Autonomous Cars - Are They Realistic?
 

We are getting waaaay out in front of our skis, in my opinion.
There is a LOT to discuss:

First point is that by far the best autonomous cars are ELECTRIC. So, if we see a lot of autonomous cars - most of the cars on the road will be electric.

Second point is that before we see Level 5 (FULL autonomous) cars, we will see MOST cars with Level 2, or Level 3 STANDARD.

Third point: autonomous cars need to be "learning" - and therefore they have to A) "talk" to each other, and B) they need to share ONE DATABASE.

Fourth point: we can barely get this to work on LIMITED ACCESS highways, that have bright painted lines. Making the jump to full autonomy on ALL KINDS of roads / highways, in ALL conditions - is a QUANTUM increase in complexity.

To illustrate this fourth point: take a gaming computer - these are right at the cutting edge of performance - and ALL it has to do is render a PRE-MADE 3D MODEL. An autonomous car has to take the inputs from 8-12+ different sensors - and it has to "build" a moving 3D model, AND it has to do all the vectors of their motion - a FOURTH DIMENSION - in the blink of an eye.

AND it has to PRIORITIZE ALL THE MOVING OBJECTS that it "sees".

AND it has to "decide" the best route, AND the second best route, AND MAYBE even the third best option - CONSTANTLY.

Fifth point - it probably needs to have a FULLY REDUNDANT set of backup hardware - AND / OR it needs a really robust "algorithm".

Sixth point: WHO IS LIABLE FOR WHAT THE CAR DOES?

Seventh point: in what world will our government have a consistent and effect set of regulations? The insurance companies have to be onboard, too.

****************

To sum up - we first will have to have a majority of cars on the road to be electric, and we will have to have a majority of cars / vehicles to have automatic safety function, that backup the driver - as STANDARD equipment.

Then, and only then - we will need a shared open source database of all / most cars, so that each vehicle gets as much advantage as possible - of driving in any / all conditions as close to perfect as possible.

The goal is - obviously - much better vehicle safety. Lets be realistic about how / when this can happen.

Predetermined routes is an important step. And the driver's role is wholly transformed - they go from being 100% responsible, to being a passenger.

And how do we enter the desired route? Have you gotten 100% accurate GPS routes, all the time?

In their moving cars, people WILL:

Eat

Sleep

Have sex

Watch a movie

Do work

Many other things we can't imagine, too.

Good points.

Processors can analyze data very quickly though. There is also already a wifi standard for autonomous vehicles coupled with the fact that 5g cell service will make data transfer much faster, among other advances.

3d gaming isn't really "premade models", that's the whole point of gaming, it's not a movie. It is completely variable each independent input doesn't necessarily happen in the same sequence or at the same times. Modern PC's can run some of the most advanced games at over 60 frames per second which clearly shows the amount of data capable of being processed in extremely short periods of time. I'm not phased at all about the capability of processors for autonomous vehicles.

With more accurate GPS and known good routes "learned" in real or near-real time, I don't see way-finding to be an issue.

I'm a bit more optimistic than you are about AV I suppose. Way too much money on the table to not happen in the next 5 to 10 years.

AV doesn't have anything to do with EVs. They already have ICE cars that auto-brake, parallel park, etc. I see no reason why AV and EV are related.

We'll probably never have shared databases (DBs). In the healthcare industry, there isn't universal databases. Every time you go to a new clinic (in the US), a different DB is utilized and your data gets replicated there. Heck, even within the same clinic, they might have multiple DBs depending on what treatments you need.

My point is, the DBs are incredibly valuable, and companies aren't going to collaborate to share their info, or get their info to work with some universal DB. Perhaps vehicles will communicate with each other, but it will be through an interface that keeps their DBs private and only offer up the relevant data to share.

I've said the smartest way to implement autonomous driving is to whitelist known routes that are simple for autonomous vehicles to drive on, such as interstates with few turns and exits. You could also blacklist routes known to be difficult for AVs. Areas where human input was required frequently suggests an unsafe place for AV to operate, and that could be blacklisted automatically. Places where human intervention is infrequent suggests safer areas to engage AV, and those could be whitelisted automatically.

There was a guy who ran into a guardrail in his Tesla using Autopilot, and it had been mentioned that the area had confounded Autopilot before. Had the user intervention been properly logged and flagged in some DB as a tricky spot, it could have blacklisted the route and refused to be in autopilot mode for that stretch.

In one of my classes we had a guest speaker talk about autonomous cars, he said at first there would probably be autonomous only sections of interstate during certain hours of the night. This would slowly expand until at some point in the distant future it would be pretty much all auto, all the time.

This was a quick summary of a few points he made.

.

5G and AI (artificial intelligence) will make it possible.

All you have to do is leave fate (yours) at the door...

>

Fully autonomous cars have another purpous which is to drive the car with no one in it. Or with a rotating group of occupants getting in and out like a custom bus service with the route determined by smart phones calling for a ride. Or to drive itself to the closest available charging station. Private car ownership will start to come to an end.

