One of the author's main criticisms seems to be that BMW et are commercialized (e.g. good looking and cheap) self-driving cars and Google is making ugly expensive ones.
But I'd argue that concern is backwards. First, go for correctness. Perfect the algorithm and sensors, get the car working. Then look at scaling down the technology and making it cheaper. The fact that Google doesn't make cars (a point the author made) is kind of irrelevant, because this is a sensor/control/software problem, not a car problem.
Recognize that BMW/Daimler/et al really don't want true self driving cars. It's not really in their interest, because it would radically reduce the need for car ownership and would open up a new world of on-the-fly car rental. Human drivers required = good business. I'm not saying its a conspiracy, just that they have no passion to disrupt their industry in this way.
Even the cost is somewhat moot if the business model is different. Let's say the Google Car sensor package costs $100,000. That's cost prohibitive for individual ownership, but it would not be a problem for a business model like ZipCar + Uber, where you call a car on your mobile phone to get you, and 'rent it' for a short self-driving or assisted trip. If car use moves from ownership to renting, then many people can spread the cost of the sensors.
Recognize that BMW/Daimler/et al really don't want true self driving cars. It's not really in their interest, because it would radically reduce the need for car ownership and would open up a new world of on-the-fly car rental. Human drivers required = good business. I'm not saying its a conspiracy, just that they have no passion to disrupt their industry in this way.
This is spot on. The car industry has done a quite wonderful job of transforming the 'mechanical horse', a contraption able to take you between locations with minimal effort from the occupants, into an extension of your personality and some form of lifestyle statement. The weak point in their advertising is that it is very often aimed at the driver and the driving experience.
Automated vehicles offer the opportunity for greater efficiencies which will translate into fewer cars as making use of a vehicle becomes much more common than ownership of one. Thus the price of purchasing these vehicles will remain high and many owners will need to ensure their vehicle is bringing in a decent income as some form of taxi in order to cover the cost of ownership (shouldn't be a problem as at the moment, most vehicles probably spend the vast majority of their time parked, waiting for their owner). However, I can see that if the whole system can be made to work as an uber-efficient taxi service most people will have no need to own a vehicle.
"Recognize that BMW/Daimler/et al really don't want true self driving cars. It's not really in their interest, because it would radically reduce the need for car ownership and would open up a new world of on-the-fly car rental."
This is exactly what is happening, without self-driving cars. E.g. Drive Now is a very popular service in several german cities, and one of the reasons why I don't feel the need to own a car while living in Berlin. I can just hop in a car any time I like, drive around, and then park it anywhere in the city.
> Recognize that BMW/Daimler/et al really don't want true self driving cars.
I highly doubt that. Self-driving cars are a revolutionary product. The first manufacturers to enter the market will definitely make a lot of money, as well as a technological advance on what is the future of cars.
They might be slow to invest in these technologies, because at this point it's still high risk and traditional car manufacturers are notorious for being very conservative, but there are definitely very high rewards for those who can make it.
Good smartphones (with capacitive touch-screens, and webkit-based browsers) are also revolutionary. But there's only two companies making much money out of them, and only one of them has been in the phone industry for long (and even Samsung was never that great at phones).
Cars don't just sell a way to get from point A to point B. They sell an experience. If that experience becomes "put your destination into your GPS, then play Angry Birds for a while", then it will turn the car itself into a commodity.
Right but BMW/Daimler/et al's company models are not based around the kind of massive 'moonshot' R&D that is needed to enter this market with a full self-driving car.
They do a lot of R&D but in a very structured way that produces the consistent results their shareholders ask for. The market might be disrupted but until that happens convincingly they will do everything they can to prevent it while adopting the low hanging fruits of automatic parking etc.
They might consider licensing the technology from Google or buying startups working on the tech but even then they might just bury the startups to delay their main business model being disrupted. I think it's going to take disruption from outside the main car manufacturers - in the same way it took Tesla entering the electric car market to shake it up.
> Recognize that BMW/Daimler/et al really don't want true self driving cars.
They don't have a passion to, but they must. Capitalism forces companies to adapt to competition, even when it isn't in their individual self-interest. BMW has to, because if they don't, Toyota will.
Companies make necessary self-defeating choices all the time. Newspaper companies have websites. Barnes & Noble sells e-books. The Empire made the Death Star.
I think Capitalism is a pretty broken concept, but I think this particular facet of it tends to work.
