Well, “in our lifetimes” is a pretty squishy time period. Aren’t most of the people on this board in their 50s and 60s?
Also, is that factoring in COVID or not?
I would argue that non-LIDAR guidance is actually necessary for a truly Level 5 system just because of the limitations of LIDAR (though it is still an excellent capability to have to supplement visual information). However, this is going to require fundamental advances in computer vision and machine intelligence science that are not on the horizon now. I do not believe that Tesla or anyone else can predict when those will occur, and without those advances automated piloting systems are going to be prone to potentially grievous errors. So, I should note, are human drivers, but I think the public in general is more tolerant of human error than it is of machines that are not as safe as promised, and that liability is why a lot of established car companies are pulling back their plans to field autonomous vehicles in the foreseeable future.
I wouldn’t sαy that we’ll never see a Level 5 autonomous vehicle “in our lifetimes” but it will take some revolutionary developments in both computer vision and the robustness of automated piloting systems, as well as the interface between autonomous piloting system and user such that the user can communicate fine-grained directions to the vehicle such as driving off pavement. It would also require several infrastructural changes, and of course, it will transform or virtually eliminate whole industries like driver’s insurance so there will be a large degree of legislative resistance from corporate lobbies on that front. I think it more likely that we’ll see the automation of over-the-road (OTR) cargo transportation and limited applications of autonomy (Level 4) long before we’ll see any real attempt at deploying Level 5 vehicles.
Stranger
While I think automatic driving through snow is going to be nigh impossible anytime soon, I do think it’s very possible self-driving cars will be able to handle other adverse weather effects such as heavy rain and heavy fog better than humans can.
What I mean is that autonomous vehicles that don’t use lidar seem unlikely to succeed. I agree that a successful AV will need multiple methods of scanning, and perhaps some peer to peer networking.
I’m 56, so we may have different thresholds. 
In any case, no disagreement on the likely path.
I have no idea how to start calculating, however the discussion is prompting two questions…
- What (or how would we calculate) % of cars will never (or so rarely it doesn’t matter) be required to drive in snow?
I’ve mentioned it here before - but again - in the last 4 years, I’ve never had to drive in snow, it is so rare here that it is not a consideration when buying a car.
- What % of trips / cars / drivers could are fully on surface streets or otherwise commuting on regular routes? Routes where you the chances of having to go off the road for any purpose are so small that you don’t need to prepare for it?
Could we get fully autonomous cars with restrictions on where / when they can be used (much the same was as some drivers are NOT allowed to drive at night). And once we get this - how will it speed the overall adoption?
Why do you think camera isn’t enough?
I know that lidar makes things easier, but I don’t think it’s impossible without lidar.
I think some kind of optical sensors are crucial, but are also fighting an uphill battle when it comes to rain, snow and fog…lidar isn’t great in those, but it’s getting better, and it’s just hard to replace the sensory quality of a known, very discrete light source (as opposed to ambient lights/headlights).
On top of that, depth perception is critical to reading objects, and lidar is terrific at that. Not sure cameras can replace that either, particularly on dark/wet roads.
It’s opinion, granted. I’m taking the part of Ford, GM, etc against Elon Musk…he’s pretty crazy, but he’s also put commercial rockets into space, so he might turn out right.
Edited because the board software turned my shrug emoji into some kind of Greek masculinity symbol!
“I feel extremely confident that it will be possible to do a drive from your home to your office most of the time with no interventions, by the end of the year. We can almost do this already with the leading edge alpha build that I’m driving in the car.”
“Everything I’ve ever said would come true, did come true. It may have come true late, but it did come true. So punctuality is not my strong suit, but I always come through in the end.”
“I think we could see robotaxis in operation with the network fleet next year.”
I should invite Mr Musk to send a robocar to my address. Will it avoid the house that Garmin GPS thinks has a road running through it? Can it evade attack by a bear on a vaguely marked, twisty, rutted dirt track? And stealthy bear traps the neighbors devised?
Amen. Folks in this thread seem obsessed with total and complete autonomy. Within a certain number or reasonable parameters, something like (random guess) 70% of all driving in the US could be done more safely by an autonomous vehicle than a human being.
It’s really the insurance liabilities that are the big stumbling block. We are used to humans killing humans with vehicles and have developed a long-standing legal and insurance industry that deals with this. However, this is not the case with autonomous vehicles, which will kill humans. We just need to wrap our heads around that and we will end up with many fewer deaths.
