My questions about driverless cars
We have the past few years seen a steady stream of articles about the inevitable arrival of driverless cars. I am skeptical, but willing to be persuaded. I have questions about how this would work. The articles I have seen don’t address, or even imagine, them. Perhaps this is a failure of journalists to ask the right questions. Most of the articles seem to fall into two categories: paraphrases of industry press releases, or Gosh! Wow! pieces by writers who have been thoroughly seduced. So take this post as a critique of journalism as much as it is about driverless cars.
There are two scenarios for driverless cars: completely autonomous with no input from the rider apart from designating the destination, and semi-autonomous cars that hand off control to a human as necessary. The fully autonomous version is far sexier, of course, and most of my questions are about that version.
But first a question about the semi-autonomous version: How does the hand-off work? The thing is, a driverless car is very attractive to those of us with a daily commute, but only if this means we can use the time doing other stuff, e.g. sleep, read, copulate, and so forth. The problem I see with the hand-off of control to the human is that a driver has to be oriented to the surroundings: road conditions, traffic, where exactly they are, and so forth. This orienting is continuous and automatic when driving. It is anything but, when sleeping, reading, or copulating. This need not be a problem with a planned hand-off. If, for example, the car is autonomous on a freeway but human-controlled on surface streets, then an alarm could sound a few minutes before the car reaches the off ramp. But I see ample potential for wackiness with unplanned, much less emergency, transitions. How much situational awareness do I have to maintain during the trip? If I have to pay constant attention to the trip, prepared to take over on short notice, then frankly that sounds agonizing. I would rather just drive.
Next I move on to fully autonomous cars. By way of background, I have a commute of about 35 miles through secondary roads. These vary from residential streets to country roads through woods and corn fields. They are perfectly good roads, mostly with one lane in either direction, most of the way with a solid double yellow line dividing the two directions.
Articles about driverless cars routinely include the pious statement that the car will of course follow traffic laws. People who question this mostly wonder if this really means the car will putter along when everyone else, including the cops, are driving ten to fifteen over the posted limit. My questions are different. I wonder about that double yellow line. Its whole point is that you aren’t allowed to cross over it, expect for left turns. So here are bog-standard routine situations in my daily commute:
(1) It is trash day. I end up behind a trash truck stopping every few feet. Present-day, this is a minor annoyance. In practice I ease over so I can see around the truck, and zoom across that double yellow line as the truck is stopped, and in a spot where I can see that there is no immediately oncoming traffic. The trash guys around here are pretty cool. Sometimes they’ll even signal me whether or not it is safe to pass. Everyone does this, including the cops and including the guys immediately ahead of a cop car. But this maneuver is technically illegal. So in our driverless future, are we going to stop and start for miles on end behind that trash truck?
(2) It is trash day. The trash truck is come and gone, and an empty trash can has rolled/been tossed/been blown all or part way into the roadway. In the real world, this hardly rises to the level of petty annoyance. If there is no oncoming traffic, I swerve around it without slowing. If there is oncoming traffic, I slow to await a gap, and in the worst case I come to a complete stop. At that point I might nudge the can off the road, or in a pinch get out of the car and move it, but in practice this is virtually never necessary. What does a driverless car do?
(3) There is an accident up ahead. Emergency personnel are on the scene, and the police are directing traffic away. There is a police car sideways in the middle of the road, and the policeman is waving cars to turn around. So far so good. I am confident that our driverless car and recognize the obstacle and avoid running the officer down. But how does it recognize and properly interpret the cop’s instructions, rather than (e.g.) treating the police car as a minor obstacle to go around, kind of like a trash can? There also is that pesky double line again. The cop is instructing traffic to make a technically illegal U-Turn. So what does the car do?
Moving on, in some of the articles I see a passing mention that the system doesn’t work in inclement weather. Please. This not only isn’t ready for prime time, it isn’t ready for the late-night infomercial block. Get back to me when you can navigate a road with three feet on snow on either shoulder and the lane markings obscured by salt and/or slush.
Come to think of it, we see mentions of how the Google car relies on 3D mapping down to within centimeters. Please. In winter in snow country the effective driving environment changes from hour to hour. Mountain View may be the center of the universe, but it isn’t the entirety.
Finally, I want to ask about failure modes. Whenever I mention this, people think I mean crashes and they respond with a discussion about how many happen with human drivers. That’s not what I mean. What I am asking about is more mundane stuff like flat tires. Currently, if your tire goes flat while driving you do a rapid situational analysis to figure out how to respond. If you are next to a nice wide shoulder you respond one way. If you are on a bridge with no shoulder you respond differently. You might respond yet differently if that bridge is one of those long ones with “Bay” in its name. What does an autonomous car do?
More broadly, when you drive your car every day you know how it should feel. If the brakes don’t feel quite right, it is time to have them looked at. Or maybe there is a vibration it didn’t have before. And so on. How does this work with our automated car? If you own it and ride it daily you might pick up on this stuff, but then again you might not, what with your sleeping and reading and copulating the trip away. How do you feel the brakes are a little off, when you aren’t applying them yourself? I have this sneaking suspicion that there is a tacit assumption that all these automated cars will be no more than three years old, and fanatically maintained. I see vast potential for wackiness when this assumption hits the real world.
