Self Driving Cars?

First you crawl, then you walk, before you run.

My 1-year old grandniece has been walking. Still wobbly, but improving everyday. Then she will run, and eventually ride a bike. She may get to enjoy a level 4 car.
 
The lanes I saw the other night in the construction zone were a mess. No mapping will keep up with it, especially considering the wind turned around a few signs.

Construction zones may have to have set up a "do not enter" zone for some of these vehicles if they are relying on maps.

I think ultimately the mapping and exact positioning will be the way to go. But there may have to be times of exclusion enforced.
Did you read this quote below that was in my post and take it into consideration?

The data captured is not just lanes. It is meta data about locations of where signs are read, "permanent" structures are located. So if this data is being sent to the "cloud" all the time then when 1+ cars as part of this fleet learning are transmitting changes (signs aren't be read, lane reroute, etc), it could trigger things like autopilot not being able to be engaged in that area or warnings displayed to the user or the 'hands on wheel' messages being much more aggressive in the case of Tesla.

Tesla stands apart from the others is the way it's acquiring this data: through drivers. Every Tesla Model S, with Autopilot or not, is connected from the cloud; the company is constantly connecting data from each of its cars. Tesla is using the data it has, and will continue to collect, to develop its maps.

Elon Musk called this a "fleet learning network" where all its cars contribute to a shared database. "When one car learns something, all learn," said Musk.
 
Last edited:
Level 4 is the interim goal, primarily due to regulations as Krafcik has noted. Not sure why that's important if level 5 is the ultimate goal. I haven't seen anyone suggest the evolution won't follow,a stepwise progression, that's just common sense, especially for something this complex.

The Firefly car is a concept demo car. It shows people the future vision of what a level 5 car will do for them, but itself is not one. When they show driving a blind man around, the car followed a pre-planned route. I recall they said so too. And then, as safety conscious as Google is, I am sure they have followers with a remote kill switch if something goes wrong. What if anything in the electronics control channel fails? What if a leaf falls down and sticks on the lidar, blinding it, etc...? A demo suddenly becomes a PR disaster. "Google car almost kills blind man", says the paper. Google is way too smart for that, besides being very cautious. I like them a lot.

PS. The Firefly demonstrates that Google can do one hell of a level-2 or 3 car long ago, and far better than what is out there. They just don't want to do it for many reasons that they keep reiterating.

Caveat: Never a Google stock owner. In fact, I dislike them for some other stuff.
 
Last edited:
Reading some of the above posts, I think people understand the definition of autonomy levels differently than what SAE intends. It is more than the capability of the car under the best conditions.

For example, a car with a single lidar may work very well, see pedestrians, bicyclists and all that stuff. But what if it encounters sudden rain? A level 4 will request the driver to take over, but if he doesn't, it has to know to pull over, and stop. How does it do that without less capable backup sensors like radar and vision cameras? Most likely it will drive on at a lower speed, using lesser sensors, just like people driving more slowly in the rain and snow.

That's what the new Waymo sensor suite provides. They use the lidars foremost, and can blend in other sensors as needed. By crosschecking between sensors, they can detect failures more easily. The lidars have mechanical spinning parts. They may not have the same lifetime as solid-state radars and vision cameras. And lidars will be blinded by a piece of paper flown up from the road. All these mode reversion logic and control strategy take time to develop and to test.
 
Last edited:
Did you read this quote below that was in my post and take it into consideration?

The data captured is not just lanes. It is meta data about locations of where signs are read, "permanent" structures are located. So if this data is being sent to the "cloud" all the time then when 1+ cars as part of this fleet learning are transmitting changes (signs aren't be read, lane reroute, etc), it could trigger things like autopilot not being able to be engaged in that area or warnings displayed to the user or the 'hands on wheel' messages being much more aggressive in the case of Tesla.

I'm not sure what the first car does when it encounters a sign blown into the middle of the lane that was just repainted. I guess it stops and waits for a human? ("Hands on the wheel"?) Of course, for level 4, take over and go round. I guess that may work. Not sure about level 5.

The cloud is great until it isn't. Cloud currently relies on cell or wifi coverage. There are a lot of uncovered roads. The cloud has issues from time to time. (See the recent AWS outage.)

These are not unsolvable issues. They are bumps on the road.
 
I notice that Waymo has a lidar on the nose looking down. They need it to come close to the curb. And there's a lidar in the rear, but no camera. The lidar is superior to detect a child or other objects, so they need no rear camera, and that makes sense. They use no ultrasonic sensors, and use side lidars besides side radars. These guys like lidars a lot.
 