One reason for the interest in total autonomy is that some of the proposed designs for self-driving cars have no way to be operated by the passenger; no steering wheel, no gas pedals, etc. So someone thinks completely autonomous cars are a possibility.
As long as such a car is smart enough to avoid snow covered roads, then you could still have cars with no steering wheels or pedals. The car could even check the weather forecast to see if snow is likely, before commencing a trip.
Snow, feh. Try slippery road surfaces covered with sand, mud, black ice, spilled oil, or roadkill guts. Will a robocar scan weather forecasts for violent thunderstorms, tornadoes, and other extreme weather? I’ve driven the I-10 around San Bernardino with 60 mph crosswinds and the I-40 in Kansas with a tornado chasing me. I see deer, turkeys or quail lurking beside my mountain highway and I slow to avoid inevitable suicidal dashes - will AVs learn animal behavior? Will AVs safely haul us from Sacramento to Guatemala City?
I’ve said this before, but when they can build a robot that can play centerfield - which I’d think would be a much more manageable feat, in a much simpler environment - I’ll at least be open to the possibility that AVs might be coming sometime. But if nobody can build such a robot, then nobody can build a decently functioning AV.
Here’s a point I’ve been meaning to mention for some time, and haven’t seen brought up often. I don’t know if this is an insuperable barrier to ever achieving Level 5, but it seems to me that it could become a potentially serious problem in Levels 3 and 4.
The very process of making cars smarter is making drivers dumber. This probably started with automatic transmission. When you have to know how to shift gears, you gain a better sense of what is going on with your car. With auto, all you have to know is “right foot down = faster.” It may not seem like a big thing, but you lose the innate sense of the engine’s power band, the understanding that you should often be in a lower gear if things are slippery, and other things that give you a connection to what the car is doing and how to handle it. And yes, you can learn these things in an auto, but the vast majority of people don’t.
The next steps in making drivers stupider, which came along at about the same time, and are now mandatory in the U.S., were ABS and stability control. For the record, they both greatly enhance safety, and I’m not opposed to them or the regulations that required them. But before they were common, drivers had to pay more attention in wet and slippery conditions, and experienced drivers developed handling skills to do some of the things that ABS and stability control handle automagically. In northern climes, pretty much all drivers learn how to handle a car in the snow. It’s a very useful skill that has application in other conditions, like rain, and on gravel or dirt roads. But elsewhere, a lot of people never get the chance to learn these skills.
Traction control can prevent or mitigate accidents caused by poor car control, but so can better driver training. As some here may know, for about ten years I was a High Performance Driver’s Ed instructor, teaching race driving techniques to people in their ordinary street cars on closed race tracks. The track where I started doing this also offered a one-day course called Accident Avoidance, intended to teach young drivers advanced car control to help them…avoid accidents. How to avoid or handle skids, do panic braking, and much more.
But that kind of training is expensive and time consuming, so most 16-year-olds only get a few hours to learn the very basics of operating a car before they’re allowed to pilot a potentially deadly 3,000-pound projectile on the public roads.
Furthermore, it seems to me that the number of people who really don’t like to drive is increasing, and a lot of younger people seem to see driving as time wasted that they could be using to Instagram or Snapchat or whatever it is the kids are doing on their phones these days. They don’t look at driving as a skill that is important to master for its own sake, and that can even be fun.
So every new “driver’s aid” that’s introduced – blind spot detection, lane minding, radar-controlled cruise control, etc., etc. – means drivers don’t have to pay as much attention while driving or know as much about how to handle their vehicle.
The problem is that, as people have posted above, there are a number of driving situations that AVs won’t be able to handle for quite a while, if ever. AIUI, at Levels 3 and 4, in some of those circumstances, the cars will expect the driver to take over. But drivers raised with all these aids will hardly be prepared to do so.
As the cars get smarter, drivers are getting dumber. As those two curves meet, there will be a place where the car is smart enough to know it can’t handle the situation, but the average driver will not be skilled enough to handle it, either. The result will be accidents. And who will be responsible?
Will this stop the development of AVs? Maybe not. But it could become a big obstacle to overcome along the way.
Agreed. It may take a couple generations, but we’re going to “Idiocracy” people with cars to the point where people will be more afraid of taking over than risking the car driving off a cliff in a snowstorm.
“Please resume control of the vehicle.”
“GO AWAY, I’M BAITIN’!”
The idea that any technology being hard to use therefore made users more competent, and therefore safer, is hard to swallow.