I’m sure I could come up with more questions, and I would be fascinated to see yours. It may well be that they can all be answered satisfactorily. But color me skeptical until they are.
Let’s say your driverless car has an override function so that you can bypass the trash truck. You press the override function, the car gives you a warning, you confirm, and then drive into the opposing lane of traffic. Boom! Someone else’s driverless car backs out of the driveway, also on override because of the trash truck. Property damage and mild personal injury occur. Who is at fault?
A) you
B) the backing-out driver
C) the corporate operator of the driverless trash truck
D) the manufacturer of your driverless car (and/or/including the author of its software and supporting hardware)
E) the manufacturer of the other driverless car (etc as above)
F) the manufacturer of the driverless trash truck (etc as above)
G) the city
H) the NTSB
I) Dude this just doesn’t work unless you’re in a no-fault jurisdictionReport
What you are describing is the semi-autonomous version. My response to this version is that it might be nice, but not actually that big a deal. The hype is over full autonomy. The latter opens up a huge range of new possibilities: mobility for blind people, ordering a car on your phone and having it pull up to the curb, etc. If the best we can come up with is an override, then we are actually saying that the full autonomy ain’t gonna happen.Report
Absolute full autonomy with no overrides will probably never exist. If it ever does, roads will have changed drastically such that your garbage day scenario is not an issue. Mostly autonomous cars with loud overrides will probably come in at some point, but they will probably require major changes in how we make roads in cities and towns, just like the introduction of the automobile 120-odd years ago did. I predict we’ll have “freeway autonomous” cars pretty soon though, soon followed by mandatory freeway autonomous cars (since it’s much easier to plan around the computer you can maybe even talk to than a mystery human driver). Exit lanes may extend somewhat where possible; I’d expect changeover to happen there.Report
Yeah, freeway autonomous vehicles come first. (Don’t have to worry about running over the family dog then).Report
There’s no mandatory override, at least in the garbage truck case. The autonomous car can happily do just as you describe, patiently waiting behind the garbage truck until it’s legal to get around. You can orient yourself to the road at your own pace, no need to pay attention for the whole trip, and people who can’t drive themselves can still use it if they’re willing to put up with being stuck behind garbage trucks occasionally.Report
With the Hand-off cars, the drivers will be expected to be paying attention at all times. At least until the questions you ask get worked out. People will likely only half pay attention, and less and less as the cars get better and better, but it will probably result in a safety premium all the same. They’ll have to at least act like they’re paying attention.
With the fully autonomous, where we will actually allow drivers to get in the car and take a nap and all that… well, we’re a long ways off from that. But I am not betting against progress advancing quickly.
While it will be a ways before it will allow us to take a nap, it won’t be that long before it can allow us to relax.Report
Naptime will occur on highways only, and with a head slap if you’re needed. (Only partly joking about the headslap — I blame the Germans, our IRB would never approve).Report
“With the Hand-off cars, the drivers will be expected to be paying attention at all times.”
This seems to me to be wildly unrealistic: sit in a seat, paying attention to ever-changing yet uninteresting visual imput, possibly for hours on end, without any likelihood of having to make use of this information. This simply isn’t how humans are wired.Report
Sounds exactly like modern airplane pilots.Report
To the extent that this works, this is a small, highly trained group. Generalizing this to the guy slogging to work after a late night out seems rather a stretch.Report
Of course. which is why you generalize it FIRST to commercial truck drivers (who will use it during the Midnight to 8 am time period, and get more stuff moved for cheaper).Report
I think we have lost the thread here. The claim I am responding to is that in the semi-autonomous version the human will have to pay attention all the time, so as to be ready to take over should the need arise. This is inconsistent with truckers on the graveyard shift. You seem to be describing a situation where the human does not need to pay attention while the vehicle is under computer control. Which returns us to my question about how confident we are that the transition to human control will only happen under predictable, controlled circumstances.Report
I think it’s a silly hypothetical, in the era of big data. There’s enough places with snow for half the year (I live near WV, trust me, we can get the test data if we need it — it may take triple as long, but we can get it).Report
Again, we have lost the thread. What is the silly hypothetical: that humans will have to pay attention, that unexpected transitions to human control might be necessary, or something else?Report
1) Humans will have to pay attention — within 5 years, we’ll be able to get humans some decent REM sleep with the robot at the wheel.