Last edited:
Are you referring to the ones indoors? Or the guy partially obscured by the street-sweeper thingy? I would hope that the car would avoid hitting those things for reasons other than correctly identifying them as pedestrians.

The latter case. Seeing a man standing by himself is different than a man carrying a billboard obstructing his lower part of the body for example. A squatting 5-year old may be a lot tougher than a full-grown adult.

When I say there are things that the human mind can do a lot better than machines, some disagreed. The problem is that the human mind sometimes gets a brain-fart, does not pay attention, or sometimes just doesn't care.
 
Last edited:
NW-Bound/ERD50, I appreciated your thoughtful review of my post. I did see them last night but wanted to think about them some more to consider a more thought out response. Sorry for a little repetition below.

Here's an overall thought. Just like you guys feel like when using regular cruise control that you can override it at anytime and it makes you comfortable and more trusting ... I feel that exact same way when using AutoPilot (AutoSteer + TrafficAware (adaptive) Cruise Control). I can manually brake or move the steering wheel and the car immediately gives me control back. It just adjust my lane centering and steering 100s of times per second and I can monitor it mentally (sense) and physically with my hands lightly on the wheel and the immediate feedback that gives my brain.

I like the cruise control analogy. With normal/regular (not speed adaptive) cruise control your foot is relaxed and you can move around. If you are on a long open stretch of highway with few cars you can relax more because you don't have to worry about slowing down for cars ahead of you going slower and cars behind you tailgating. However, if you are on a busy highway normal cruise can be annoying you have to pay a lot of attention because you can run into the back of a car.

Steps of improvement and why: Faith, confidence, and comfort level after a few hours of use happens quickly, AND you can override at anytime in any of these (i.e. you still have the sense you are in control! - key point) :
1) Manual driving vs normal cruise controls : Great on non-busy highway with few cars occasionally ahead or behind
2) Normal cruise control vs adaptive cruise control (ACC) : Even great on busy highway including stop and go. Super nice not to have to constantly adjust your speed with the +/- or turning it off/on.
3) Adaptive cruise control (ACC) vs AutoPilot (AutoSteer lane centering & ACC) : If you are conscientiously manually staying in the center of the lane it takes a lot of micro adjustments ... with AutoPilot it is doing that 100s times per second and literally feels like someone is driving as you lightly hold the steering wheel AND you feel/sense it working.

So, I think what is going on is that by being attentive, a conscientious Tesla owner experiences the scenarios that may cause trouble, such as construction zones, mismarked lanes, etc... and gets prepared the next time he encounters the same, and it is no longer a big deal.
Here's the thing ... I don't think it is that much different for anyone driving manually. We humans adjust to situation dynamically and conscientiously. If you were driving down very low traffic highway your alertness and constant input to the car would be very different than if you were in a construction zone on a highway with concrete dividers very close to you and lanes being rerouted. Similar in urban driving at 1pm after lunch hours over vs 5pm when people are rushing home after work. Different defensive driving and hectic environments. Or another situation is on a short roadtrip of maybe an hour on a highway vs long road trip of multiple long 8+ hour days. You get lulled into driving after few hours and are less alert.

Now compare this to AutoPilot (AutoSteer lane centering and Traffic Aware (Adaptive) Cruise control) which is *always* alert. How many times per second is it correcting the steering to stay centered in the lane. 100s for sure. I am just lightly holding on to the steering wheel and it is magically staying centered. It is like someone is driving for me even tho my hands are on the wheel. HTH (see my response below too to round this out)

("hands on the steering wheel AT ALL TIMEs *and* eyes on the road just like you were manually driving,"), how is the drive any less fatiguing? I think it was discussed earlier, if I'm constantly wondering if the system is going to need assistance, couldn't that be more fatiguing? It would seem to me you'd wait a little before deciding it wasn't acting properly, so then you have to react even faster, due to that lag. That just doesn't sound relaxing to me, but I haven't actually experienced it either.
The reality is that it is not just like manual driving as the car is maintaining and processing the distance to not only the car just ahead of me but the one ahead of it 100s of times per second. It is also adjusting the steering 100s of times per second to stay centered in the lane. My hands are on the wheel and I'm paying reasonable attention but I am more relaxed and can look at scenery left or right or look/adjust the nav system or radio for a *several* seconds vs a *couple* seconds because I'm not as afraid of what may catch me by surprise (car merging, car stopping ahead of me, etc). (see my response above too to round this out)
 
Don't know how it is in the US, but what boggles my mind is that today you generally only have to demonstrate capable driving once in your lifetime. After that, no-one will check whether you are a poor driver and major disaster waiting to happen. One is presumed capable. Insane.

