Self Driving Cars?

[NW-Bound]

Exactly. And we are back to having to define new laws to address all this new technology.

 

[Midpack]

Since 50% of todays fatalities are due to folks not wearing their seatbelt, (for 13% of drivers) that is the low hanging fruit. Note that todays forward crash mitigation systems tension the seat belts if a collision is expected.
So it appears that primary seat belt laws and moving to seat belt not fasten car does not move, are the current low hanging fruit for reducing fatalities.

I found a CDC summary that supports your 50% rate, but it's from 2009.

Fortunately 2015 (latest) data shows 28% of auto crash fatalities were unrestrained occupants, so we've gotten much better? Or air bags have helped?

Unfortunately alcohol was involved in 47% in 2015. But I assume there's overlapping data - e.g. unrestrained drunks...

And distracted driving fatality stats can't be reported with the same accuracy.

Fatality Facts

 

[Totoro]

Lots already written re: insurance and self-driving cars.

One side note: Many (most?) car insurance is actually already structurally loss making today. (including investing profit, so not just underwriting). Makes one wonder .. why bother?

 

[NW-Bound]

...What I occasionally get is the same as the test below would demonstrate to you. Try it...

Just noted the above. In case you were talking to me, no I do not own a Tesla to enable the capability.

I am not into cars, and a Tesla is too expensive for me. If I were to go out to buy a new car, $40K is all that I am willing to spend. My wife is more frugal, and she would say $30K, even though we can afford more. We just do not want to get mad if some envious guy scratches our car at the supermarket.

My son got an Audi S4 two or three years ago. He kept it meticulously shiny for the 1st year. Now it is all dirty as he does not care anymore.

 

[eroscott]

Just noted the above. In case you were talking to me, no I do not own a Tesla to enable the capability.

I am not into cars, and a Tesla is too expensive for me. If I were to go out to buy a new car, $40K is all that I am willing to spend. My wife is more frugal, and she would say $30K, even though we can afford more. We just do not want to get mad if some envious guy scratches our car at the supermarket.

My son got an Audi S4 two or three years ago. He kept it meticulously shiny for the 1st year. Now it is all dirty as he does not care anymore.

I was talking to you (or anyone else that wanted to try the test).

The test described DOES NOT require a Tesla. Just a normal car in regular cruise control. The tester is looking straight ahead with their hands on the steering wheel. The passengers (left) hand is on the steering wheel. After a minute of traveling likes this the passenger moves the steering wheel just a little toward either line on the right or left. The tester (driver) will see how QUICKLY they feel a change when it deviates even a TINY bit from what their perception of the situation is. HTH

 

[NW-Bound]

OK. But if I asked my wife to assist, she would say "Are you crazy? Want to get us killed?"

Just joking. But I know what you mean. And I agree that a gentle "loss of lock" is OK, as long as you are looking at the road. I worry mainly about something less benign.

And I thought about this some more. Of course when on the freeway, the car knows that it is on a nice straight road with very gentle curves (it has the map). So, it can be programmed to make accordingly smooth and small steering corrections, because that should be all that is needed.

But if you go on a rural highway with sharp bends, the car knows that it has to make large and fast steering wheel movements. And it will allow itself to do that. The situation is now different.

 

[explanade]

Maybe the first SDCs will be expensive but the kind of resources they're bringing to bear on it -- paying premium on compensation, buying up startups, doing very high-resolution 3D maps, extensive testing, etc. -- suggests they are aiming for scale.

If they can only deliver this tech in $100k cars, the market would be too small to spend this kind of money to make it happen.

Plus the more SDCs on the road, the more they can leverage telemetry data-collection, 3D mapping and V2V communications to improve the reliability of the tech.

So I would think within the first 5 years of launch, there will be $30-40k SDCs available and then gradually migrating downwards to lower price segments.

 

[NW-Bound]

All this talk about maps gets me to ponder the problem some more. When I first heard of the self-driving capability, I was thinking of a car that uses its vision cameras, assisted by lidars or radars, to drive on the road. It is the same as humans driving cars today. You use GPS only to place where you are on a digital map, which tells you where to make a left turn or right turn to get to that shopping mall that you want to be. If your handheld GPS or smartphone suddenly goes dead, you are not going to crash. You drive around a bit, orient yourself, read street signs or use memory, and still get from point A to point B.

Now, it dawns on me that many car makers are not doing that at all. They use GPS to guide the car, and so they need good maps. GPS only gives you a latitude, longitude and altitude. They then place that coordinate on a road, and even on a particular lane on a highway or freeway. They do use the camera, but it is not as critical, as they only use the camera to correct for GPS errors.