2) Unexpected transitions to human control? — I doubt these will be a usual part of driving. I REALLY don’t expect you to know how to drive a car like a fucking boat. Do you know how? Have you ever done it? Computers are better at crises than humans are — cars are a hell of a lot easier than planes. We already have automated trucks (not street legal but in industry)
Legal conditions are a lot harder to fix than computer programming. We have the computer programming Down (that is, we still need to finish coding it, but Primary Routes are a hell of a lot easier to code than every last city street.)Report
I expect that people will not exactly comply with the law. Most likely people will triage how they pay attention. Not on freeways, but yes in residential neighborhoods.Report
Straight from the Horses Mouth: Don’t worry about cars. Think about commercial trucks. There you already have established “who takes responsibility”, insurance, and the ability of insurance to offer discounts for increasing automation. Once they get to the point where “it works at night on highways — better than people”, you’ll see it pushing over to full automation.Report
Self-driving cars will be a magnificent development for certain situations — in particular, they are good news indeed for the suburbs. Tiny cars like the one in the picture, with software handling things, double the capacity of existing freeways and main thoroughfares. Given a 40-mile electric range and a bit more storage space, they can address a very large percentage of the typical suburban driving patterns. From a policy perspective, I anticipate the real problems arising when we reach the point where there are social advantages to excluding large non-autonomous vehicles from a significant portion of the road network.Report
I think this is exactly right. Google cars right now drive super cautiously, such that it would be annoying to have a meaningful number of them on the road. That said, if we get past that to where you can devote significant real-estate to automated-only use, crazy-awesome things happen.Report
I think that if the car is doing the driving, it’s going to significantly change the need the humans feel to hurry. Because you can be doing something else while in the car – be that productive or fun or social.
So you might not stress out or care so much about being stuck behind a trash truck, or going only the speed limit.
The snow driving is an issue, I’ll grant that. I’m sure it will get worked on, but there’s a lot of what might be called “low-hanging fruit” that doesn’t require snow-driving capability.
For instance, my wife is from Minneapolis. When we visit there in the winter, the roads are cleared after snowfall almost instantaneously, a couple of hours at the most. I do not consider it difficult to train an autonomous vehicle to drive on cleared roads with snow on the shoulders. AV’s may, in fact, be better at detecting ice on the pavement than humans are.
Granted, this isn’t universal. There are places, like rural NW WA where I grew up, that the snow doesn’t get cleared up for possibly several days when it happens. That seems like a problem that will get eventually get solved, but doesn’t need to get solved for the vehicles to be useful to some people in some places.
(Full disclosure: I live in Mountain View. Once upon a time I worked for Google. I see the cars on the street all the time. They don’t seem like a big deal.)Report
“So you might not stress out or care so much about being stuck behind a trash truck, or going only the speed limit.”
Only true if the added time is predictable. The ability to make use of my commute time would compensate for having to leave ten minutes earlier, but getting to work late because I got stuck behind a trash truck is another matter.
As for snow, I too would think that snow banks on the shoulder wouldn’t be a big deal, but actual driving lanes getting nudged around by snow is another matter, and pretty common in my snow-driving experience (Flagstaff, Arizona; western Pennsylvania; and now central Maryland). It is not that uncommon to lose lanes entirely, so that four-lane road becomes a de facto two-lane road, with cars following in the tire tracks of the cars ahead. Even after the snow has stopped, the ploughs make multiple runs over the course of the next few days. The day after the storm, you might not yet have the right lane. Two days later, that lane is fully cleared.
Most of my skepticism is about the claims for full autonomy. I find claims of semi-autonomous cars far more believable, but much less impressive. “This car will drive itself so long as the weather is nice!” is not much of a sales pitch.
It also occurs to me that this would result in a generation of partially-trained drivers. The car drives itself when the weather is nice, so the first the that teenager has to drive for real since he finished driver’s ed is in a blizzard. What could possibly go wrong?Report
It’s not much of a sales pitch in Western Pennsylvania, but it’s a great sales pitch in New Orleans.
I don’t think this is an all-or-nothing. I think there are a bunch of tasks that we will be happy to hand over to AV’s and only be mildly inconvenienced by when they can’t do it because of conditions. And there’s another set of tasks that will be done a bit worse, but a lot more cheaply, and we’ll be willing to put up with that, too.
For instance, Uber is working on AVs. I can imagine them structuring the services as to provide either a driver and car, or an AV, if available. The driver will cost more. And if sometimes the AV isn’t available, that’s business as usual.Report
I am not personally familiar with New Orleans weather, but I am given to understand that it has its own versions of weather that does not constitute “nice.”
For that matter, here in Maryland there is a time of year where you frequently go from nothing to a driving rainstorm and back to nothing again in the space of about five minutes. We call this time of year “summer.”
I refer again to my question about how the transition from computer to human control works. My car has just driven into a storm cell. What next?
Uber: I have called for my automated ride and am happily on my way home when the snow starts, going from nothing to heavy in no time. This happens all the time. So what happens next?Report
Decrease speed until you’ve achieved a safe level of friction with the road. (as this is a car, with sensors, it can measure that…) — also put on your fourways so that the people behind you know that you’re stopping, and quickly.
(unless your tires have completely left the road, I’m not sure they’re putting fording into the cars yet. at least I bloody well hope they’re not prioritizing Arkansas roads).Report
Coming to a stop in the road when both visibility and traction are reduced? Is this really the answer?Report
Obviously you pull off the road, silly. That still doesn’t stop the person behind you from hitting you if he’s still moving at three times your clip. And you leave the lights on, even on the side of the road, so the next person pulling off doesn’t hit you.
