Nearly half of Americans don't want a self-driving car

[David Megginson]

There is a continuous discussion about this topic here, so I thought I'd post this.

Looks like the self-driving idea is cooling off for now. Although not with millennials. Interestingly, it appears to be independent of the recent publicized crashes.


https://www.foxbusiness.com/technology/nearly-half-of-americans-dont-want-a-self-driving-car-survey

I don't want to share the road with self-driving cars—especially as a pedestrian or cyclist—until they've had a decade or two of widespread use on segregated roadways. As Boeing showed us with MCAS, you don't find most problems until tech goes into widespread use.

 

[tspear]

I don't want to share the road with self-driving cars—especially as a pedestrian or cyclist—until they've had a decade or two of widespread use on segregated roadways. As Boeing showed us with MCAS, you don't find most problems until tech goes into widespread use.

Actually pedestrian and cyclist are generally much easier to solve for technically, due to speed and the fact that they actually follow more consistent rules. Further, in general most of the sensor systems these cars use work better than the mark one eyeball.
The problem is much more a case of other drivers.

Tim

 

[David Megginson]

Actually pedestrian and cyclist are generally much easier to solve for technically, due to speed and the fact that they actually follow more consistent rules. Further, in general most of the sensor systems these cars use work better than the mark one eyeball.
The problem is much more a case of other drivers.

Tim

I believe that's true on paper and in controlled tests, but I don't believe it's true in real life (at least not in dense urban areas with heavy pedestrian and bicycle traffic, where we all share the same physical space). Human car drivers can be frustrating a**holes, but they behave more predictably and follow fewer potential paths than cyclists and pedestrians.

It will also take kids no time to figure out that self-driving cars are easy to manipulate. Act like you're lunging to run across the road, watch all the self-driving cars slam on their brakes, laugh, and repeat.

 

[David Megginson]

Also, self-driving airplanes are a much simpler problem than self-driving cars--for most of a flight, it will be rare to have a hazard within 10 nm of you laterally or several thousand feet vertically--so if we really were close with passenger-carrying autonomous cars, shouldn't we already have had small passenger-carrying autonomous aircraft in service for 10+ years?

 

[jsstevens]

I don't want to share the road with self-driving cars—especially as a pedestrian or cyclist—until they've had a decade or two of widespread use on segregated roadways. As Boeing showed us with MCAS, you don't find most problems until tech goes into widespread use.

While I'm of similar sentiment, I do object to the "you don't find most problems until tech goes into widespread use" comment. I'd say you don't find all the problems until tech goes into wide spread use. In reality, hundreds of thousands of bugs were probably found and fixed before MCAS went into production. (I didn't work MCAS but I've been working software systems for over 4 decades now.) So many bugs are found in testing, code and system walkthroughs, etc. YOu never find most problems when they're in wide spread use-just the more obscure interactions and boundary cases.

 

[tspear]

Also, self-driving airplanes are a much simpler problem than self-driving cars--for most of a flight, it will be rare to have a hazard within 10 nm of you laterally or several thousand feet vertically--so if we really were close with passenger-carrying autonomous cars, shouldn't we already have had small passenger-carrying autonomous aircraft in service for 10+ years?

1. I never said we are close.
2. Flight is much more complicated when near to the ground than driving.
3. FAA is much stricter than DOT for the level of safety required. DOT basically says you have to be safer than human which is pretty low standard. So far FAA has held the position that you need to be a magnitude safer than a crewed airplane.
4. Pedestrians can Jay walk. Otherwise they do not share the road.
5. Reaction time and speed matter. Cars out accelerate, and go faster, share the same road... Much harder to solve.

Tim

Sent from my HD1907 using Tapatalk

 

[Daniel Katz]

I teach Computer Science and I've told all of my students that the one job offer that would cause me to drop my current 19 year long career and take is if I can make a substantial contribution to self driving vehicles. Sorry, but that's going to have the most major impact on all facets of society in the next 10-20 years. Just like the Internet has in the last 20 and that's why I got into doing what I do.

 

[Everskyward]

Actually pedestrian and cyclist are generally much easier to solve for technically, due to speed and the fact that they actually follow more consistent rules.