GPS errors, you ask? The Waymo guy has a video to show how GPS can put you on the wrong side of the 2-lane rural road. This should not be a surprise to anyone who spends time watching the GPS position on his smartphone. It wanders about several feet sometimes, and in a random manner. When Tesla combines data from the fleet to make the map, the errors average out when you have thousands of cars traversing that lane over time.

So, the map eventually gets very good. But your present position, where you are right now as GPS tells you, may have several feet of errors. That is where the camera data comes in, for the car to seek to the center of the lane. On an unmarked rural road, it must see the road borders, then keeps itself to the right lane. On a very narrow bridge with a single lane, it must know the protocol to pass by yielding to the car in the opposite direction when appropriate. That last step, that's level 4.

If one wants to find out how well the camera vision works by itself, meaning driving the same way humans drive with eyes, he should find the position of the GPS antenna and cover it with tin foil. Then goes out on a drive to see how the car can see the road boundary, or the center of the lane.

 

[NW-Bound]

Maybe the first SDCs will be expensive but the kind of resources they're bringing to bear on it -- paying premium on compensation, buying up startups, doing very high-resolution 3D maps, extensive testing, etc. -- suggests they are aiming for scale.

If they can only deliver this tech in $100k cars, the market would be too small to spend this kind of money to make it happen...

Oh, I am sure they plan to eventually have the cars cheaper than that. The question is how fast, and eventually how cheap.

Would you not love to see that spreadsheet in the laptop of the chief engineers, showing how much they budget for each new item in a level-4 car, and their projection how they drop over time? And then to compare one car maker to another?

Waymo has a lot of sensors compared to Tesla. It is going to be interesting to see how Waymo competes.

 

[meierlde]

To partly look at the cost of a camera look at how the cost of a TV camera has fallen, given that a large number of cell phones will act as a tv camera (at about the old analog tv resolution). Or a web cam for $20 for example (albeit you need a computer also)

 

[NW-Bound]

Yes, cameras are dirt cheap. It is very obvious. The question is how well a computer can extract data out of it. I talked about it in an earlier post.

Waymo prefers to use lidars. A lidar can measure distances very precisely. Radars are no match as they have much poorer resolution. A camera picture has no direct distance information in it. One has to deduce the distance to an object in some way with a picture.

Waymo says its new lidar has such fine resolution they can tell if a pedestrian face is turned towards them or not, and say that that is useful. Lidars do not see colors, so they still need vision cameras for other tasks, such as reading traffic lights.

Now, some companies use all vision cameras and no lidars. So, maybe they have some tricks up their sleeve that Waymo can't? Maybe they have some patents that lock up the technology?

But that gets away from my earlier post. It's not at all about cost. It is about driving without GPS. One can do it with lidar, radar, camera, or all 3 as Waymo will be doing. But no GPS. GPS can get jammed or spoof'ed very easily. Or it simply does not work in urban canyons such as Manhattan, or in tunnels.

 

[Elbata]

I have been driving a Tesla Model X (SUV/CUV) for just over 1 year and just turned over 27K miles today. I probably have driven 90%-95% of those 27K on Auto-Steer/Pilot.

Yes, I've taken multiple road trips to get that many miles and this is one place the car shines. It stays steady centered in the lane and with a 2.5 second delay behind the car in front of me (if there is one). It makes for a MUCH more relaxing drive. You still need your hands on the wheel but you can look around more and have better conversations.

I also use it a lot around Chicagoland and even on 35-50 MPH roads. Why? It is looks at TWO cars ahead with the latest changes of using radar primarily and the camera secondary. So even when a van is in front of me it detects the little car in front of it (van) slowing down or braking even if the van brake lights have not come on yet. It is also very steady in the center of the lane and keeps a safe distance (again 2.5 seconds).

FYI, I use the maximum following distance unit of 7 which roughly translates to 2.5 seconds (so more distance at higher speeds and less distance at lower speed).

I drove 100+ miles today to a sporting event where 70% was on the highway and 30% was on 35-55 MPH roads. The Tesla Model X drove 95% of this 100+ miles. Relaxing drive.

ASIDE and FYI ---- EVERY TIME you engage autopilot a message pops up to remind you to KEEP YOUR HANDS on the wheel. If you don't and ignore it's warnings 3 times THEN you have to pull over and put the car in park before you can then drive and re-engage autopilot.

Below image is agreement I had to click on to enable AutoPilot -- done at walkthrough.

Love this. Thank you both for sharing. With adaptive cruise control and center lane driving, that alone would account for most of my driving.

And to be able to travel coast to coast with such a wonderful car, how nice would that be.

I don't see anything but upside to self driving cars. It'd relieve congestion, save on car insurance, and save lives. No more threads about road rage

Interesting article: https://www.technologyreview.com/s/603853/machine-learning-and-data-are-fueling-a-new-kind-of-car/

 

[NW-Bound]

Another thing about lidars is that they can detect a deer standing off the road at night, and know how big it is. That might help Ron with his deer collision avoidance problem. Well, a camera can too, but it sees only the eyes of the deer in the dark, and only when it is looking at you.