It should have read "Competition forces companies to adapt to market demand". Fortunately, there's a lot of competition in the automotive industry; less so in the production of LCD panels.
Well, they could start shoveling money over to politicians and lobbyists in order to create regulatory barriers to new entrants. Not exactly 'capitalism' in the truest sense of the definition, but certainly a common practice in our version of it.
Innovation does emerge from the competitive forces in capitalism, but it doesn't have to come from the old companies (and usually doesn't). B&N sells e-books, but it was a new market entrant, Amazon, that brought the innovations in e-books and readers
You've misunderstood. Automation allows greater throughput given a set road real-estate, but that does not directly transfer into more total cars owned.
In particular, if totally automatic driving is legalized, then many people will have a strong incentive to use on demand car services as they'll be cheaper and more dependable than existing taxi options. In parts of the world where labor is still extremely cheap this may not be the case, but it is the most likely outcome in the developed world.
And an engineering study published last year concluded that automation could theoretically allow nearly four times as many cars to travel on a given stretch of highway.
I am pretty sure it means what it means: the possibility of selling more cars.
> allow nearly four times as many cars to travel on a given stretch of highway
That's throughput not total ownership. A manufacturer doesn't monetize throughput. Moreover a study about allowed throughput is about the civil engineering constraints. It doesn't say anything about purchasing demand. Throughput can increase while total ownership decreases if average car utilization increases. The vast majority of cars spend the bulk of their time parked. Autonomy invalidates the reasons why those cars sit parked. If you work in software it's important for you to understand these distinctions. That or you're just trolling. I can't decide.
My point is that if there are less traffic jam, owning a car becomes interesting. I currently do not own a car because of traffic jams. Remove traffic jams and I'll buy one right away.
If you work in software it's important for you to understand these distinctions. That or you're just trolling. I can't decide.
You never know who is at the other end. In my case, I just felt it was inappropriate. But I could have been an insecure teenager and this comment could have hurt a lot.
You say that now but wouldn't you much rather pay a fraction of the cost of a car to own "part" of a car in a car-sharing program similar to NetJets?
> You never know who is at the other end. In my case, I just felt it was inappropriate. But I could have been an insecure teenager and this comment could have hurt a lot.
It might actually be quite appealing to them in the "fewer people own cars" scenario. The remaining people who want to own their own car will probably be willing to pay more, and businesses renting them out will almost certainly be willing to do so (since they'll be making money off them). Their margins would probably go up a lot.
That would be against all economic theories I know. I'd demand is lower, profits use to go down too.
I understand part of your reasoning, buying a car for the joy of driving, and probably some sport cars wouldn't be as affected, but for the automobile industry, it's going to be a hard blow.
It really doesn't go against economic theories. Demand has no direct relationship with profit. Supply and demand affect price, which has an effect on profit. When demand goes down production slows and supply eventually reaches a new equilibrium with demand at a new price. Cars that are already constrained in supply could continue to do well while commodity cars would go unsold. There would be a short term downward pressure on price, but it wouldn't last forever.
I'm not saying it wouldn't be a hard blow for the car industry at all. It would decimate it. But what would be left standing would be the most expensive cars and cars equipped to be rented, and the margins on those are much higher. Companies like BMW are more well equipped to take on that new market reality.
Perhaps the part I was unclear about is that the "them" in my OP was high end car manufacturers like BMW.
A self-driving car never gets tired or drunk. It never gets bored. It never talks on the phone, and it's never distracted by passengers or music. With the right sensors, it can see better than any human. It can react faster and more accurately than a Formula 1 driver.
Even with all these advantages, some automated vehicles will crash and kill their occupants. Some will even kill pedestrians, and sometimes this will be due to software error.
But humans do this already, and we do it so often that it doesn't even get on the news. Around 1.2 million people die in traffic accidents each year. That's just over 2% of all deaths[1]. Unless the automated vehicles of the future are orders of magnitude worse than current ones, switching to them will save millions of lives and prevent tens of millions of injuries.
It's sad that bureaucracy and human irrationality cause so much unnecessary death and suffering.
It's not entirely irrational - One driver suddenly goes crazy -> One car crash. Broken software update for thousands of self driving cars -> hundreds of crashes.
That can already happen. Modern cars have fewer mechanical linkages than you might think.
