Volvo's is starting a 100 vehicle test in one city in Sweden. Automation is only available on some major roads. The driver is not required to pay attention in automatic mode. There's a progress bar - minutes until the end of automatic driving. There are previously located points at the end of automatic driving regions where the vehicle can pull over and stop, should the driver not take over when requested. So that's what entry level full autonomy looks like. That's a good level 3. Those go on the road any day now; the vehicles have been built and tested, and the first of the 100 drivers, who are car purchasers, not Volvo employees, have already been chosen.[1]

And yes, they work in snow. The test is in Gothenburg, Sweden; they have to work in snow.

[1] http://www.volvocars.com/intl/about/our-innovation-brands/in...

That said, I agree that the liability transfer is a big issue. Although, at some point, expect to see some case that breaks through "It's assistive driving only. Nudge-nudge. Wink-wink."

Even if you are right because of the above all manufactures would be stupid to not have a significant investment: the day when they have no choice might be out of their control and if they are not ready they go out of business (or confine their business to poor countries with loser laws - but these countries tend not to sell many high profit new cars).

https://www.youtube.com/shared?ci=ALqiEkdD36Q

I'll bet that in the near future (next decade or so), there will be rising incidents of traffic being blocked due to a stopped automated car who's driver is not responsive, but the computer couldn't handle some construction or weather, and went into fail-safe mode, simply slowing down and stopping (but not pulling over due to lack of space).

Heck, we're halfway there with people checking FB on their phones at stoplights, so anticipating this seems reasonable. We haven't solved that problem, and I don't think we'll solve the one you raise (not for a much longer time, at least). Valid problem, but a solution is likely far off.

Edit: I would gladly accept more failure in uncommon edge cases if it means reduced failure in normal operation (ie: drunk guy slams in to a house at 1 AM on a clear calm night)

This is coming; I've seen mentions of a system with passive alcohol detection (just from the air inside the car) plus active "blow in tube" test if the passive test triggers that will start rolling out to cars soon.

But it seems if stopping drunk drivers was really a major concern, we would have done this years ago.

Remember, the breathalyzer is only a backup in the situation where the car has detected alcohol vapors in the air, and is wanting to confirm that the driver is not the one who has been drinking. So it's not something you do every day, ergo it can be a bit of a hassle.

Otherwise, what happens if someone (sober) is giving a ride to a passed out friend, can't wake that friend up, and ends up leaving the car (with the drunk friend still in it)? In that case, everyone involved was being responsible (no drunk driving was involved at all, actual or potential).

http://www.autoblog.com/2017/01/10/nissan-self-driving-cars-...

> For example, Sierhuis described a situation where a temporary construction site, something like a crew repairing a traffic light, would confuse the car. Not only would the traffic signal be missing for the AI to process, but we would also have a worker waving cars through the intersection. It's a prime, real-world example of a challenge that is basically impossible to program into software. With SAM, when a self-driving Nissan approaches a situation like this, it stops and sends a signal to a human operator. This operator, called a Mobility Manager, checks the camera feeds from the car, high-definition maps, and whatever else he can, and then plots a new route for the car to take. This information is then sent to all the other Nissan AI machines in the area so that they don't have a problem when they approach the same intersection.

The California DMV continues to publish autonomous vehicle crash reports. The last one was in October.[2] A Google vehicle was rear-ended at 3mph when it was entering an intersection with poor visibility, detected cross-traffic, and stopped.

I'm looking forward to the 2016 self-driving disengagement reports appearing on on the California DMV site.[3] Those were due January 1, covering the year ending 1 December 2016. Then we'll have comparative info on how all the players are doing.

[1] https://www.youtube.com/watch?v=Uj-rK8V-rik [2] https://www.dmv.ca.gov/portal/dmv/detail/vr/autonomous/auton... [3] https://www.dmv.ca.gov/portal/dmv/detail/vr/autonomous/disen...

Volvo loaned a family a Level 3 car yesterday.[1]

[1] https://www.engadget.com/2017/01/09/volvo-tests-self-driving...

It's not clear to me how much L4 buys you as a consumer product. "No driver during defined use case." If you can't go door to door in a wide range of weather conditions, what's the difference between having a non-attentive driver and no driver?

ADDED: Yes, as was written elsewhere, low-speed campus use cases make sense. But that's not what most people consider self-driving.

Level 5. Many decades.

Outsource Uber driving to lowest-wage drivers in India, Bangladesh, or Nigeria.