(if traction is reduced entirely to zero, it may be prudent to come to a stop in the road, rather than continue off a cliff — at which point one ought to be making enough of a ruckus that the entire neighborhood knows there’s an issue).Report
What if there is no shoulder?Report
Depends. If you can get off the road without wrecking the car, that might be the least costly option. If there are trees lining the path (or god forbid, a cliff), then you slow the hell down and hope everyone behind you does too.
But you’d do that anyway, wouldn’t you?
Computers GOTTA be able to do better than a half second reaction time. Once you prove that, you’re going to want them in crisis situations (particularly when they can talk to one another).Report
The claim is that semi-autonomous vehicles cannot handle inclement weather. You seem at this point to be asserting that they can, so what’s the problem? It seems to me that I see some goalposts moving off in the distance.Report
Triple the training time, and they can. At least as well as humans. (For Highways, we should understand. I’m not willing to put the family dog at risk just yet).
Solvable problems are solvable.Report
Nobody trains people in freaking blizzards. That’d be stupid.
In a blizzard, you pull over, and hope nobody runs into you.
Hell, nobody trains kids on ice either. And that’s also extremely negligent.
I’m told there are courses where one can practice drifting (and presumably driving on ice), but… how many parents use them?Report
I’m not sure if you are agreeing or disagreeing with me. When I first moved to snow country, it was very educational. Fortunately I already had a lot of experience in driving under various conditions, and was able to adapt. If the human only has to take control under crappy conditions, there is no baseline of experience to draw on.Report
Humans are prone to panicking in unfamiliar conditions. I predict it will be pretty easy to get automated vehicles to perform better than humans (and particularly impaired humans), under a wide variety of conditions. (Is fog permeable to non visible light?)Report
Again, we have lost the thread. I am responding to the claim of cars that are semi-autonomous, snow being a condition where it turns control over to a human.Report
I don’t think that we’re talking about cars that can’t drive at all in adverse conditions, merely that they don’t do so particularly safely. Which means that the car will need to be turned over to human control. Often, the person will themselves recognize the adverse conditions and know to take control. Otherwise, if the car is sensing traction problems, it will keep going but will beep the driver informing them that they need to take control.
We deal with this somewhat when it comes to cruise control, when to use it and when not to use it. When to hand over speed control over to it, and when not to.
Semi-automatic in this context takes it to the next level (lane control, distance management, turning, etc), but the overall principle is likely to be the same. And some aspects of the automation can and may remain in force (lane management, for example, where the car will keep you in your lane unless you signal… that sort of thing).Report
We already use antilock brakes, though. Isn’t that a device to deal with poor traction?Report
“I don’t think that we’re talking about cars that can’t drive at all in adverse conditions, merely that they don’t do so particularly safely.”
It’s not clear to me what are the extent of the posited limitations. Why can’t the system handle snow? Is it traction or visibility? If the problem is visibility, I can see how this might shut things down pretty definitively.
(This makes me wonder about that sensor array on top of the car, and how quickly it will get blinded by snow buildup.) (Or road spray from the truck ahead of it.) (Or one well-aimed bird poop.)Report
Why do we expect that humans have better vision than the car does? I expect they have worse. Humans don’t often drive like they’re playing a video game, keeping every single other car in their heads, and reassessing as they regain visibility.
Plus, a car can use IR, or other signaling measures.Report
“Why do we expect that humans have better vision than the car does?”
Because the human is looking through a large windshield. It has a triple system for clearing snow and salt: the wipers; the windshield fluid; and the defroster blowing full strength at high heat. These work imperfectly, but the human can scrunch around to find the one clear spot while looking for a safe place to pull over and work on the windshield and the wipers with a scraper.
What does the computer sensor array have? On the Google car, apparently nothing. That is designed for Mountain View, of course. Presumably the production model will have something more robust, but I have not seen any description of what it might be.Report
The computer sensor array has at minimum wireless, judging by them doing doughnuts in the parking lot when someone hacks it (yes, they fixed it… eventually).
Humans have the issue (generally) of single points of paying attention. And being generally pretty shit at modeling other cars.
You give me a computer, and I can get the velocity of a car, and model its movement (particularly if a signal says it’s autonomous too).
Besides, it’s not like you really need to be able to see a car well. Luma increases during a fogbank? Guess who’s braking.
Pick up a big blur that’s moving in the same direction as you? Assume it’s a car or at least something worth worrying about.
Humans are pretty good at dealing with “suboptimal” conditions (you should see our facial recognition, that’s hardwired), but please don’t assume the computers won’t be.Report
I assume (perhaps incorrectly) that we’re talking about diminished capability, rather than out-and-out non-functioning. Even if it’s a matter of “visibility” it’ll still be able to see, just not with a recommended level of safety (a lower level of safety than for people). But I would think enough to find a place to pull over, or stay on autopilot while it beeps the driver into taking control (perhaps pinging the driver while trying to find a place to pull over safely).