There are solutions to your proposed problem. You could require the car be taken into a shop for updates. This would make the update gradual (so the eggs are never all in one basket) and it would let mechanics ensure the update was applied correctly. You could also design the software to have a core that is never updated. It would monitor basic things like "Am I going too fast?" "Can I brake in time for any obstacles?" etc. Google uses this strategy IIRC.
>Modern cars have fewer mechanical linkages than you might think.
Even better, the more software and chips there are in your car the "better" the bugs get. Friends of mine had fun problems with a new Audi A5: about 10 times, it randomly turned off while driving taking about 20 seconds to come back, and handling a car with power-steering with the engine off is very hard! AFAIK, that problem hasn't been solved yet.
At least in Germany though, steering and braking have to be mechanically connected which maybe limits innovation in that area a bit (or reduces incidents where you cannot control your car anymore).
Yes, but auto-pilots are easier problems to solve, due to predictability.
It boils down to maintaining speed and bearing, while considering some other factors such as weather, terrain and collision prevention with other planes (transponder based?).
In case of a truly unpredictable event, the pilot takes over.
Cars on the other hand need to do much more than that.
The process of automatically determining where the road actually is, is not trivial by itself.
Cars also need to handle other unpredictable factors to a much higher degree; other manually driven cars, bicycles, pedestrians to name a few.
Not to mention road obstacles, and animals (in my country moose collisions are common, and deadly).
> Not to mention road obstacles, and animals (in my country moose collisions are common, and deadly).
It always fascinates me that this argument comes up from the negative side of self-driving cars pretty much every time, but it's actually the poster-child reason self-driving cars are better.
A self-driving car is at least as likely to recognize a potential collision with a moose as a human is (given current sensor tech) and probably much more so since it can't be distracted.
The advantage of a full omnidirectional sensor array is that it can pay attention to the kid on the bike, the moose crossing the street, the 5 cars around you, the alien spaceship in the sky, and the kid using your bumper to accelerate his skateboard all at once (or at least in timeslices small enough to count as such). You can only pay attention to a couple of those things.
No, don't get me wrong.
I am all for better sensors, drivers assist systems, or self-driving cars. I'm just pointing out that self-driving cars are probably more complex than air plane auto-pilots.
"A virus has been detected in your driving software. Please upgrade now to remove this threat" ... and the consequences are potentially a lot more damaging than PCs.
When the target gets big enough, it will happen. And how to protect against this, when we don't really understand the weaknesses yet - a rogue radar/lidar imitator sending false echoes, strong EM fields messing with electronic components - it could even be an intermittent connection fault in a communications bus. Plenty of unknown unknowns in this field.
Nitpicking: I should point out that far fewer than 1.2 million -- more like 35,000 these days [1] -- die in traffic accidents, although there are many injuries.
I understand where you are coming from but when you get on the road you put your fate in not only your own hands but the hands of the drivers around you as well. You have to remember that while you might be an amazing driver with a perfect record there are plenty of other drivers out there that aren't and when it comes down to it they are a bigger threat to you than you are to yourself (provided you are not under the influences or otherwise incapacitated).
Whatever accidents happen on a car driven by human, it is usually the fault of the driver. For self-driving cars, then it becomes that the manufacturers are responsible for all accidents. This is unmanageable liability.
You know what will be an unmanageable liability? When the insurance adjusters start taking into account the fact that you could be reducing your risk of taking their money by switching to a self-driving car, not to mention the fact that if you get into an accident with a self-driving car and you were in control the liability will be all on you (and them).
Once someone starts selling these things, insurance for city drivers who insist on driving themselves will skyrocket.
Where's your data? According to Google, their cars have gone 300,000 miles without an accident under autonomous control. At this rate, we wouldn't need auto insurance... so probably this liability cost will simply be part of the car's purchase price or a new type of insurance product offered to those who buy these cars.
On well-maintained roads that the software has been specifically trained on, with very expensive sensors and equipment, with a human driver that can take over instantly. Not that the work isn't impressive, but this 300k-no-accidents claim is just marketing spin and doesn't tell us anything about how close the system is to being ready for use.
It is more like the manufacturer is suddenly in the role of a public transport company. Self driving cars are more like small buses than personal cars.