Obviously "do not stop in the middle on the road" will be high on the priorities of self driving algorithms...

The edge of the road is soft, not well marked, contains rocks and nails and housepets. The road proper is hard, well-marked, and vehicles should want to stay there always.

Of course this will occasionally result in inconvenience for other drivers as vehicle 1 has been stationary for 25 minutes now as it applies a mandatory software service pack. But, oh well. That is THEIR problem. It's obviously maximal for MY vehicle to stay on the road at all times.

Self driving cars already know how to park and get out of the road.

The rest of the argument goes into bizarro land. Can't tell if trolling...

First, a stop vehicle in the middle on the road is a surefire way to a multi-car collision at worse, and traffic fines at best. Both of which would be avoided by any self driving algorithm not written (or approved if it's something like a neural network) by Harpo Marx.

Second, "It's obviously maximal for MY vehicle to stay on the road at all times" is not an argument, since self driving algorithms are not written and optimized for the car owners sole convenience but also for overall traffic, traffic regulations, and several other considerations besides...

i fear much of the enthusiasm behind self-driving cars is from people who understand neither driving, nor computer software, very well.

There are places where it takes some skill just to comply with an officer's request for you to pull over. Not everywhere is a bunch of standardized interstates.

Have the car GO and find a suitable area. Driving for a few more km perhaps slower and with the alarm lights flashing to find a spot to stop, will surely be safer than stopping right then and there just because you can't find a convenient place to pull out.

Lower priority than stopping then and there in the middle of the road/interstate (as suggested by parent) as opposed to a service/rest-area/shoulder/side of the road/etc...

If you're going to look at your phone at a red, know the light timings or look at the transverse pedestrian crossing signal or transverse yellow light so you know how long you have.

I think it puts a positive spin on itself. Traffic lights work best when the green duration is utilized to the fullest extent.

If the car came to a situation that it needed human input then that human could be drunkenly slurring advice.

But I am failing to see a situation that the car wouldn't be able to figure out, but the human inside could figure out... I've been in situations where I didn't know what to do and had to wait for the emergency services to come and help.

When self driving cars are that smart the emergency services AI will just fire 30 or 40 drones towards the emergency and those will coordinate with the cars below to fix the situation.

Great point. It's one that not enough people are emphasizing.

I need to be physically engaged in the driving experience or I won't be attentive in the least. It's been a trait since childhood that my mental state is tied to physical activity. A car that keeps me physically uninvolved until it decides that it needs me would be a death trap for me and anyone around me.

I do think there's the potential to enable full autonomy in well-defined use cases while requiring manual control elsewhere. (Still problematic--how do you enforce?)

But the idea that a driver can take control on short notice from an automated system is a non-starter. Maybe you can assume there's an adult with supervisory control in the vehicle but you certainly can't assume they're prepared to deal with urgent emergencies.

That's probably fine though, if the automation can reliably discover when the environment changes. Especially since perhaps a majority of the time spent driving is in the very few "easy" scenarios.

Human drivers are probably happy to drive 10% of the commute time for the next 2-4 decades.

What this means however is that the dream of completely autonomous Ubers and transport trucks is probably much further out than some might hope.

That sounds about right. The ability to read/write/etc. while driving on the highway would actually be a huge win for a lot of people. And, equally important, it seems achievable in, perhaps, a single digit number of years.

But there would need to be an unambiguous mode change from "you can do Facebook now" to "you're driving" with plenty of time to make that shift.

This isn't as sexy as roboUber, shared cars, etc. but I'm inclined to agree that getting the last 10% of the way to unsupervised self-driving in essentially all situations is going to take a very long time.

If you do not accept and manually turn off auto-pilot, it slows down safely to a stop in the break-down lane.

http://www.volvocars.com/intl/about/our-innovation-brands/in...

Uber could probably also just have defensive systems that simply park the car and wait for a human driver to arrive, if there is a problem. It's good business for them to have 3 cars per driver. It still doesn't approach the interesting problem though, which is autonomous driving on a human level (Autonomous Uber without backup drivers, in an entire city). That I think is far away.

But would a car's connectivity and latency allow for that to be safe?

Edit: If not there would still be some form control needed for the passenger.

Serious question.

a. 1 fatality per 100 million miles driven as an average driver[0]

b. .5 fatalities per 100 million miles driven as robo driver

c. .25 fatalities per 100 million miles driven as 20-70 sober well rested driver

I'm 31yr old male so I have an average life expectancy of 47 more years.