It reminds me of the idea that cars should have a giant spike in the steering wheel instead of an airbag, the argument being that drivers would be so careful that accident rates would plummet. I think that’s absurd: you’d just end up with more dead drivers because a squirrel ran into the street or whatever.
It isn’t that driver’s assistance technology makes cars easier or safer to drive that makes drivers less competent or aware; it is that as features become available that automate essential functions, drivers naturally tend to pay less attention because they often don’t have to. An example is lane holding driver assistance; many vehicles have a land hold feature that will keep the vehicle centered on the lane so the driver can take their hands off the wheel. The intent is to permit a driver to do other things with their hands, like operate music and environment controls without drifting around, but the net effect is that drivers decide that they can engage in other activities requiring a significant amount of their attention such as reading or watching videos even though the instruction manual specifically says that lane holding is not an ‘automatic driving’ feature. An extreme case of this is the Tesla Autopilot which is notorious for holding lane accurately for long durations and then without warning suddenly changing lanes or drifting if lane markings aren’t clear or it just gets confused. I’ve had two incidents earlier this year of Tesla vehicles ping-ponging across multiple lanes while the driver was totally oblivious, and it is an example of how an automated driving feature that is 98% reliable just isn’t adequate and is actually more hazardous than relying on a human driver to pay attention despite the fact that people suck at attentiveness for long durations.
Subaru introduced a Lane Assist feature on vehicles starting in 2016. As it is explained in the instruction manual, when activated the feature prevents the vehicle from drifting across marked lanes (unless the driver sharply turns the wheel which deactivates the system) but crucially does not center the car in the lane because the system is not intended for hands free driving; it is simply a tool to prevent an inattentive driver from drifting. Despite repeated warnings in the instruction manual (which apparently nobody reads) and a large publicity campaign including mailings to owners, there was a hue and cry by owners about how the functionality didn’t work, because they wanted to be able to not touch the wheel.
Level 2 vehicles are prone to driver inattentiveness (or sometimes driver confusion because of the additional workload of warnings and alarms), and Level 3 vehicles are problematic because they require the driver to take over in an emergency even though the features they offer virtually assure that the driver will not be paying full attention to driving. Level 4 and Level 5 vehicles are presumably able to handle contingency situations via some failsafe protocols, but such systems have yet to be demonstrated, nor is there a comprehensive test protocol to validate the functionality and reliability of such systems. When Elon Musk assures us that Tesla will be able to deploy a fully automated Level 4/5 vehicle with software upgrades, that really needs to be backed up by more than just marketing bluster, and specifically, it needs to be shown that such a system can deal with the kind of driving hazards that commonly result in accidents or near-accidents with an efficacy that is at least as good as if not significantly better than a human driver. I have yet to see any demonstration of this capability whatsoever, and as noted above, the Tesla system has displayed consistent and pervasive errors that give pause to any wide scale deployment of a fully autonomous piloting system.
Stranger
Then it’s good no one is saying that. The less cars did automatically, the more skills drivers had to have just to operate them competently, and the more attention they had to pay while driving. But no one’s saying older cars were ipso facto safer. They weren’t.
Of course, many of the reasons cars are safer today than 60 years ago – e.g., seat belts, air bags, crumple zones, collapsible steering columns – have nothing to do with actually controlling the car.
But ABS and stability control were major advances in safety largely because they could perform in ways that not even the best drivers could match. Before those innovations, we were taught to pump the brakes rather than let them lock up in the wet, and we were told “steer into a skid.” Both are pretty useless as instruction unless you actually put a driver on wet pavement and have them try to brake, or put them into a skid and show them how to control it. (Which I have done.) Not one driving student in 10,000 was ever taught these things on purpose, and even professional drivers who have practiced them can’t outperform ABS.
So no one’s saying we should drop those systems. They have saved lives.
However, the “driver’s aids” I mentioned are far more oriented to convenience than safety. If your mirrors are adjusted properly you won’t have blind spots. If you’re paying attention while driving, as you should be, you don’t need lane minding. And so on. They have had the overall effect of making drivers lazier, less attentive, and more distracted than they already were by their entertainment systems and cell phones.
Again, no one has ever seriously suggested that, because it is absurd. (I had actually considered mentioning it in my post.) But as a thought experiment it highlights the importance of paying close attention to what you’re doing while in control of a dangerous piece of machinery.
Safety improvements and driver conveniences have lulled us into a false sense that hurling human bodies around at high speeds can be done without risk. It can’t.
And in the meantime, the skill level of the average human driver is going down.