> for all its expertise in developing search technology and software, Google has zero experience building cars. To understand how autonomous driving is more likely to emerge, it is more instructive to see what some of the world’s most advanced automakers are working on
What is this bullshit? He basically dismisses the entire Google effort right off the bat, and provides no information about it whatsoever? Just like that? As far as I know, Google is in fact much farther along than BMW, and I'd really like to know what they're capable of. It doesn't matter that Google isn't a car company. If they can manage reliable driverless cars, car companies will be lining up to license their technology.
I've also read that LIDAR doesn't work in rain, or even fog. I never see weather brought up in any detail in any writeups of autonomous vehicles. How close can we be if the sensors they're using only function on sunny California days? How will any car deal with the rest of the country where several months a year, a coating of snow can block any sensor from seeing lane markings, signs and signals, and even the outline of the roadway?
LIDAR can be augmented or replaced by other sensing technologies that penetrate precipitation better than human eyesight can; the algorithms will most likely still work. When the road is coated with snow, the cars could do what human drivers do: analyze the behavior of other cars, look for emergent traffic flows (in my experience a four-lane freeway becomes three well-spaced lanes in well-packed but not deep snow, and flows surprisingly well), and follow the pack.
Right, the beautiful thing about probabilistic machine learning/AI is that it can use the signals it gets, and even ignore some conflicting evidence, without having to be explicitly programmed with rules. I think a lot of people underestimate what's possible there, or take an overly simplistic view of what it'd be capable of, since we're so used to having to design behavior.
I'm almost certain Google's self-driving car is leaps and bounds ahead of BMWs.
Particularly, his BMW demo occurred on the highway, not city streets. Conditions on the highway are far more predictable, there are no turning choices to make (just follow the road!), and no pedestrians. BMW also doesn't seem to be interested in self-driving on city streets. FTA: 'He said the company aims to be “one of the first in the world” to introduce highway autonomy.'
I thought it was a pretty thoughtless statement too. Google doesn't need experience building cars. We have cars. The driver-less technology is what is missing and Google has the talent and money to do it.
It's not entirely unjustified. I don't think people appreciate the sheer difference between making one self-driving car, and making a million self-driving cars.
One of the Good things about autonomous cars is that we're close to having cars that can drive, when people want them to succeed. One of the bad things is that messing with them will be too over powering to resist and that will slow their deployment.
The typical "mess with this" kind of action would be walk up on the edge of a freeway and pop up a full sized STOP sign on the shoulder. Humans will say 'wtf?' and keep driving but autonomous cars, unable to know if it isn't a legit stop sign, will slam on the brakes to stop. Resulting in hugely funny (to some) traffic jams. Similarly with inflatable stoplights.
Then there are the things where even human drivers have issues (a person standing on the side of the road signalling traffic to slow down).
There are many situations that autonomous cars will need to be able to handle, that are handled in a fail safe way today by humans, yet to be programmed. Definitely will take longer than you think.
And something tells me that would be made very illegal assuming it isn't now (which it most likely is) very very quickly. It would be funny until it got someone killed. Then, not so much
Three defendants were sentenced to 15 years each in state prison Friday for uprooting a stop sign at an intersection where three teen-agers were killed in a crash a few hours later.
I would have thought in the future all kinds of traffic circumstances would be handled by broadcasting information (such as "slow down") in a format native to the onboard computer. Then the guy in the hi-vis suit is just a redundancy for humans.
But then our pranksters will merely hack together a parabolic dish and Raspberry Pi to emit faux stop-sign signals, then set up shop on the nearest mountaintop and snipe Benz's for fun. So now this is a signing/crypto and signal reliability problem, which is a bit harder. But still doable.
Easier said than done. What happens when the power goes out to the city's system? Can we trust the city to keep traffic info up-to-date? What happens when there's too much interference? When you're in a tunnel with the roads broadcasting above you?
Self-driving cars are the future, but these solutions that seem simple will still need a lot of work.
Memristor RFID chips. The cars have to be powered to move -- so you get the cars to ping road-embedded sensors, which then respond with the latest value they've been programmed with. They could get get programmed either over the 'net, or by a maintenance truck passing by and pinging them with a cryptographically signed message.
>> The typical "mess with this" kind of action would be walk up on the edge of a freeway and pop up a full sized STOP sign on the shoulder.
I am pretty sure, these autonomous cars will get info about road signs, etc through an authorized satellite system. So you can pop up with whatever sign you want, it is not going to matter.