That means I lose about .5/100 million * 47 yrs * 365days * 24hrs * 60mins * 60s = 7.4 seconds life lost per mile in robo-car. Sober/well-rested that drops to 3.7 seconds of life lost per mile.

So I'd be giving up 3.7 seconds of life to be able to read/work/sleep instead of drive for every mile. For me this is a good deal @ high speeds(trade 3.7 seconds of life for a minute of time), and an absolute steal at lower speeds(usual commute averages 4 minutes/mile.)

[0]https://www-fars.nhtsa.dot.gov/Main/index.aspx

Your expected gain would be nearly 5 additional years of life! Should I sign you up?

There are certain strategic times in life where you'd want to take this, though. Ideally after the kids are grown up and don't need you anymore (so you either drop dead painlessly & predictably, or get a nice long life in great health so you can see the grandchildren grow up), or if that's cheating, in childhood or right after college so you haven't invested much into your life yet. Same reason they sell life insurance.

Not taking the pill already of course risks dying in your sleep.

I've invented a teleporter that will reduce your commute to 1 second every day. The only glitch is that your molecular makeup may be incompatible with my device, and there's no way to tell for sure. If it is, you'll disintegrate instantly the first time you try to use it.

Don't worry, only 1% of the population is incompatible. From a probability standpoint, you should expect to gain more than 1% of your life in extra time. Care to give her a whirl?

Because your proposal is very different for two reasons. One risk vs reward. In the robo-car example the pay off is 20-80x of lost life to gained time. You get a lot of time for very little risk. But your example is 1.03x.

The second is the distribution of risk over the years. With the driving I take a little risk to gain a little time each day for 30 years. In your example I take all of the risk up front and then have small a pay out over the next 30 odd years.

But I don't think that most people would get into a car that was twice as likely to crash as an average human driver (including drunks and so forth). Which suggests that most people have a different approach to this question.

The gains from coordinated robotic driving could very well be enough to actually reduce speed limits by a significant value, which not only reduces overall fatalities (due to laws of physics), but also makes the equation better for the robotic cars, which might convince more people to switch, leading to a positive feedback loop.

That's an optimistic scenario, though.

[0]http://www.npr.org/sections/health-shots/2016/12/06/50444863...

I grew up driving with my grandparents who were over 70, and my friends/random people with a car who were under 20. I think that the decision would realistically be made mostly on how the question is framed.

"Will you get in the car that's twice as likely to crash?"

"Will you get in the car with grandma to go to the zoo?"

Talking about weird edge cases misses the point. Edge case = slam on the brakes.

Because, duh, something went wrong.

Human errors tend to be misjudging something and then getting into a situation that, due to physics, they can't quite get out of. So you can't quite stop the car in time and you hit the car in front of you at 15mph.

Car errors are going to be things like "I thought the tractor trailer was a billboard or a bird so I hit it at 60mph with no braking". Or "I lost all sensor data and got stuck in a loop so I drove off the bridge". It's a fundamentally different sort of error.

I watched someone driving on a nice sunny day on a low traffic interstate suddenly make a sharp right turn crossing our lane to fly off the highway and start rolling. I have no idea why this happened, but everyone in the car survived so it's not even part of the traffic fatality statistics.

For example, what if it's dense fast traffic (say, one car space between each car, 50mph) in random black ice conditions and sensors suddenly begin failing so the software no longer trusts is inputs? Based on recent good inputs, accelerating, swerving, or breaking are all dangerous, and based on the loss of "vision" continuing at current speed is dangerous. The safest thing to do might be to begin honking, flashing hazards, and slowing down very slowly, but could still be way riskier than just handing control to a human driver who (in this scenario) can still see.

I can think of tons of similar situations. I have a feeling a lot of folks who think it'll be so easy are people who have mainly experience driving in easy areas and conditions.

EDIT: I could have easily made the scenario way worse: the road is curvy, or random lanes are closing and reopening due to construction. Loss of visibility then means continuing straight has a high chance of causing a serious wreck quickly rather than maybe eventually

Still, no driverless car should ever let itself be in dense fast traffic to begin with. Only a human driver would ever be dumb enough to do something so stupid and dangerous.

I would expect all high speed roads to be continuously validated by the car network, mapped in detail and kept up to date by the sensors of each car as it drives it. Your car should be streaming a model of the upcoming road from the cloud and be able to fall back onto it in the event of sensor failure.

Handing over control to a human driver only works when one is actually in the car. And paying attention. And competent.

