switches to turn traction control off, and if they don't, it will get mentioned as a negative in any review done by publications focused on those audiences.
I mean, sure, you can come up with some scenario where the liability exceeded the insurance coverage, but I haven't heard of many of those. Anyway, it comes from somewhere. If bankruptcy protects drivers, it also exposes them to the risk that they will suffer damage that isn't compensated.
Regardless, expecting a legal loophole to preserve the status quo indefinitely seems quite unrealistic and inherently unstable. If that actually holds up something that could massively benefit society (both economically and in saving lives), we simply legislate liability limits.
Liability you can incur while driving is almost arbitrarily high. Individual drivers rely on the existence of bankruptcy protection to cover these rare scenarios, or simply don't think about or plan for this at all.
> but I haven't heard of many of those.
How many do you think it takes to put a self driving car manufacturer out of business?
I'm not saying the status quo is a great situation, or that this is a good or bad argument for or against self driving cars. Only that it's a description of the current situation, and why legal issues might be a much bigger problem for self driving car manufacturers than individual drivers.
It's a different type of potential error and a much more scary one. I can mitigate human drivers as a pedestrian by taking care walking, but I cannot mitigate an AI that thinks my shirt makes me look invisible due to its learning being deceived by a pattern.
I'm not sure why people even embrace self-driving cars. By now we know that centralization, lack of maintenance, and fragmentation in software are serious risks, as well as how software can increase the attack vector on people as well as provide benefit. I'm not sure if the risks are worth the modest efficiency increase, and this isn't even getting into existential risks like external parties being able to control when you drive, or attacks on the networks or technology.
However, I think the profit incentives of the trucking industry will manage to carve out some regulatory exceptions; something like “freight trucks can self-drive between 11a and 5a on these specific Nebraska highways, with warning signs on both roads and vehicles”. This sort of lobbying will be the thin end of the wedge for both iterating the tech, and normalizing its acceptance.
Insurance solved that problem, both by eliminating the possibility of a catastrophic financial loss, and by creating a buffer between me and all those lawyers.
I predict that insurance will solve the lawyer problem for self-driving cars as well. At some point, it will cost me $5,000 a year to drive my own car, and $500 to let it do all the driving.
And on top of all that, if my car drives itself into an accident, the lawyers will talk to my insurance company, not to me. I see the insurance companies as the enabler for this tech. And they will want to enable it, it will put them in control of the market.
Indeed we can which is why some cities on the planet have gone carless and more and more opposition is mounting in the face of traffic deaths, pollution and so so on. Even in the US, maybe the most car dominated country in the world, there is a political revival of talks about high speed rail and alternative forms of transportation.
The other important difference being that the step from having no cars at all to having cars was one of the largest leaps in mobility in human history. Self-driving cars are nice, but not that much of a leap, and they have much more ambiguous implications when it comes to the job-market. They will face significantly larger hurdles with significantly less payoff in sight.
https://www.theguardian.com/cities/2018/sep/18/paradise-life...
With a self driving car you know exactly what happened in an accident as video and telemetry is recorded. This is a tremendous advantage over having to reconstruct it without this data.
On the contrary I expect legislation forcing every car to include telemetry like the Chinese are forcing every car to be connected.
This accident's data is evidence, not opinion, not a belief, not a prejudice.
The usefulness of this has already been proven with airplanes.
While self-driving is obviously a challenging problem, I am still bemused by people who think "it will never happen".
An auto exec who believes that, is unfit for the job and definitely unfit for the future of that job.
It's very easy for execs to talk up utopian smack, it's harder the other way around.
I for one agree: L5 may not be viable for many decades at least, not with the tech we have. It may never happen because conditions will change to the point wherein it won't make sense.
What is viable, that nobody really talks about are 'controlled areas' - for example, highways built for long-haul trucks with 0 human drivers. This is a great place for AI because the conditions are set for all the crazy unknowns (esp. humans) to be controlled.
Special roadways within cities, where there are no pedestrians, AI drivers only, that can communicate with 'the grid' etc - could not only mean full L5 'driverless' but no stop-signs either - traffic could be designed to flow much, much more efficiently using 'today's tech'. Cars won't have steering wheels, they can be called 'on demand' and 'parking' will be very different, more like 'temp storage' and won't involve humans. All viable.
This is a point that always seems to be missing. All of our roads are/were constructed at a specific size with specific engineering requirements and specific materials. If we have a major change in the technology of the cars, it seems to be perfectly reasonable that we could have a new type of road to accommodate that. For example, before interstate highways it probably would have sounded strange to imagine high-speed roads with limited entry and exit. And before that, may have sounded strange to have automated stoplights and traffic signals.
There we go...most of the skepticism about self-driving cars appears to be driven by ideology and less by the actual evidence. "Mainstream opinions"...it is not an opinion, there are literally thousands of these cars being tested/piloted and improved upon constantly.
Might it take a little longer to get to nirvana? probably, but it borders on delusion at this point for anyone to say it will never happen.
You can't provide evidence of something that hasn't happened, and level 5 autonomous vehicles haven't happened. All you can provide is evidence of things we have already achieved, and use that to support a strong opinion that we will be able to achieve more things.
Anyways, back to my original point, firing someone for having an informed counter-mainstream opinion is the sign of a weak and petty leader.