This article is full of details and well worth a read. It's kind of a bummer for me, because I can't wait for the day when driverless cars replace the clueless drivers of Minnesota among whom I have to commute. But the technical challenges of bringing driverless cars into routine consumer use are still immense.
The article notes, about an expected transitional phase of development when driverless cars augment rather than replace human driving, "An important challenge with a system that drives all by itself, but only some of the time, is that it must be able to predict when it may be about to fail, to give the driver enough time to take over. This ability is limited by the range of a car’s sensors and by the inherent difficulty of predicting the outcome of a complex situation. 'Maybe the driver is completely distracted,' Werner Huber said. 'He takes five, six, seven seconds to come back to the driving task—that means the car has to know [in advance] when its limitation is reached. The challenge is very big.'"
My dream is the dream of fully door-to-door driverless cars. I think the article "Why Driverless Cars Are Inevitable--and a Good Thing"
by Dan Neil of the Wall Street Journal, published last year, is a good commentary on why ordinary people will mostly be glad to use driverless cars, and regulators and insurers will be glad to nudge drivers to use them. But that's only if they work, and it's not clear how soon driverless cars will work reliably and be manufactured inexpensively enough to become routine on our streets and roads.
a) As it says in OP's text, the most dangerous point in driving these automated cars is when a half-distracted driver takes over; having a fully automated car prevents this.
b) OP's text seems to see the biggest danger in automated cars not reacting properly to other humans driving erratically. This is easy to fix - once 100% of a road's cars are automated (and ideally communicate with each other), the problem of erratic driving simply disappears.
A question for thought: If I'm drunk in a driverless car, current laws would say that's illegal, since I'm still the main driver. But if it's fully automated, why should I stay sober?
I understood that the biggest problem is driving in non-highway, e.g., urban, situations. How will you deal with children playing at the side of the road? Bicycles? Mopeds?
"But for all its expertise in developing search technology and software, Google has zero experience building cars..."
The article is fundamentally missing the point of autonomous vehicles---namely, that this is a _software_ problem, not a mechanical problem, so car building is an irrelevant skill. Automakers have decades of experience with all sorts of useful things, like machine design, manufacturing, fit and finish, etc. But they suck at software, as evidenced by the hilariously low quality of all* dashboard media/navigation/tech clusters ever. Google, on the other hand, is pretty good at software, and is therefore much better positioned to win in this space. Empiricism verifies this, because the Google autonomous cars work better than the automakers'.
In fairness, the article's claim about the high cost of Google LIDAR is solid.
*Yeah, I know. Tesla's doesn't suck. But they're not a real automaker yet.
The LIDAR is expensive because it's not a commodity product. Once you take that LIDAR unit and spread the development costs over millions, embed all the digital electronics into one chip and start churning them out the cost will fall significantly. It might still be 5 figures, but it'll be at the low end.
Google doesn't even need experience in building cars.
They can license their technology to UPS and Fedex and make billions. They can license their technology to Uber, ZipCar and taxi companies and make billions. They can license the technology to traditional automakers and make billions.
The self-driving car ecosystem might end up very similar to Android. Google makes the software, and others make the hardware. Or they can just buy an automaker if they want to sell their own cars.
the terrible irony [is] that when the car is driving autonomously it is much safer, but because of the inability of humans to get back in the loop it may ultimately be less safe.
Driverless cars are only 90% there for complete autonomy, and the last 10% could take another 50 years.
The pilots' confusion in the Asiana 214 crash [1] [2] was likely due to uncertainty about how/if autopilot was managing the airplane, and what the human crew still had responsibility for.
"The difficulty of re-engaging distracted drivers" -
This is the danger that will be faced in driverless cars just like Asiana 214. If the expectation is that drivers must retain full "Situational Awareness", it seems like a lot of the benefits of being "driverless" are lost.
Perhaps it should be called "Assisted Driving" instead under given this expectation.
I can't believe they'd call such a vehicle a "driverless car", when the software simply gives up when there are too many things going on. Seriously, that software problem needs to be solved.
Wow, this comes off as the first of many hit pieces against Google more than an objective review of the current state of affairs. It wouldn't be surprising to see this as competition picks up.
The entire article is summed up fairly well in this one line "But for all its expertise in developing search technology and software, Google has zero experience building cars."
He effectively ignores the fact that Google is demonstrably farther along in development than any other company. Not to mention this isn't even a problem of building cars, it is a problem of software and AI, both of which Google is slightly more experienced than BMW, Audi, or any of the other companies he mentions.