Different danger thresholds may yield different responses. It could start pinging the hell out of the driver with a really low danger threshold so that drivers are driving even when the system could possibly handle it. Or it could be getting weather data and in case of even a Thunderstorm warning, the driver has to take control.
As we start getting closer to completely autonomous vehicles, we’ll have a better idea of what the specific limitations are. In the meantime, I suspect the first generation of self-driving cars will be rather aggressive in turning control over to drivers, and will get less so over time as special condition performance improves.
As @francis says, I am pretty sure we’re looking at a system of constant. But I think the end-game will be completely autonomous vehicles. There are an awful lot of incremental steps between here and there, though. I have no idea how long it will take (other than to say “longer than 10 years” and “less than 100”).Report
Or we could play it the other way, and have autonomous vehicles simply stop (by which I mean pull off the road). I mean, if you’re in sleep mode, it’s probably a BAD IDEA to wake you up and expect you to function better than the computer.Report
1. Full autonomous anywhere outside golf-cart style neighborhood vehicles operated only in non-snowy environments is probably one to two orders of magnitude away in computing situational awareness. But that is a great place to start. There are plenty of people — elderly, handicapped — who could get enormous value, like staying in their home of choice, with that kind of service.
2. Millions upon millions of Americans live in an environment where partial autonomous will be a huge step forward. Me and Burt, for example. Drive the car to the freeway on-ramp and hit the switch? Fabulous (especially if the car computes the speed limit on the basis of the cars around it). You live in an environment which makes you not an early adopter? Wait another decade.
3. State Legislatures will absolutely be forced to adopt no-fault traffic accident laws. With the ACA in place removing lifetime caps on insurance policies, hopefully some of the driving force (ugh) behind accident litigation goes away.Report
Best place to start it increasing the “Oh Crap!” hardware in cars now. Collision detection, adaptive speed, lane guidance, backup and side cameras, and blind-sport warnings — the stuff that’ll prevent the largest classes of accidents.
Read-ending people, changing lanes without looking, backing into things….
You can do that without looping out the driver at all.Report
1. Sure. There absolutely are useful applications for this. It is something different than what is being hyped.
2. As a point of information, I was raised in southern California, moving away when I was about 30. I fully understand a California commute. For that matter, I could make my commute today via freeways for much of the trip. This is slower during rush hour, hence my preferred route. That being said, I am not convinced that a car driving itself for the freeway portion, but with my being required to pay full attention the whole way, would be an improvement. It strikes me more like a torture technique. I think it far more likely that people would cheat and set their attention elsewhere. The question then is how much and what sort of wackiness will ensue.
3. No fault insurance laws suck, for many and varied reasons. Aside from this, one of the promises of all the hype is how safe these things will be. If so, why would the insurance laws have to be changed to accommodate those now non-existent accidents?Report
The freeways, particularly at rush hour, are exactly the place where we want to get to full autonomy for all vehicles. Following distances set for software reaction speeds, not human. Automatic adjustment of cars merging from on-ramps. Information shared about obstructions in lanes.
Tangentially, years ago I regularly drove a 15-20 mile stretch of the Garden State Parkway in NJ where vehicles over 6,000 pounds were forbidden. What I always noticed was that (a) it was as crowded, or more so, than anywhere else I drove in the state and (b) traffic flowed much more smoothly. Almost never saw an accident, despite very close following distances. Ever since, I’ve continued to notice just how much of an impediment big vehicles traveling at odd speeds are to the traffic flow.Report
*snort* only places in PA that forbid trucks are heading over mountains (even small ones). Traffic flow isnt’ improved when people forget to switch to second gear. (particularly with a freaking hairpin at the bottom. ride the brakes the whole way down, and you hope to hell you have enough to spare to actually hit 15mph).Report
Mountains are one of the places where large relatively under-powered vehicles are a nightmare. Heading west on I-70 from Denver is almost always illustrative. At some point, you come up on one big truck in the right lane struggling to maintain 40 mph up the hill, another big truck in the middle lane trying to pass at 45 mph, and all the cars that can routinely maintain 65 trying to squeeze past the “obstruction” using only the left lane.
You and I could sit down for an afternoon and put together the rules that software could enforce on truck behavior that would greatly alleviate the problem.Report
Re 2: our newest car has adaptive cruise control, which is basically that once you’ve gotten into the appropriate lane. It’s fantastic for commuting, especially with kids in the car.
Re 3: that’s a big issue, but keep in mind there’ll also be a LOT more objective information about what happened than there is now. Google’s cars, for example, would be able to create a holographic replay of what happened that would seem to make assessing fault pretty straightforward.Report
First is sensors. Right now the available sensors are such that highway driving is the low hanging fruit. As they get better & can parse more information, more complicated situations can be addressed without human intervention. Connected to this is software to process sensor data as well decision trees for what to do.
Couple that with vehicle networking & other supporting infrastructure changes & it’ll get better.Report
I know a guy writing the algorithms now… (don’t I always?)Report
Most hype is wrong, pretty much by definition. So if you’re focusing on what’s being hyped and worrying about that, I think you’re entirely on the wrong track.