??? I don't find that to be the case. Cyclist around here don't seem to follow any rules. Pedestrians are only marginally better, probably because they don't want to be hit by vehicles or bicycles.

 

[jsstevens]

My wife and I took a trip to Sweden to pick up our new Volvo. We were both concerned about local road customs as it was our first experience driving in another country. So we asked our driver (who was taking us to pick up to car) about it. He said "Bicycles have the right of way, always and everywhere." He was right and the cyclists know it! That's probably kind of hard to program in...

 

[Juliet Hotel]

It will also take kids no time to figure out that self-driving cars are easy to manipulate. Act like you're lunging to run across the road, watch all the self-driving cars slam on their brakes, laugh, and repeat.

If kids wanted to entertain themselves by running at moving cars to make them swerve unexpectedly, they can do that now. But they're not. I don't really see how the cars being self-driving will magically make the activity more attractive to them.

 

[jsstevens]

If kids wanted to entertain themselves by running at moving cars to make them swerve unexpectedly, they can do that now. But they're not. I don't really see how the cars being self-driving will magically make the activity more attractive to them.

By being more predictable and by not having an angry driver hop out and take exception to the kids. In Palo Alto, where Google's self driving cars were begin tested pretty extensively (at least a couple of years ago), teenagers figured out that you could take a red octagon of cardboard, stick it on a mop handle and get the self driving cars to stop in the middle of a block. Repeatedly. (I was at an AI conference in San Francisco and chatting with a Sr. Engineer in the Google self driving car program over lunch. She related that story and others about how it was working at the time...)

John

 

[smv]

Actually pedestrian and cyclist are generally much easier to solve for technically, due to speed and the fact that they actually follow more consistent rules.

Ha! How often do you see a pedal pusher stop at a stop sign (or even a red light) when they are not forced to stop by opposing traffic?

 

[tspear]

Ha! How often do you see a pedal pusher stop at a stop sign (or even a red light) when they are not forced to stop by opposing traffic?

Almost never. But regardless they are a slow target for a car. While you have many more variables to watch for in terms of other cars, with much higher rates of impact and reaction times.

Guys, when flying IFR on approach and you break out. What is easier to land? The Cessna 172 at 65 knots, 350 FPM vertical or the Lancair IV at 120 KIAS and 1000 fpm vertical? Reaction time, thinking time... whatever you want to call it. Speed matters and more time to think, fewer variables (a car can go from zero to 60 MPH in eight seconds, a human can zero to 8 MPH in the same eight seconds... what is harder to adjust for?), simpler rules (do you have parallel parked pedestrians?)...

Tim

 

[Zeldman]

If kids wanted to entertain themselves by running at moving cars to make them swerve unexpectedly, they can do that now.

You mean like this fake youtube.??

 

[Juliet Hotel]

By being more predictable and by not having an angry driver hop out and take exception to the kids.

Why exactly would the vehicle being in an autonomous driving mode prevent the driver from getting angry or from stopping the car and getting out?

 

[jsstevens]

Why exactly would the vehicle being in an autonomous driving mode prevent the driver from getting angry or from stopping the car and getting out?

I suppose it wouldn't stop the passenger from getting out, but an autonomous car IS the driver. I suspect you could make the car hit the brakes, then drive on when you run away. Would the passenger have time to look up, figure out what happened, stop the car (which is heading on now), get out and chase the kids? I'm speaking of level 5 autonomous cars here, not the autopilot level 2-3 cars we have today-which do require monitoring.

 

[Juliet Hotel]

Would the passenger have time to look up, figure out what happened, stop the car (which is heading on now), get out and chase the kids? I'm speaking of level 5 autonomous cars here, not the autopilot level 2-3 cars we have today-which do require monitoring.

Why wouldn't they? If the car swerves suddenly, the operator is going to look up and likely look up quickly. Even if the car keeps going, what's to stop the operator from commanding the car to stop and/or backup? Autonomous, even level 5, changes nothing in this scenario. That kids aren't doing it now makes me strongly suspect they won't be doing it then either.

And here's the other thing. Level 5 autonomous vehicles will very likely capture video constantly from all 4 sides. They will very likely store that video for some useful period of time. And very likely provide the operators means to access that video, save it and send it on to other devices. Kids ain't stupid. How likely are they to try to prank passing cars when they know those cars will be capturing video of them in the act?