All of his concerns about the timeline of this work seem to be based on some idea that in ten years, automation will only be ready for "limited highway driving" and that there are a lot of "Uncertain road" issues that aren't accounted for in these estimates. This differs greatly from Google's claims I read last year.
I couldn't bring myself to do more than skim the second half of the article but I saw no actual evidence to support his thesis of automation being farther away than expected (i.e., ten years). I only saw evidence to suggest I shouldn't bank on the auto-makers to get there first.
People keep throwing around the liability/cost issues, but the savings to our system in the healthcare and auto insurance sectors alone will more than offset any potential downside, to say nothing of the lives that will be saved. My prediction is that it will be so good, that it will eventually become a requirement in all new cars and it will fundamentally change the face of our society. Trauma bays, EMTs, auto insurance companies and the like will all have to cope with this change.
It seems to me that diverless cars will [at first] only work on prepared roads.
These are roads where only driverless cars will be allowed (perhaps just a lane on an existing road), and all the traffic, and lane, markings are designed for the computer to read. The roads are very carefully mapped out.
In some ways kind of like a train track.
The computer knows when it's almost time to exit the road and go to a regular road, and can alert the driver in advance. (Or simply pull off into a parking lot.)
I doubt that, with Accurate gps and a mindstorm kit, I could build a "car" that follows any road. I don't see why it would need to be 'prepared'.
With sonar, infrared, sign reading, gps, rain sensor, traffic management integration, line reading... exactly what does the cpu not "know" that a driver does?
The most interesting aspect that the article raises is the hand-over from automated driving mode to human controlled mode and potential difficulty for humans to shift their attention. What a fascinating problem that is going to play a part in every AI assisted job in the future.
Imagine reading HN from a mobile phone and then suddenly shifting to driving mode: it's obvious that you will be very disoriented for a few seconds. Thus, we need to constraint the consumption setting: projecting a screen to the front window is likely to shave a couple of seconds off from the disorientation phase. initially both mobile phones and sleeping are banned while driving self-driving cars. Constantly projecting extra peripheral sensory information to the screen is also needed. This would be immensely useful already today without self-driving cars.
However, it's obvious that the shift of attention will take a couple of seconds thus the car can't change the control mode in dangerous situations. It's likely that there will be self-driving zones like highways and the
control is changed mostly at borders of these
This is almost entirely an algorithm arms race, since that will enable better capabilities with cheaper sensors/noisier data sources. As such, I predict that Google will mop the floor with the traditional car makers' efforts. For some perspective, most car makers can barely design decent entertainment center software, what makes anyone think they can compete in the computer vision and AI category?
A fully autonomous car with the ability to make complex decisions on local roads is miles different from the adaptive cruise control and lane following/changing they're talking about here, and Google is the only one tackling the whole package.
And the gulf in safety between something that can drive all the time and something that needs the driver to jump back in when it hits a situation it doesn't know how to solve is huge to the point that I probably wouldn't touch the halfway solution with a pole.
What the author does not understand is that the vector for self driving cars will be Zipcar, Getaround, goget and other 'car sharing' companies. Cab companies might get in on it but they will struggle to deal with their large existing workforce.
There was another article on here the other day in which Elon Musk said that the last 10% was very difficult but it's easy to see how a car sharing company could offer a self driving service by avoiding that last 10%. They could do this because:
* The car might not have to go very far from it's pod to your location
* They don't necessarily have to go in driveways and other hard to reach places. "closer than public transport would get you" is good enough.
* The service can be restricted to urban areas with good map coverage
As other people have noted the opinions and even the technology of the major car companies is irrelevant.
My brother is a CHP officer and he told that he cannot wait for autonomous vehicles. He said at least 80% of all traffic accidents are directly due to human error (I would venture to guess that nearly 100% of accidents are due to some form of human error).
> But for all its expertise in developing search technology and software, Google has zero experience building cars. To understand how autonomous driving is more likely to emerge, it is more instructive to see what some of the world’s most advanced automakers are working on...
"But for all its expertise in developing automobiles, Ford has zero experience feeding horses. To understand where the world is going, it's more instructive to interview coachmen"
Wat? How is Google not relevant or "instructive"??
...and these are just the technical limitations of the autonomous car + driver system. There would appear to be immense potential liability in selling a product that is explicitly billed as being safer than human drivers, but which is a) different and b) going to fail, sometimes catastrophically.