To return to my earlier comment, I see the next level of driver assist being rolled out in two types of vehicles: Neighborhood Electrics and true cars.
Limit the range of the NEV to 10-15 miles and its top speed to 25 mph and there are still plenty of suburban Americans who would buy the autonomous model. Just being able to go out to dinner and get home safe when drunk/over-tired would be cool.
On regular cars, based on the ads I’m seeing on TV a true semi-autonomous version is essentially already available. People just need to be willing to turn over all the steering, not just accident avoidance. That’s not hype; that’s progress.
And if (Honda) programs the car so that autonomous function does not work in anomalous situations — stuck behind the garbage truck, heavy snow — then you’re absolutely no worse off than you are today.
The idea that semi-autonomous cars will be like modern commercial jets — requiring constant attention but no engagement — is I think absurd. Not only would no one buy them, I very much doubt that the federal government would allow them on the road. The times and places where the car would be suddenly overwhelmed and need to dump control onto an unprepared driver while at dangerous speeds seem to me to be so rare that the system could be designed not to get into that situation in the first place. (and if it does, that’s what product liability insurance is for.) Instead, I see this as a field that is ripe for incremental improvement.
Because being able to read a book while my car is navigating rush hour traffic (at speeds ranging from 0 to 45 mph) would be great.Report
“Most hype is wrong, pretty much by definition. So if you’re focusing on what’s being hyped and worrying about that, I think you’re entirely on the wrong track.”
If what you are saying is that fully autonomous vehicles are obvious bullshit and we should treat all the articles about them accordingly, I am OK with that.
“Limit the range of the NEV to 10-15 miles and its top speed to 25 mph and there are still plenty of suburban Americans who would buy the autonomous model.”
I am skeptical of the size of the market. The primary purpose for owning a car for most people is getting to and from work. 10-15 miles and 25 mph would severely limit how many people could use this vehicle for that purpose. How many people are able and willing to pay for an additional vehicle so that they can go drinking?
“And if (Honda) programs the car so that autonomous function does not work in anomalous situations — stuck behind the garbage truck, heavy snow — then you’re absolutely no worse off than you are today.”
I dispute the characterization that heavy snow and/or garbage trucks are anomalous. They represent a small part of overall driving, but they are utterly normal experiences. I also question the idea that we are no worse of. I have mentioned in another comment that this would result in inexperienced drivers being faced with snow.
“The times and places where the car would be suddenly overwhelmed and need to dump control onto an unprepared driver while at dangerous speeds seem to me to be so rare that the system could be designed not to get into that situation in the first place.”
I have no idea what this means. These conditions would presumably be due to unexpected external events. How do you design a system not to get into situations caused by unexpected external events?Report
If you’re talking about pure self-driving cars, the goal is for them not to have user controls at all. And we are a long way from an all-purpose version.
Here’s a review that’s stuck with me.Report
OK, now that I have a keyboard…
Sensors:
Humans – Two, maybe three if the radio isn’t blasting, and they have limited range & sensitivity.
Cars – As many as we want across whole spectrums of inputs and sensitivity.
When I drive I mostly use my eyes & my overall tactile sense. It’s sufficient, but a car can be fitted with optical, IR, UV, ultrasonic, radar, accelerometers, strain gauges, etc. My 2008 Subaru knows before I do that I’m losing traction, first via the AWD system, and then through the traction control system, and finally the ALB system, all of which can assess & react to a loss of traction far faster than I could ever hope to (& I grew up in WI in the 80’s driving cars older than I was, I openly mock people in the PNW for their pathetic winter driving skills). So a modern car can know more about the road conditions than the driver does, it just doesn’t tell the driver.
Outfit a car with an array of optical, radar, and ultrasonic sensors, and it can have an extremely clear 360 degree picture of it’s immediate surroundings at all times, no head on a swivel, and with the ability to ‘see’ better in adverse weather.
Couple all of that with wireless networking, and a car could ‘see’ much further down the road than we can.
Fully autonomous vehicles are possible, but it will require a degree of technological saturation before it would be really effective (a SWAG of ~80% of the vehicles on the road outfitted with advanced sensor suites & networking would be necessary). Once that saturation is hit, cars would just drive according to their algorithms, maintaining safe following, zipper merging, opening up adequate space for larger, slower vehicles that are merging or moving lanes, etc.
The garbage truck? You car knows where it is (because the truck told it), and it knows what is coming in the opposing lane (because those cars also told it, and it has them on the sensors mounted in the wing mirror), and the software driving the car knows when it will have a safe window to pass the truck, and if it needs to it can slow down or speed up a little bit to expand that window.
That accident up the road? The police cars have already broadcast the accident, and the bypass route around it. If the broadcast wasn’t received, your car still can ‘see’ the squad cars emergency lights, as well as the bypass route being displayed by the squad car in the IR band (or perhaps by a drone hovering above the scene).