 

[David Megginson]

If kids wanted to entertain themselves by running at moving cars to make them swerve unexpectedly, they can do that now. But they're not. I don't really see how the cars being self-driving will magically make the activity more attractive to them.

Because autonomous vehicles have to behave more conservatively, so it will be fun and predictable for them. Just wait and see. It will also work for pedestrians who want to jaywalk: take one aggressive step out, watch all the autonomous vehicles slam on their brakes, then walk on across the street.

They'll sort all of this out eventually, but it will take many more years than the optimists think. Right now, they're focussing on only the easy problems, though those are hard enough, based on my own experience with AI training. Things will get really tricky when you move to more complex human behaviour and system change (stuff engineers don't like to think about, except in the most oversimplified terms).

 

[Juliet Hotel]

Because autonomous vehicles have to behave more conservatively,

Do they? An autonomous vehicle will recognize exactly what the threat is, exactly how fast its moving and exactly how much braking and/or steering is required to avoid it and it will do all of that in a fraction of a second. Far faster, more reliably and more accurately than any human could muster. Honestly I think if the game is to get the car to swerve a lot, you're going to want to run out toward one driven by a human. And like I said earlier, the cars will have cameras. Kids love to be destructive and bad but they tend to like to do it when they know there will be no video evidence that could be used to convict them.

 

[David Megginson]

Do they? An autonomous vehicle will recognize exactly what the threat is, exactly how fast its moving and exactly how much braking and/or steering is required to avoid it and it will do all of that in a fraction of a second. Far faster, more reliably and more accurately than any human could muster. Honestly I think if the game is to get the car to swerve a lot, you're going to want to run out toward one driven by a human. And like I said earlier, the cars will have cameras. Kids love to be destructive and bad but they tend to like to do it when they know there will be no video evidence that could be used to convict them.

My sophomore roommate used almost the same words when he was explaining cruise control to me back in 1983—it reacts to speed changes faster than a human possibly can, etc. etc.

Then later, when I actually drove a car with cruise control, I realised that I could always outperform it, because I knew the hill was coming while it didn't. It was reacting, while I was anticipating.

It's similar to flying an ILS or LPV approach, where you know there's going to be a bit of descending air over a stretch of cold water: the human anticipates by adding a bit of power before they're over the water, while an autopilot has to wait until the plane starts dropping below the glideslope and then react to the change.

 

[Juliet Hotel]

My sophomore roommate used almost the same words when he was explaining cruise control to me back in 1983—it reacts to speed changes faster than a human possibly can, etc. etc.

Then later, when I actually drove a car with cruise control, I realised that I could always outperform it, because I knew the hill was coming while it didn't. It was reacting, while I was anticipating.

Really? You're really going to reference an anecdote about 80's era mechanically controlled cruise control technology and claim that since it was wrong then, it must still be wrong now? Really?

Well ok suit yourself. So here it is. Your friend was wrong. Mechanical cruise control could not react faster or more accurately than a human. But it could and still can outperform a human in terms of fuel efficiency. Keep the vehicle in cruise on the highway and you'll get better fuel economy. Been proven many times.

It's similar to flying an ILS or LPV approach, where you know there's going to be a bit of descending air over a stretch of cold water: the human anticipates by adding a bit of power before they're over the water, while an autopilot has to wait until the plane starts dropping below the glideslope and then react to the change.

Yeah, no. Not like that at all. Here's why. A driver scans the road ahead using a pair of mark I eyeballs. Autonomous vehicle technology makes use of cameras for visual reference but also incorporates lidar and/or radar to scan the road ahead. Do you have radar in your head? Can you see people through fog? Can you see people hidden on the other side of solid objects? I know can't and no one that I happen to know can. If you can, then yep you may be right, maybe you can react faster and more accurately. But no one I know can do that.

Oh and getting back to your 80's cruise control argument there is one other thing I should mention. Many class 8 semi trucks being built today come with adaptive cruise control technology. In these systems the vehicle's cruise control system utilizes a GPS and a database to inform the system about the terrain ahead so the system can make more fuel efficient choices about when to downshift (on upgrades) and when to coast (on downgrades). So the very things you talk about that you could do better than cruise control back in the 80's? Yeah cruise control does that now. And yes, it does it better than you because it knows exactly what the grades are and exactly how much extra fuel to add and exactly how coasting can be allowed in order to maximize fuel efficiency.