I'd be surprised if autonomous cars become widespread before the last human train conductor is decruited. Driving a train is a simpler problem in a lot of ways, and yet we still feel the need for people in those jobs.
Tangential, I hope that this trend will bring some genericity and open access to car parts (as much as possible) to avoid the weird scam that is fixing anything on it right now. Even though I doubt it will be and might end up just like cellphones: features and look first, "fixability" on a few models.
Eventually, software will handle that situation better than any human.
In the video you show 2 out of 3 affected cars crashed. In an automated car world eventually 0 out of 3 cars would have crashed.
Compared to any driver, a driverless car will react faster, and knows more accurately its own speed, and the speed and position of the object moving across in front of it.
A driverless also knows better than most drivers what is possible, including braking distance, turn radius etc.
So when the software is written to handle that situation it will always do better than I would, and sometimes do better than a fully alert Sebastian Vettel.
Quite possibly better than a human, since adrenaline is only so fast at moving humans from a relaxed state to hyperalertness, while computers can always perform at full capacity, and computers are probably better at predicting how to accelerate an object moving at high speed in order to avoid collision with multiple other objects moving at high speed.
I've always wondered if there isn't an intermediate step, a sort of 80/20 rule to gain most of the benefits for 20% of the work, without your car actually taking you for a drive while you sleep in the back seat.
For example, lets assume highway traffic is caused primarily from exceeding road capacity, and road capacity is a function primarily of the size of the gap that we are taught to leave between cars. Why are we taught to leave a 2 second gap between cars? In theory, that's how much time/space you need if the car in front of you decides to go full-tilt on the breaks with no warning.
I've read about autonomous caravans (the article calls them platoons) where cars line up behind a follow vehicle. The follow-cars in that scenario typically also take over steering. Like in the 2014 Mercedes S-Class from the article;
A jovial safety engineer drove me around a test track, showing how the car can lock
onto a vehicle in front and follow it along the road at a safe distance. To follow
at a constant distance, the car’s computers take over not only braking and
accelerating, as with conventional adaptive cruise control, but steering too.
But why does it have to steer? If the car takes over JUST breaking and acceleration the software and sensors required are vastly simplified. Given a target maximum speed, set by the user (so it can be set at 85mph and not 65mph), the car drives at the designated speed, or else maintains a close follow (250ms is ~30ft at 85mph). Why can't the driver still be responsible for steering while this is happening?
I think the key is giving the driver very high confidence that "no, there is no way that my car will let me rear-end the guy in front of me" even if you're just 30ft back at 85mph. That's not "scary close" but at that distance, you are trusting the car in front not to apply maximum breaks given human reaction times. Computers could apply sufficient stopping force in time, although the responsiveness required of the algorithm might make regular driving a bit "twitchy" depending on the human driver you're following.
I've never even driven a vehicle with adaptive cruise control, so I have no idea how "aggressive" the system is, or how it feels as the driver. I don't think any of the adaptive cruise systems out there will take you down to 0mph and then also start moving again, which seems like a must-have. But I bet if Tesla added "maintain Xmph or close follow" to their Model S, owners would trust it, use it, and look quite badass in the process.
You might benefit from some obvious (but not distracting) signal to other cars when this mode is active, and spend a boat-load of money on awareness, to try to avoid the inevitable "oh this asshole is tail-gating me, I better slow down."
Another caveat is that it's easier to steer smoothly at high speeds when you look far down the road ahead of you, which is kind of hard to do when you're breathing down the neck of the car in front of you. Obviously if drivers start losing the ability to stay in lane when following that closely, the idea falls apart.
Ultimately I think "cruise control" is something every driver understands and trusts. Make cruise control better. Call it "super cruise" and put it in the Tesla Model S. Try to educate other drivers about "super cruising" so that they know you're not actually driving like an ass. More brands will follow.
If most highway traffic is caused by exceeding road capacity by just 1 or 2% (personally hard to believe, but that's what experts say) then in theory if "super cruise" reduced inter-car gap by 50% for 5% of cars on the highway, then POOF no more traffic jams. Of course as adoption increased much past 5%, only then you would need to add software to support zipper merging ;-)
Audi has a gearbox that changes it's shift pattern based on upcoming corners identified by the GPS.