However, that is an enhancement. Level 4 or 5 says that you have to match human's ability first. The other stuff is extra but not required.

 

[meierlde]

Yes, cameras are dirt cheap. It is very obvious. The question is how well a computer can extract data out of it. I talked about it in an earlier post.

Waymo prefers to use lidars. A lidar can measure distances very precisely. Radars are no match as they have much poorer resolution. A camera picture has no direct distance information in it. One has to deduce the distance to an object in some way.

Waymo says its new lidar has such fine resolution they can tell if a pedestrian face is turned towards them or not, and that is useful. Lidars do not see colors, so they still need vision cameras for other tasks.

Now, some companies use all vision cameras. So, maybe they have some tricks up their sleeve that Waymo can't? Maybe they have some patents that lock up the technology?

But that gets away from my earlier post. It's not at all about cost. It is about driving without GPS. One can do it with lidar, radar, camera, or all 3 as Waymo will be doing. But no GPS. GPS can get jammed or spoof'ed very easily. Or it simply does not work in urban canyons, or in tunnels.

It looks like there is the short term issue of auto navigation similar to what a driver does when following a road, and long term navigation i.e. when to turn.
But the difficulties with GPS navigation showed up in the recent bus crash in MS where the GPS took the bus over a railroad crossing that hung the bus up due to steep ramps on both sides. Or the reports of the guy following GPS in northern nevada into the absolute middle of nowhere where he died of exposure. Or the number of times drivers used car gps units and ran their truck into a bridge....
So I suspect the gps provides the needed turns that don't follow the road while cameras, lidars, radars are concerned with the next 500 feet.

 

[NW-Bound]

Yes. I tried to say something similar in previous posts, but people have different ways of phrasing it.

There is an engineering term call GNC, short for Guidance, Navigation, and Control (see Wikipedia). These are 3 different tasks for a hierarchy in driving or flying a vehicle from a point A to a point B.

A human driver performs Control to keep the car in the lane, and to control its speed. It's a low-level task that a beginning driver learns first thing.

He also plans the car path, to change lane as necessary, making turns, avoiding objects. That's Guidance. That's also his wife yelling at him that he just missed a turn.

His smartphone with GPS and map display is only for Navigation and also Guidance, telling him where he is, and what general direction he should be heading to get to where he wants to be.

There's some overlap, and a sensor or subsystem can be used in a mixed way.

PS. My career in aerospace has been mostly in GNC (airborne things, but no car), though I also did a few other related odd jobs.

 

[NW-Bound]

Umm... Detecting and avoiding steep railroad crossings that can hang a bus has no current classification in GNC. This is something new!

I think it should be in Guidance, as it is a higher level task, if he knows to look for another route.

Or it is also Guidance, if he plans to cross at an angle to the railroad for better clearance.

Control is when he tries to execute the above slant-angle crossing in a manner to avoid a rollover, or does a U-turn to find another road.

 

[tinlizzy]

Edit - wrong thread!

 

[ERD50]

Another thing about lidars is that they can detect a deer standing off the road at night, and know how big it is. That might help Ron with his deer collision avoidance problem. Well, a camera can too, but it sees only the eyes of the deer in the dark, and only when it is looking at you.

However, that is an enhancement. Level 4 or 5 says that you have to match human's ability first. The other stuff is extra but not required.

True, but I think we should keep focused on the goal - reducing accidents/injury/death. If the SDC can detect things we would miss, that's a positive, and could help bring down the averages. We shouldn't only think in terms of replicating what humans do. That kind of thinking is too narrow, like designing robots with only two hands, when three or more can be more effective.

Along those lines, I feel that while the pursuit of Level 4-5 is interesting and will have benefits along the way, I'm a more pragmatic thinker (the good in the near term over the perfect future). I fall back on the 80/20 Pareto Principle, we should be looking at the biggest problems (some of that has been reported in this thread), and looking at what current and near term technology can do to help in those areas.

Just to illustrate with numbers, if a 30% problem is difficult and expensive to detect/respond so is only 5% effective, but the 10% problem is relatively easy and 90% effective, there is more gain in that low hanging fruit - 9% solved easily, versus 1.5% solved at high cost.

Take impaired driving (not just drunk, could be other substances, a medical condition, sleep deprived, etc). The driver-awareness monitoring I've been talking about could be relatively easy to implement today. It was mentioned that users cheat the breathalyzer tests with a sober accomplice. But a monitor in the car, checking the driver's responses could detect impairment, and force the car to slow and stop at some point, with warning lights flashing. And could call for help.