 

[flyingcheesehead]

A driver scans the road ahead using a pair of mark I eyeballs. Autonomous vehicle technology makes use of cameras for visual reference but also incorporates lidar and/or radar to scan the road ahead. Do you have radar in your head? Can you see people through fog? Can you see people hidden on the other side of solid objects?

The most impressive example of this is this one, from shortly after Elon first mentioned something about bouncing the radar under the vehicle in front of you to "see" vehicles further forward - It predicted the crash and hit the brakes before the accident even happened (when you hear the first four beeps):

 

[IK04]

Several auto makers have adaptive cruise control/collision avoidance systems that can "see" a sudden deceleration of a vehicle in front of the vehicle in front of itself.

My Toureg does it and it sometimes gets annoying when that loud beep goes off because Nancy missed her exit and hits the brakes!

 

[tspear]

Several auto makers have adaptive cruise control/collision avoidance systems that can "see" a sudden deceleration of a vehicle in front of the vehicle in front of itself.

My Toureg does it and it sometimes gets annoying when that loud beep goes off because Nancy missed her exit and hits the brakes!

I have tried multiple adaptive cruise controls in rental cars. BMW, MB, Toyota to name a few. I absolutely detest them; they get way to close to the vehicle in front of them. Sure, in theory the radar might be able to see under the car in front of them, but the result is always jerky behavior; or an accident because no automation so far is perfect and the machine did not leave enough room for the human to respond.

Tim

 

[denverpilot]

Several auto makers have adaptive cruise control/collision avoidance systems that can "see" a sudden deceleration of a vehicle in front of the vehicle in front of itself.

My Toureg does it and it sometimes gets annoying when that loud beep goes off because Nancy missed her exit and hits the brakes!

The Subaru has an absolute hissy fit about vehicles slowing for left turn lanes in front of it. After watching it for a while I believe the Eyesight system sees brake lights get brighter and starts to brake. It then holds on to that vehicle as a lead vehicle even when it has fully changed lanes to the left turn lane. Part of its ability to track the lead vehicle around sharp corners, but it messes with you when in straight lanes and the lead moves over out of it. Takes it a while to release and see the empty lane ahead.

I’ve learned to just stomp on the gas pedal before it brakes too much.

I have tried multiple adaptive cruise controls in rental cars. BMW, MB, Toyota to name a few. I absolutely detest them; they get way to close to the vehicle in front of them. Sure, in theory the radar might be able to see under the car in front of them, but the result is always jerky behavior; or an accident because no automation so far is perfect and the machine did not leave enough room for the human to respond.

Most have a distance setting. Did you use that?

The Subaru has four. They range from:

4. The vehicle ahead is in the next county.
3. Comfortable in stop and go traffic but lane dodgers will jump in front of you.
2. Slightly cringey at speed and braking applications have to be slightly uncomfortable.
1. Tailgating / Why does it even have this?

I use 2 or 3 depending on how much I feel like being tossed around the cabin. LOL. Honestly 3 is great. Very relaxed ride watching morons dart around.

Since it doesn’t understand stop signs or lights unless there’s a car stopped ahead of you there, I’ve noticed I’ve already built a habit of just bumping the speed down to the minimum of 25 as I’m approaching the stop. It decelerates very smoothly and if you wait too long, applies brakes lightly, but with planning you can kinda emulate the “hypermile” people and have it slow consistently and smoothly to 25 without braking and then manually brake to the stop.

If you’re coming up on stopped cars ahead, same deal... just bump the speed down and give it time to acquire the stopped car ahead and it’ll smoothly complete the entire full stop and go into hold mode, holding the brake pedal for you. When the lead car moves, tap either the speed up toggle once with right thumb, or the gas pedal, and it goes back into follow the leader mode.

It’s wild. Can do an entire drive anywhere without touching pedals except when there’s no car ahead at any stop, and parking at the far end.

But yeah. Jerky ride or too close following is possible on most of them if you don’t set the following distance.