Case in point: My grandmother has never driven a car at 86 years old, yet has a drivers license and is legally allowed to take a 4000+ lbs death machine on a public road...

It is the same or worse in the US. The wife of my brother-in-law does not drive, yet she has a license. I asked, and learned that the day after she got her license, she panicked on the road (don't know the details), got out from the driver's seat and has not driven since, for many years!

Now, it is good that she knows her limit, but for the life of me I cannot understand how she got the license.

France has them mostly on N-roads (national roads) and secondary routes. Germany just opened up their first section in 2016. Luxembourg none as far as I know (it's so small anyway :D). Italy (Tutor) everywhere on the major highways. Switzerland in some locations like tunnels. Belgium is expanding quickly. Started with the tunnels around Brussels and highway around Gent.

If you want a more detailed overview you might want to try and toss this in Google Translate - also references therein. Happy to translate (from Dutch or French) any specifics.
https://www.wegenwiki.nl/Trajectcontrole

There is one aspect potentially relevant for you: breaking the speed limit is one thing, getting the fine is another ... information sharing is still far from optimal across the EU.

If you break the limit in Switzerland, you'll get chased towards the end of the earth for example. But if you have an Italian car in Belgium you are pretty much immune.
Thanks for info. I can handle a bit of French (in reading), but Dutch is Greek to me.
 
Last edited:
Seeing a man standing by himself is different than a man carrying a billboard obstructing his lower part of the body for example. A squatting 5-year old may be a lot tougher than a full-grown adult.

Agreed. And as I said, I would hope that a self-driving car would make every effort to avoid hitting any of them, or anything else that could damage the car or cause destruction, whether it assesses it as a human, a tree, another car, a runaway shopping cart, or anything else it detects passing in front of it. Whether or not it successfully identifies it as a human is of little material consequence, except for the (hopefully) extremely rare scenario where an unavoidable crash is imminent, and the car must decide which obstruction represents the best candidate to minimize damage. Should it choose to hit the human pedestrian, or the cinder block wall? Should it hit the elderly, infirm human, or the puppy? Such situations will happen, but hopefully extremely rarely.

When I say there are things that the human mind can do a lot better than machines, some disagreed.

Not this guy! :) As a computer programmer, I'm well aware of the limitations of such devices. A couple of amusing expressions in the field of computer science come to mind, such as:


"Things that are easy for a human are difficult for a computer, while things that are easy for a computer are difficult for a human."


I'd cite some obvious examples, like catching a ball, riding a bike, or recognizing people as things that humans take for granted, but are incredibly difficult to train a computer to do. Conversely, a computer can perform perfect welds on an automotive frame, in total darkness, 24/7, or instantly tell you whether or not 7,893,097,171 is a prime number.

Another one I always liked is, "Asking whether or not a computer can think is like asking whether or not a submarine can swim."
 
Last edited:
Eroscott, I understand what you wrote above about having the car doing the work for you is very relaxing, with you only loosely "supervising" it. The problem nagging in my mind is when the car misbehaves, what would the human reaction be?

At the risk of being called some name, I cannot help bringing up this aviation analogy. A real tragic accident happened shortly after take-off, caused by a mechanical failure. The big jet rolled and crashed, killing everybody. One of the questions was whether the pilots could have handled it.

So, they tried to duplicate the failure in a motion-base simulator, using many pilots. As I recall, some were able to correct for it, and saved the plane. But then people say, when you put pilots on a simulator of the same aircraft after the fact, they all knew what you were up to, so were prepared for it. If it were a true surprise, it would be a lot worse. The question was not truly answered.

So, these "hiccups" of the car system can be handled by yourself very well, but would my 84-year old mother react in time to save herself?

Is anybody aware of any study of this human takeover problem by simulation, to see how the average Joe and also Jane Sr. react?
 
Last edited:
Not this guy! :) As a computer programmer, I'm well aware of the limitations of such devices...

I wrote software too, real-time embedded code that went in those things that flew (not always have humans in them), as well as desktop analytical software for my own use.

So, it bothers me when people say computers can make decisions 1000 times per second. Well, it does not matter if it is wrong all those 1000 times. Or that the decision it makes is based on faulty sensor readings.

But then, I am not saying anything engineers and programmers do not already know.
 