Level 5 means that the car can drive autonomously > 95% of the time, and it will absolutely be possible. Level 4/5 distinction doesn't really make any sense once the autonomy gets beyond certain percentage. It doesn't mean that it's level 4 until it hits 100% (which is impossible).
Level 5 car is designed to drive in all conditions, but there are always statistically unlikely corner cases or situations that require high-level decision making, which the car can't handle by itself. A single driver may never hit such case, and for them the experience is full self-driving.
[0] https://newsroom.intel.com/news-releases/volkswagen-mobileye...
People claiming most of AI hype barely know what the "AI" people mention is.
And people who go as far as drawing rosy pictures of human like general AI being your personal chauffeur are past ridiculous.
The entire idea of human-like general AI for practical applications is like trying to make people using horses for transport in 21st century, by trying to make a horse than if better than a car.
(full autonomy on some/many roads)
Also, while claiming these innovations are only a few years way may be insane, saying they will never happen feels probably just as non-sense. History hasn't been on the side of such absolute claims.
It will take time, but Waymo already has working 100% self-driving cars out in the wild. Yes it's in a small area, with perfect weather, and with technicians ready to jump in, but it's still a good start and easily paves the way for it to extend over the next decade or so.
All and every piece of electronics we did in the automotive sector had some lame "AI platform" attached somewhere, for no apparent reason.
The most blatant "AI washing" I saw was an "AI power steering" which was a plain PID, inferior on all fronts to purpose made systems already on the market.
Second to it was an "AI airbag," which was basically an OpenCV hello world that will trigger the airbag ahead of time. Terribly unreliable, and would've probably killed more people than saved.
I totally understand mistrust. Less so if there are somewhat capable alpha products already available with huge economic incentives behind made by somewhat reputable companies. It might not happen in the current hype cycle, then it will be a future one. But I don't ever see self driving cars quietly being deployed. That's just a way to well known game changer hardly anyone will want to sit out.
Waymo ceo made a similar statement.
The Waymo CEO was very clearly referring to extreme conditions like storms, and honestly, most humans can't and shouldn't drive in said conditions, so it's not that crazy to say Waymo may never be able to either.
I wish there was a collection of stubborn dismissals of things that did come to pass and became bigger and better than originally imagined, so we could point and laugh at those cynical sticks in the muds.
Steve Ballmer's dismissal of the iPhone being a popular relatively recent case.
I still remember that quote I discovered through Civilization, apparently made by some popular leader (Napoleon?) about steam engines: "You would make a ship of iron and without sails move by lighting a bonfire under its deck?? Absurd!"
Like we never wondered about self-driving trains, essentially. It just became a fact once we could do remotely supervized L4, and that novelty was short-lived too.
Driving a vehicle is not an AI problem and none of the serious autonomous vehicle teams are using AI.
Unless you are using a super loose definition of AI that includes anything with an if statement.
But that's just because the AI we currently have - basically idiot-savants made of silicon - is mostly useless to them, since the kind of AI they would need for actual, 100%, no-compromise level 5 self driving capabilities hasn't been invented yet, nor does anyone have a promising idea as to how to invent one.
I am pretty sure that the only way to actually reach this goal requires a full-blown AGI. Because what we call "driving" and what is often misunderstood as a rather mundane and repetitive mechanical operation of machinery governed by a few simple ground rules is actually an insanely complex task that requires several higher-level capabilities of the human intelligence: besides pattern recognition, quick evaluations of situations with an unbounded number of variables and application of knowledge and concepts to an unlimited number of ever-new traffic situations it requires foresight, communication (of the difficult nonverbal kind) and sometimes even carefully dosed brazenness - to know when rules should better be ignored or bent instead of followed.
This is incidentally one reason for Tesla's huge market value. The company actually has a plan to transition from individual ownership to fleet, so when this happens it will be prepared to deal with a new manufacturing reality.
Just try to imagine VW without all those ads to sell a positive self image because you drive a sexy cool car they make.
Cuts both ways. Tesla execs have the same reason to overhype autonomous driving, because their market value depends on being perceived as a hypermodern tech company.
So far reality has proven VW right. Elon himself had to walk back on grandiose claims about full autonomy and robotaxi fleets, or advertising cars with slogans like "the driver is only there for legal reasons, the car drives itself" a bunch of years ago.
Full autonomy on a human level requires human levels of intelligence and 'common sense', it's a ridiculously hard problem that requires several leaps in AI and plenty of other fields.
Why on earth would anyone use this over public transit? That would be a hugely expensive way to travel.
Due to these advantages, I usually take Uber/Lyft instead of BART or Muni. Autonomous vehicles would reduce the cost of such services even more.
I've been following this stuff casually since about 1980 when I totally failed to write an AI program and since then processor performance has been steadily increasing, first with clock speed and now more with multi cores. Hans Moravec, a robot guy did a reasonable estimate that to get equivalent hardware to the brain you'd need about 100 teraflops and just recently the first 100 teraflop GPU has come out (https://www.nvidia.com/en-us/data-center/v100/) which is kinda historic in a way, maybe a billion times faster than a cheap computer in 1980. Of course this will keep going for a while so the situation will flip from we can't do much because processors are much slower than the brain to the other way around.
(update NIDIA may have fudged the numbers a bit but anyway.)