My prediction? Once the software is worked out (which is what Google cars are doing – figuring out what the software is lacking), I suspect that cars will come outfitted standard with the sensor suites and networking capability, plus the auto drive features (all advertised as driving aids or assistants, if advertised at all), and when there is sufficient saturation of new cars, or retrofitted vehicles, the paradigm will begin to fully shift to an autonomous system.Report
First off, thank you for actually addressing the questions. I would have to go back and check to be sure, but I think you are the first.
As you point out, this scheme relies on saturation, with vehicles talking to each other, and also infrastructure investment, such as the special systems for the police. The hype articles I have seen talk in terms of ten years or so. I don’t see it happening. I see ten years from now local governments starting to talk about maybe funding the upgrades. But I could be wrong.
How much more would this cost than a conventional car? I wonder about reaching the tipping point. Before then, you are asking for people to pay for technology that won’t be fully usable until years later. That seems to me a hard sell, depending on how useful the intermediate stage is.Report
The real cost will be the infrastructure. The sensor suites are amazingly inexpensive & would probably add a fraction of a percent to the overall cost of the car, once the car companies have things setup for it. Cars already come outfitted with a number of these sensors as driving aids (blind spot sensors, collision sensors, traction control, engine monitoring, auto parking systems, etc). Even the networking is there (On-Star). By the time the software is ready, the hardware will be old hat.Report
Richard,
If you can draft another car, you save a bundle on gas. So not fully usable, but close enough.
Oscar’s right, with enough sensors, you do better than a human, pretty much under all circumstances.
Scheme’s not really reliant on saturation, but it is reliant on “good maps” atm. Knowing where to drive isn’t something the computer comes up with in real time. Hence why trucks and major highway routes are such a lowhanging fruit. Double your shipping with little risk? Win win.Report
OMG, I am sooooooooo skeptical of driverless cars, mostly because we’ve had driverless cars for quite some time in the form of trains, buses, taxis, limousines, airplanes, personal drivers for rich people, Dad, etc.. What we need to do is stop referring to them as “driverless cars” and start discussing “robot drivers”. When we do that, I feel like the concept takes on a more appropriately ridiculous connotation.Report
That depends. Is it R2D2 driving, or C3PO?Report
I would prefer IG-88, but I imagine we’ll be stuck with Asimo.Report
You’ll get Twiki, and you’ll like it!Report
I just have to congratulate myself on how great this comment ^^^^^ goes along with my current gravatar.Report
Number of Buck Rogers references I expected to see this year: 0.Report
I’m astounded at the number of otherwise intelligent people that think that the engineers involved in these projects haven’t thought about things like random road debris, children playing, or inclement weather.
Legalities prevent me from commenting in detail but I’d rather have a driverless car reacting to Bambi crossing the road in bumfuck Michigan than oh, let’s pick a number, 83% of drivers.
We’re still a solid decade away from a practical solution to the problem but limited applications will appear within that decade and will advance the technology.
Much like electric cars. The Tesla’s not perfect. It’s not even the best at everything. What it is is a real-world example of the rubber meeting the road in an a way most Americans still think is unfeasible.Report
“I’m astounded at the number of otherwise intelligent people that think that the engineers involved in these projects haven’t thought about things like random road debris, children playing, or inclement weather.”
You appear to have inadvertently overlooked the first and last paragraphs of the post. Or to have answered a single one of the questions.Report
Sorry. I thought the answers were implicit. I’ll make them explicit.
1) Yes. In the near term, no automatic system will break the law for you unless loss of life or bodily harm is imminent (and maybe not even then depending)
2) It steers around the obstruction legally. If it cannot It either pulls to the side of the road to allow the obstruction to be cleared or to allow you to assume control.
3) Likely it is either signaled ahead of time by the police car as to what it should do or it again pulls over and allows the human to take control. Even more likely it knows about the accident at the moment it happened and routes you around it even before the police get there (having been notified by the car in the accident and by your car as well no doubt).
None of these are problems that are not being discussed and solved by the teams working on these vehicles.Report
I can imagine the police adopting some form of standardized emergency signaling system for manual override traffic control once the cars become common enough.Report
Yeah. We can actually expect “human takeover” during detour situations, because the computer is at the moment fairly ill-equipped to do so.
That’s a lot more sensible than “wake human after 2 hours sleep and part drunk, and hope he’s better than the computer”Report
Most of the the things you list are problems to be solved, and will add time to solving them. It’s possible that in between now and then events will overtake everything and we won’t have personal cars anymore because we’re all taking the train. But I doubt that. More likely, as with previous advances (automatic transmission and cruise control come to mind) drivers will be increasingly relieved – or have the option of being relieved – of responsibilities. Until, at some point in the future, we will be wondering what we need drivers for again and laws against sleeping behind the wheel are archaic.
The next things are parking, automatic buffering, and lane control. I think these are all within the realm of possibility (and parking is already done, from what I understand). Then combine that with GPS, and automatic entering and exits, slowing down for particular turns, and making the turns. All of which are possible, and none of which are likely to be prohibited by law. At that point, in good weather at least, you only need to worry about garbage truck situations, getting out of parking lots, and trying to keep an eye out for hazards.