Far easier to detect that the driver is weaving, that responses are slow, that they are drifting off, than all these other scenarios that a L4-5 must cope with. The car could even do an effective 'sobriety test' before allowing you to pull away. Tap the wheel with your left hand 3 times, the right 4 times, blink 7 times, what is 3 + 5?, recite "Peter Piper", (voice detection), .... and you would have to be behind the wheel - no way to cheat that with a camera and seat & seat belt sensors.

-ERD50

 

[Music Lover]

Far easier to detect that the driver is weaving, that responses are slow, that they are drifting off, than all these other scenarios that a L4-5 must cope with. The car could even do an effective 'sobriety test' before allowing you to pull away. Tap the wheel with your left hand 3 times, the right 4 times, blink 7 times, what is 3 + 5?, recite "Peter Piper", (voice detection), .... and you would have to be behind the wheel - no way to cheat that with a camera and seat & seat belt sensors.

-ERD50

How many sober people would that test or a similar one disqualify?

 

[NW-Bound]

True, but I think we should keep focused on the goal - reducing accidents/injury/death. If the SDC can detect things we would miss, that's a positive, and could help bring down the averages. We shouldn't only think in terms of replicating what humans do. That kind of thinking is too narrow, like designing robots with only two hands, when three or more can be more effective...

Not disagreeing with you at all. But I am an engineer much more than a sociologist. When looking at the driving cars, I ponder how I would do something, how these companies solve these technical problems. That's foremost in my mind.

I researched and quoted some accident statistics because people stress the accident-free benefits. In my mind I think about the self-driving car as providing comfort and relief from the tedium of driving. Hence, I am still interested and would pay more for it (within reason), even if idiots driving normal cars can still hit me.

 

[Midpack]

To partly look at the cost of a camera look at how the cost of a TV camera has fallen, given that a large number of cell phones will act as a tv camera (at about the old analog tv resolution).

True. But a wealthy friend of mine bought one of the first flat screen TV's before anyone else - it cost him $20,000. You can buy a decent sized 4K TV today for under $1,000. SDC's costs will fall dramatically, but it may take quite a while, and they may start out within $ reach of only the very wealthy.

 

[Midpack]

Very interesting IMO but lengthy read for those so inclined. They walk through each of the "misconceptions." FWIW
Top misconceptions of autonomous cars and self-driving vehicles | Driverless car market watch

Self-driving cars are a rapidly evolving technology which only a few years ago was still considered science fiction. In such a dynamic context, quick intuitions can be very misleading and misconceptions about the technology, its impact, and the nature of the innovation process abound. In the following we address some of the most widely held misconceptions about autonomous vehicles:

Top misconceptions

1. Driver assistance systems will evolve gradually into fully autonomous cars
2. The first models of fully autonomous cars will be targeted to the consumer and will be available for purchase
3. It will take decades until most of the vehicles on the road are capable of autonomous driving
4. Self-driving cars are controlled by classical computer algorithms (if-then rules)
5. Public demonstrations of self-driving cars provide an indication of their capabilities
6. Self-driving cars need to make the right ethical judgements
7. To convince us that they are safe, self-driving cars must drive hundreds of millions of miles

 

[ERD50]

How many sober people would that test or a similar one disqualify?

If they can't pass the test, the car should shut down, regardless the cause. Impaired is impaired, and impairment results in more accidents.

-ERD50

 

[Music Lover]

If they can't pass the test, the car should shut down, regardless the cause. Impaired is impaired, and that causes more accidents.

-ERD50

I agree. Yet, even suggesting re-testing for drivers is met with an immediate backlash and is called age discrimination. Forcing drivers to pass a "test" every time they get behind the wheel is much more invasive than simply re-testing every few years. I have nothing against it, but don't see it changing any time soon.

 

[ERD50]

I agree. Yet, even suggesting re-testing for drivers is met with an immediate backlash and is called age discrimination. Forcing drivers to pass a "test" every time they get behind the wheel is much more invasive than simply re-testing every few years. I have nothing against it, but don't see it changing any time soon.

I'm not saying it would be accepted, just pointing out what could be possible and effective.

It wouldn't need to check you every time, unless maybe the courts said you were under that restriction. It could do it as a double check - if it was sensing you weaving (maybe there was a reason, stuff on the road it couldn't sense, whatever), it might give you a chance to prove you are OK? Or maybe you were drowsy, pulled over and took a nap - now prove you are alert before it starts up again? A conditional test.

I'd like it. If I'm impaired, or anyone who gets behind the wheel of my car, I should not be driving, and sometimes that impairment means you don't fully comprehend how impaired you are. Although rare, people do have strokes or other medical emergencies while driving. It wouldn't take super-advanced tech to flash emergency lights, slow the car down and have it pull over, stop, and call for help.

-ERD50