When I say there are things that the human mind can do a lot better than machines, some disagreed. The problem is that mind sometimes gets a brain-fart, does not pay attention, or sometimes just doesn't care.
I didn't see anyone categorically dispute the human mind has advantages over computers/software. However, computers/software also have advantages over the human mind - as your second sentence touches on. Sensors can "see" more, way more with connected cars & off vehicle sensors, and much faster than humans. Software can make decisions much faster than humans, but the underlying knowledge base and AI required is substantial. Computers/sensors aren't distracted or tired. Self-driving cars will never be perfect - whether self-driving cars can clearly net outperform humans remains to be seen, but there's already a lot of evidence machines may prevail. Should be settled within 30 years regarding driving.
 
Last edited:
Turkish Air flight 1951 crashed short of Amsterdam because a faulty radio altimeter caused the autopilot to reduce the throttles under the mistaken belief that the plane was already on the ground...
Forgot to comment on this. Bad design!

If radar altimeter is used for such a function, it should have been dual. For autoland, you need dual radar altimeters for the autopilot anyway, which is more critical than the autothrottle. The system I worked on used something much simpler: wheel spin-up sensors and gear strut compression switches. The chance of them failing simultaneously is null. Plus, if they indicate "gear compressed" and "wheel spinning" while in-flight before landing, the computer knows that they are bad, and inhibits some modes.

So, the risk of pulling back the throttle at the wrong time is zilch. And the risk of not pulling back after touch-down is very small, and also not as critical as pilots can pull it back. Actually, they may operate thrust reversers anyway.
 
Last edited:
I didn't see anyone categorically dispute the human mind has advantages over computers/software. However, computers/software also have advantages over the human mind - as your second sentence touches on. Sensors can "see" more, way more with connected cars & off vehicle sensors, and much faster than humans. Software can make decisions much faster than humans, but the underlying knowledge base and AI required is substantial. Computers/sensors aren't distracted or tired. Self-driving cars will never be perfect - whether self-driving cars can clearly net outperform humans remains to be seen, but there's already a lot of evidence machines may prevail. Should be settled within 30 years regarding driving.

I don't think we disagree at all. But laymen do not know.
 
I don't think we disagree at all. But laymen do not know.
I try not to assume folks here are "laymen" - most of them are probably smarter than I am.
 
Not all thread readers post. Many readers are true laymen.

One can go to youtube videos, and read the comments to see so many misconceptions tossed about.
 
Eroscott, I understand what you wrote above about having the car doing the work for you is very relaxing, with you only loosely "supervising" it. The problem nagging in my mind is when the car misbehaves, what would the human reaction be?
If were relaxed driving and you've had to adjust quickly from sudden wind gust or from a semi/RV passing you the 1st time that day in traveling down a 2 lane road, then you have an idea of your reaction ability.

A good test would be to drive down a highway on normal cruise control and with you having a light grip on the steering wheel and focusing on staying in the center of the lane. Let your trusted passengers keep their hand on your steering wheel while you are doing this and within a 1-2 minute timeframe (enough for your to think you are in control), have them minorly adjust the steering wheel WITHOUT them giving you any warning. Your eyes will be watching the road straight ahead all this time so you will sense or feel the steering wheel move WITHOUT your input.

You will realize it is happening VERY VERY quickly and react!!!
 
NW-Bound/ERD50, I appreciated your thoughtful review of my post. I did see them last night but wanted to think about them some more to consider a more thought out response. Sorry for a little repetition below. ...

And thanks for your reply. I had to think about it, and I think I understand better now, and it dovetails in with NW-B's follow up, so...

Eroscott, I understand what you wrote above about having the car doing the work for you is very relaxing, with you only loosely "supervising" it. The problem nagging in my mind is when the car misbehaves, what would the human reaction be? ...

Yes, I was thinking that all along, but try from a little different perspective. Picture this scenario:

You're using the Tesla AP on a clear day, on a long, long stretch of highway, with clear, well defined lane markings. You have a clear view of the road ahead. In this circumstance, I can understand that the current capabilities of AP could do a great job of keeping you centered in your lane. There's not a lot of variation to deal with. And you have enough experience with it to feel confident of its capabilities in these conditions.

So just like with cruise control, you can relax a little. You know you can take your eyes off the road just a little longer, and the car will be fine. In fact, this might even enhance safety, because you could take an extra second or so to concentrate your vision on something ahead, w/o needing to use any mental energy keeping the car centered.

I guess that's kind of a re-hash of what eroscott described, but I think this is the key - this is a pretty static environment, one you feel confident in with AP. If you did see some construction ahead, or some extra traffic, then you would go on a higher level of alert. But I can see how you could be more relaxed from maybe many, many miles of routine driving. But if you had to second guess the AP most of the time, I don't think that would be relaxing at all.