Mercedes has advanced pedestrian/object detection integrated into their higher-end cars to prevent collisions.
and there are probably many other advances already present in cars that basically allow them to drive themselves without much user-concentration these days.
Latest BMW 7-series: "A new Enhanced Active Cruise Control system builds on the last-generation Active Cruise Control System with Stop & Go to add the ability to brake to a complete stop if the driver doesn't react to stopping traffic in time. The system alerts the driver to the situation along the way, but will ultimately take over if it deems it necessary."
That sounds quite lame actually. Not the kind of system that instills trust and confidence in the driver. I don't want to hear the damn thing beeping at me every time it has to do its job.
Thanks for the link, interesting to see it in action.
But this is actually the complete opposite of what I want. This system reduces road capacity by almost half. Look at the gap it's leaving. Look at how slow it is to start up from a stop. I think I'd be on the horn if this guy was in front of me.
When you leave that much space in front of you ("to be nice" or to "smooth it out") the empty road where 2 or 3 cars could have been is actually reducing road capacity causing significantly worse traffic behind you.
Most of this video is shot below 25mph. Average follow distance of a competent computer "super cruiser" should be < 10ft, which would make the car essentially "free" from a capacity standpoint. When the speed finally gets up to 50mph, the follow distance is practically too long to count, something like 5 seconds back.
The driver even comments just after t=5:00, "See how big the gap is? I have it set at 75, it's doing about 50." That car has a 300hp engine, why is the computer driving like it's drunk?
"I'll just sit back and relax and get to work when I get to work." No! You probably impose a several thousand dollar "productivity tax" on SoCal commuters by driving like that over the course of a year.
After driving 2000km over the weekend and then getting stuck in traffic each day since, I've been wondering the same thing. We're a way off trusting our cars enough to sleep while it's moving at high speed, but a car that could handle the crawl into the CBD would free up drivers to check email, read a book or eat breakfast. A lot of that time is waiting and watching for a light change, or moving slowly down a busy road.
Something a bit like cruise control (in the way that it can be quickly overridden) would be a start. Put it in robo-mode and get a warning if you need to take the wheel within the next few seconds.
I think it is better the driver is either fully engaged or taken out of the loop entirely in case they get too distracted or fall asleep.
I don't even like cruise control (although I can see the use when covering massive distances with little traffic) and drive a manual (stick shift car).
“The first generations [of autonomous cars] are going to require a driver to intervene at certain points,” Clifford Nass, codirector of Stanford University’s Center for Automotive Research, told me. “It turns out that may be the most dangerous moment for autonomous vehicles. We may have this terrible irony that when the car is driving autonomously it is much safer, but because of the inability of humans to get back in the loop it may ultimately be less safe.”
This reminds of the frame problem in earlier AI[1]. Artificial systems can be built to deal well with a given frame having given specification but they can to the boundary of these frames, they fail to gracefully change their approach.
Edit: changed link since the Minsky article is more descriptive. Wikipedia only describes a frame as a data structure but I (and I think Minsky) would see them as a metaphor for the structure and limitation of AI system.
Yes, it is ironic but whilst human operated vehicles are still using the roads it actually requires much more sophisticated autonomous vehicles to 'join in'. Automated vehicles will greatly benefit us be being able communicate their intentions to each other allowing roads to run more freely as continuous adjustments to speed and direction make stopping for traffic virtually obsolete. Now throw an essentially unpredictable human driven vehicle into the mix and you get problems and the onus will be on automated vehicles to deal with this.
But I'd argue that concern is backwards. First, go for correctness. Perfect the algorithm and sensors, get the car working. Then look at scaling down the technology and making it cheaper. The fact that Google doesn't make cars (a point the author made) is kind of irrelevant, because this is a sensor/control/software problem, not a car problem.
Recognize that BMW/Daimler/et al really don't want true self driving cars. It's not really in their interest, because it would radically reduce the need for car ownership and would open up a new world of on-the-fly car rental. Human drivers required = good business. I'm not saying its a conspiracy, just that they have no passion to disrupt their industry in this way.
Even the cost is somewhat moot if the business model is different. Let's say the Google Car sensor package costs $100,000. That's cost prohibitive for individual ownership, but it would not be a problem for a business model like ZipCar + Uber, where you call a car on your mobile phone to get you, and 'rent it' for a short self-driving or assisted trip. If car use moves from ownership to renting, then many people can spread the cost of the sensors.