I would assume that a self-driving car would allow more than 1 space between cars when going 50 mph, and probably wouldn't be going 50 mph when there's black ice around.

Hell, most humans wouldn't be.

Then the human is no better or worse, no?

One of the improvements is that an autonomous car will avoid getting into a situation it can't dig itself out of. And, at times, it will detect those situations long before a human will.

My favorite counterexample for people like you is driving in Southern California in the rain. An autonomous car NOW is probably better than 25-30% of the drivers on the road in that situation.

It should be continuously preparing and revising a stopping strategy. It should even be able to do this with any half its sensors returning spurious or misleading data. Or even if all sensors are suddenly jammed, it should be able to blindly execute the stopping strategy prepared moments prior.

Well said.

GM didn't get to play the engineering is complicated and there are always tradeoffs card with its ignition switches. (And, yes, I know this was a case of a known problem but product defects still usually trigger liability in any case.)

Although, another way to look at is: If you fall below the bar for the average human, including drunks/etc, aren't you by definition more safe using this AI system?

If your question/argument is: What about those who are safer than average - then i think the answer is, they shouldn't use self driving, only the "poor" drivers should. Of course the next question is how do you determine that.

Perhaps infractions on your driving record would inhibit you from driving. If you get caught for drunk driving, that's it, no more license, purely automated driving (even if that means no rain driving, etc). If you're above 70, or etc.

We, as a society, are about to learn some hard lessons about software. As in, deuteronomy-level lessons - things that will be (figuratively) carved into stone.

We will gain a new appreciation for the value of human decision making.

A pretty big chunk of automobile collisions are directly attributable to poor human decision making.

For example, nearly 1/3 of fatal collisions in the US involve a driver with a BAC of 0.08 or higher.

PS: Drunk driving is stupid and dangerous, but poor statistics bother me. Drunk drivers are more often night time drivers adjusting for that closer to ~1/4.

(According to the National Highway Traffic Safety Administration, 35,092 people died in traffic crashes in 2015 in the United States (latest figures available), including an estimated 10,265 people who were killed in drunk driving crashes involving a driver with an illegal BAC (.08 or greater). Among the people killed in these drunk driving crashes, 67% (6,865) were in crashes in which at least one driver in the crash had a BAC of .15 or higher.

From https://responsibility.org/get-the-facts/research/statistics...

So unless they are gravely misrepresenting the numbers, it's 29% of fatalities involving a driver with an illegal BAC.

(again not trying to argue, I'm sure there are fatalities involving alcohol that are not attributable to the impairment, just adding clarification about where my statement came from).

Anyway, to be pedantic 29.0% is closer to 25.0% -4% than 33.3% + 4.3%. But, it's also 'close to' 1/4, so I don't have an issue with your phrasing.

However, as you note this is 'involving' a drunk driver. So, that 29% is inflated by cases where the sober driver struck a drunk one. </ done beating a decaying equine.>

For example - I was nearly rear ended by another driver, but I noticed they were failing to brake and pulled into an oncoming traffic lane (after checking there was no oncoming traffic of course). I would not have been at fault but also managed to avoid the accident altogether.

Also just because you aren't "legally" at fault doesn't necessarily mean the accident wasn't a result of poor judgement on your behalf.

For example, slamming on one's brakes noticing you're about to miss a turn you need to take and you get rear ended. You won't be legally found "at fault" but the accident is largely "your fault" for unpredictable driving, slamming on your brakes, and having been so inattentive to have nearly missed your turn.

So I think we're at least comparing apples to apples.

Do you have a good source that tracks fault in alcohol involved collisions? (I'm not trying to argue, I'm curious to look at it)

121 million self-reported episodes of alcohol-impaired driving among U.S. adults each year

Just remember, even if you are sober, driving home from a party at 3AM is still significantly more dangerous than average.

I sloppily implied causation above, but I didn't actually say 1/3 of crashes are attributable to alcohol, I said that a "big chunk" are attributable to poor decisions and separately pointed out the large percentage of fatal crashes involving alcohol.

Anyway, 1/4 is from someone with access to raw data who spent some time trying to get less biased numbers. So, wider error bars than NTSB estimates, but less biased assumptions.

Drunk driving is still a major issue, but there is a reason people are still spouting 20 year old data. When trying to make a point there is a tendency to cherry pick to the point of distortion.

Would you elaborate on this? I read this as meaning that the statistic includes drunk drivers who were involved in a fatal accident but weren't determined to be at fault for causing the accident. Am I understanding your intent correctly?