Unfortunately, without more of the kinds of driver awareness monitoring I've been talking about, I think there are a lot of lesser drivers who would just allow themselves to mentally drift off, and just not even be ready for the more questionable sections, or be over-confident, and suddenly find themselves in trouble.

Just like regular cruise control - when traffic gets congested, I just turn it off, it's more trouble than it's worth.

-ERD50
 
They are everywhere in the Netherlands on highways. The cameras are at every onramp and offramp. Took a while and several thousands in fines to get used to :angel:. I like the fairness and clarity, but very much dislike the sometimes unreasonable limits.

A speed limit of below 50mph (80 kph) on an empty road with four lanes per direction, who came up with that? Especially when on a similar road a bit up ahead they increased speed limits up to 130kph (80 mph).

The German variable limits are the best solution: they get adapted based on traffic conditions. When the skies are clear and one is away from the city, no speed limit at all. While it can feel dangerous (if you drive on the left, you will have a Porsche flying past at 150mph pretty often) the road safety track record there is pretty great.
They could do it also in the toll by plate areas of Miami. In the Miami area you get a bill in the mail if you don't have a sunpass sticker and go on toll roads in the miami area. It would not be hard to extend this to calculate speed between the toll points.
 
2) Normal cruise control vs adaptive cruise control (ACC) : Even great on busy highway including stop and go. Super nice not to have to constantly adjust your speed with the +/- or turning it off/on.

Just like regular cruise control - when traffic gets congested, I just turn it off, it's more trouble than it's worth.
But I think this is important point. If you had adaptive (speed) cruise control you would LOVE it in congested traffic. It would be much easier than manual driving and controlling the throttle and brake. It would maintain a 2.5 second or whatever setting you have it at distance to the car ahead.

If you had an advanced adaptive (ala traffic aware) cruise control like Tesla that would actually allow you to come to a complete stop and then go again when traffic moved you would love it even more. Some systems make you press the throttle after stopping. (Tesla normally does not make you do that).
 
Thanks, eroscott for the post. If I interpret what you were saying, it's that when the car "loses lock", if you will, it drifts gently and not violently as I feared.

There's a big difference between the above and a sudden twitch, like someone jerking the steering wheel. I have more concern of a more violent upset, like when my RV was hit broadside by a gust of wind. I was able to handle that, else would not be posting. But it is very tiring and uncomfortable, even though they have road signs warning of gusty areas ahead for me to be prepared.

About advanced adaptive cruise control, you do not have to sell me on that even though I have not driven one. I know I would enjoy it (but I do not drive in congested areas often). In my mind, I know how something like that works, and have more confidence that the chance of it failing catastrophically being small.

PS. My RV is blown much worse by passing semi-trailers than any car I have owned. It has a much bigger side area (duh!).
 
Last edited:
Don't know how it is in the US, but what boggles my mind is that today you generally only have to demonstrate capable driving once in your lifetime. After that, no-one will check whether you are a poor driver and major disaster waiting to happen. One is presumed capable. Insane.

Case in point: My grandmother has never driven a car at 86 years old, yet has a drivers license and is legally allowed to take a 4000+ lbs death machine on a public road.

Sadly, driving a car within the legally enforced rules has limited bearing on actual driver capability. There are people routinely treating the sidewalk as extended lane capacity that never get any citation.

If somehow we are willing to use current tech advances to open that discussion and do something about it, we'll be very far ahead already.

Imagine if your level 2 or 3 car signals a central evaluation center that you just hit a curb, with associated pictures. Or that you are dangerously tailgating, or slaloming (drunk driving). I hate big brother as much as the next person, it would save many lives I think.

All state are different, but here in Texas when you get to 79 you have to show up to renew and they might retest... vision for sure...
 
I know there is a lot of talk about the various levels etc. and how great it will be when we get there....

But... right now the avg age of a vehicle is 11.5 years (so much for the people who talk about keeping one for 10.... you are not even getting average age!!!)...

So, say level 3 comes out... that does not mean all cars sold will be level 3... just a small number... it might take 10 to 15 years to get that down to the cheap cars... but there are millions of vehicles that are still OLD... with no assists at all... it will take an additional 12 years to get rid of half of those...

IMO, to fully (say 95% of vehicles) implement even level 3 might take 30 years...

My point is that there are still going to be people driving without aids and those people can crash into us even if we have a level 3... I have been hit at least 3 times just sitting still.... it does not matter what level I had I still would have been hit...
 
Back
Top Bottom