Shorter term, as the above changes are implemented, they will fall into the manual override category.
Longer term:
1) Cars will recognize that they are stuck behind a slow moving vehicle and try to find an alternate route. Garbage truck schedules may even be included in the mapping systems and you may be sent a different route. Garbage trucks may be programmed to pull over and let everybody from behind pass every five minutes or so.
2) Debris and pedestrians are a trickier issue. We’ll likely need a lot better sensor technology. I don’t doubt that technology will reach the point where it will “notice” things on the road that shouldn’t be there. That’s a pretty long ways off, but still in the category of “problem to be solved.”
3) Procedures will be changed. Police will notify guidance systems that will redirect cars. At some point, every lane of every road would be mapped and cars will know which lane they need to be in or where they need to be within the context of the road. This won’t be solved by the cars, but is a systems problem and will involve communication.Report
Sensors are cheap. if we can put them on Wheelchairs, we can put them on cars.Report
The point about things that we all do out of necessity that are technically illegal is an interesting one. I expect it will probably present more of an interesting legal problem than a technical one, though. I don’t think there’s a good solution that doesn’t involve writing an exception into the law (which may exist already in a lot of places). That’s really how that problem should be solved for human drivers as well. Having a law with obvious failure cases that we just overlook isn’t really a great situation to begin with.
Any questions having to do with quickly deciding how to turn the wheels and where to put the car will almost certainly be better handled by the computer than by a person. I’d bet good money that even today, self-driving car computers will handle a blowout better than a human. It’s initially a physics and spatial problem that the computer can have the optimal solution to before a human has fully decided that the problem is a blown tire. Once the initial crisis is over, “what do I do when there’s no shoulder” is a matter of following reasonable rules, just like it is with a human. My guess is that the majority of people whose tires blow out on the Bay Bridge lose their shit and do the wrong thing, even if they get the car under control. Worst case, once it brings the vehicle to a stop, it could ask the human whether to stay put or try to find a safer location to wait for help.
Figuring out when something is wrong with the vehicle is an interesting one. I wouldn’t worry too much about brakes and brake feel–the thing knows exactly how hard it’s pressing on the brakes and it knows exactly how much the car is responding, so anomalies along those lines should be easily detected. Weird noises and the like are another matter. I suppose the question is, what do you want to detect and how important is it that somebody programs the car to detect it? Just about anything a human can sense should be detectable by the car. It’s a question of whether it’s real engineering line item.
Bad weather is going to be the real killer. When the ground is covered in snow and you can’t see the road or lines, you have a serious problem, not to mention the slipperiness and instrument failure that comes along with cold weather. But there are a lot of humans who live in areas where that’s simply not a problem. A large percentage of Americans never have to drive in snow and only rarely end up in a really serious thunderstorm. It’s unfortunate for the people up north, but not every problem is easy to solve. I don’t expect to see that one solved before widespread adoption of these cars in more hospitable places.Report
tf,
” When the ground is covered in snow and you can’t see the road or lines, you have a serious problem, ”
You got GPS. Right now, that solves everything. Routes come preprogrammed– that’s why it’s trucks first, then cars. and Major Routes then minor ones.Report
I could imagine a UI specifically for failover modes in fully autonomous. The car may sit while you respond, but it may prompt the driver for things like 1 – Obstacle: Pass when safe (this works for garbage truck and trash can) 2 – Impassable: Turn around, etc. Eventually, there may be more intelligence or the police will have a way to communicate to cars when widespread enough. Police are not wed to hand signals, they only use them because that’s what communicates to drivers. To cars, it would likely go over the network (which brings a host of security issues).Report
The issue of the brakes not feeling quite right is straightforward: the car monitors them. Is the fluid level OK? How long since it’s been changed? How much brake pad is left? When the car brakes, does it pull in one direction? Does it shimmy? Anything you can do by feel, it can do by direct measurement and rule application.Report
@richard-hershberger
RelevantReport
As the agnostic said of the Bible, “interesting if true.” What I see is a press release from over a year ago put out by a company I have never heard of. I have been unable to find any evidence of rubber on the road.
In any case, what exactly are we looking at here? The promotional material is rather vague (though excellently produced: they clearly didn’t skimp on the production budget). But buried on their site I found this FAQ:
“Can I text while the RP-1 drives?”
“No. You still have to be paying attention to the road and still need to be in the drivers seat. All state and federal driving laws still apply while the RP-1 is in operation.”
In other words, what we really have here are enhanced safety features: collision avoidance, lane control, and so forth. This is far from nothing, but it also is far from being able to do something else with your commute time. As I have previously noted, I find the idea of being required to pay attention to the drive while not actually driving quite unappealing. It would to me be a considerably worse experience than just driving.Report
Here.
Report
I was referring specifically to the company the linked press release/news article was about.Report
Ok. Just seems relevant to note that self-driving cars exist and have extensive road-testing in an article that sounds skeptical of the possibility of such things.Report
This is how it begins. Remember how I talked about retrofitting cars with autonomous units as part of the effort to achieve saturation?Report