Assuming this to be the case, I agree that there should be some additional nuance here. That said, it would also need to take into account whether or not their inebriation prevented them from acting defensively to avoid the accident. I'm not sure how this would be done.

https://en.wikipedia.org/wiki/Drunk_driving_in_the_United_St...

Also of note: "On average, about 60% of the BAC values are missing or unknown. To analyze what they believe is the complete data, statisticians simulate BAC information.[7] Drivers with a BAC of 0.10% are 6 to 12 times more likely to get into a fatal crash or injury than drivers with no alcohol."

This is a flawed comparison because failing to update your software or keep your tires inflated or clean off your sensors would also be human decisions that could lead to adverse outcomes with self-driving cars.

Non-driving-operation decisions like that are applicable to both regular and self-driving cars.

It's incredibly simple to remove humans from those decisions. Just have the self-driving code refuse to engage until the issue is resolved.

Calling an ambulance is currently the recommended thing to do in severe medical emergencies anyway.

For other concerns about availability, adjust maintenance inspections and routines as necessary. For example, if you live somewhere there is no ambulance service and someone is of uncertain health, maybe inspect the tires every day (this sounds tedious but look at the alternative you are proposing, having a tire blow out while delivering dying Grandma to the ER).

I also expect that automatic systems will greatly improve the level of maintenance of vehicles (for example, mandatory tire pressure monitoring already does...).

Bad example--the majority of the lessons in Deuteronomy are shit, especially in a modern society, and were almost never taken literally through most of their history.

As an added bonus I can turn it on in a traffic jam and relax as that's stressful but not fast enough to be dangerous.

Btw, what is the death/accident rates for non-drunk people? Including accudents from drunk drivers crashing into them.

In fact some vehicles already have front crash prevention systems and stability control systems that automatically brake and cut the throttle (2 control axes).

http://www.iihs.org/iihs/topics/t/automation-and-crash-avoid...

So manufacturers could build upon that baseline by adding more sensor inputs and control axes for collision avoidance. Instead of just braking when a large obstacle appears in front the vehicle could steer around obstacles, exert force on the steering wheel to prevent running off the road, limit throttle inputs when approaching curves, etc. Some aircraft autopilots already have similar functionality by automatically climbing to avoid terrain.

Obviously, a lot of people still die in traffic accidents. But the bar is actually still fairly high for automated systems. I also suspect (though I don't have the numbers) that even if automation were just for highway driving that would, by itself, significantly reduce the fatalities.

But...

It isn't just deaths that need to be prevented. It is a combination of total damages, delays, deaths etc. If self driving cars take forever to make a turn or have other AI failures like driving too slow, it won't be acceptable to a large number of people.

In addition, you will see a situation in the future where partially automated cars Level 2 or 3 are safer than current driving modes. That means that in order to go from level 3 to level 4, you need the car to be safer than level 3.

I want a fully automated self driving car so badly, but I suspect I have a long way before I can hop in a car with no steering wheel and sip an IPA while being rushed home from a tough day at the office.

In fairness, your operation of the vehicle has a substantial impact on the outcome of operating the vehicle, the the decision loop that your brain implements while driving is largely incapable of effectively processing and executing on the volume of information and variables that are in play in the circumstances that lead to high speed collisions.

Anecdotally you might feel in control, but it is probably worth noting that most people, drunk or sober, tired on fully awake, fully alert or texting, felt like they were in control when they were suddenly the cause of, or involved in a collision.

So now you are in an accident caused by someone else. If that accident would have been prevented had they turned on the automated driving system that is in their vehicle, then do you still prefer this? Is having someone to blame really worth being in an accident that would not have happened if the computer was in control?

So, is it truly reasonable for you to refuse to engage the self-driving system in your car, putting other people on the road at greater risk?

I wonder if there will be increased liability, both financial (insurance) and legal, when someone willingly chooses to take control in a situation that would be well-handled by their self driving vehicle.

So? That's something humans tend to do, and anyone trying to sell self-driving cars will have to account for it.

You also can't compare deaths per million miles if the self driving cars don't operate in dangerous conditions. You'd have to compare apple miles to apple miles and orange miles to orange miles. A lot of people die in car wrecks in bad conditions.

A car that only kills .5 people per million miles, but doesn't work in the rain, snow, dark, whatever, likely isn't safer than a person.

Also: hacked in...3...2...1...

It could even be that the safety improvements are so dramatic that they become the pressure for us to engineer around the other issues. (Like marking lanes electronically and giving traffic cops new equipment and whatever else.)