I would say diminish, I doubt it will ever end.

I am curious what that means for the auto industry. While total vehicles may be reduced, riders will demand/prefer the newer vehicles so taxi/transit companies may be pressed to purchase more often. That brings about a good question around the used market....

Complete disruption. Sales numbers will fall drastically when autonomous ride sharing makes serious cuts into private ownership since the idea is that cars will seldom be parked. Parking lots, insurance, gas stations, all of this will suffer. Uber, Lyft, Waymo, ect are only slightly ahead of their opportunity of profitability right now.

I'd like to see the auto industry consolidate a bit. There's no reason to have thousands of different models of passenger vehicles to choose from. Seems like 20 would be a better number, or perhaps 50, but not thousands.

This will replace buses, maybe. Predetermined routes are relatively easy, remember. Truly FULL autonomy is way more difficult than predetermined routes.

The problem with that is that 5g's limited range makes it only practical for congested urban areas. 5g's range is (optimistically) about 1500 ft, and its short wavelengths are easily blocked: https://www.computerworld.com/articl...y-from-5g.html 3g & 4g range is anywhere from 3-30 miles, perhaps more if you're on a mountain with a good line of sight - and there are still lots of places in the US that have roads but no cell service.

So the only way to make that workable is for every car to in effect be its own cell tower, adding yet more cost & complexity.

We're on an 16' wide gravel road. Had a delivery truck call me today and ask. "I am on your road but there is no bridge?" Just a ford with water crossing over it.

He was following gps. I directed him several miles around the the creek crossing. There may be a lot easier things to automate than trying to make cars do what they are not suited for. The human brain is much more powerful and versatile than any computer. The experts are wrong.

5G won't be restricted to high frequency, which is why they killed off 2g and have already set drop dead dates for 3g.

Also, beam forming technology from wifi will likley be implemented and will improve transfer rates and range.

Not from what I've read, though I'm certainly no expert.

Adding further cost & complexity :-)

Longer wavelengths (low frequency) transmit further, and through more objects, but support lower data rates. Shorter wavelengths (high frequency) transmit shorter distances and have less object penetration.

Long range isn't desirable because only 1 conversation can take place on a frequency at a time. That's the idea behind "cellular" towers. They have short range so that other conversations can be made on the same frequency not too far away.

The pertinent data a vehicle needs to communicate to others is small. It needs to transmit position, and intended course and speed; a vector. Other stuff too, but the data is still minimal.

Communication with other vehicles is a relatively minor technical consideration compared with the vastly difficult task of dealing with a dynamic environment and accurately interpreting sensor inputs into actionable or ignorable responses.

Realistically they are still taking in data from human drivers to develop their autonomy. Unless you are in an area like the bay area most of the devices will struggle to drive similar to what a 14 year old learning to drive would do. But thats not to say i'm a naysayer here. The devices right now are really fantastic at level two autonomy, bumper to bumper traffic, highway cruising, and interstate bombing.
Comma Ai to me is the self driving car for the every man.
Check them out on youtube!

If interstate travel was all AI ever got extremely safe at controlling, I'd be happy with that. Take off at 8pm when there is no traffic on a long road trip, sleep while the car puts in the miles, and wake up only to refuel or when we near the destination.

Your seven points are good ones, but for me it is all mooted for the same reason as cell phones and 5G -- the potential for abuse of the citizenry by corporate entities is too great.

Not that that has stopped anyone else so far.

The proper way to introduce [the removal of your] autonomy would be as a back-seat driver. Letting you drive the car until you get in over your head and then stepping in at the last minute. Plus the lane keeping, etc.

Exactly,

The only reason I even want one is just to be able to have a job an hour away and not have to actually drive the interstate there and back everyday and fight falling asleep.

The problem, though, is that the shorter the range, the more towers you need for complete coverage. Cut the range in half, and you need at least four times as many towers - probably more when you take terrain & other obstacles into account. And then you need to run data cables and power lines to those towers... Not cheap, and probably not cost-effective outside of congested urban areas and such.

On 90/94 yesterday in rush hour traffic, a Model 3 got behind me. I noticed the driver didn't have his hands on the wheel, was looking out the window or at his cell phone most of the time, so he must have been using Autopilot. The Model 3 maintained a car length distance or so behind me, and if I moved up just a few feet and stopped it didn't budge. I felt much more comfortable with the computer driving behind me than the Silverado HD later who insisted on getting as close to my bumper as possible every time traffic stopped.

You can actually accomplish a lot of this yourself. I've seen a guy control his epas through an arduino board through the comma ai unit with its camera's, sensors, and gps. He made a electric motor mount to control his hydraulic steering shaft and low and behold it did it. it was funny hearing the servo react to steering inputs though, but nonetheless. This isn't as necessary now with newer cars though as most already have EPAS and just lack the radar plate for distance control. Which brings me to the next point where they guy rigged up a radar plate to some relays which controlled the cruise control to maintain a follow distance of 1.8 seconds behind the car in front, and target speed was the speed limit if no obstacles were detected.
But the city aspect is the monster. We are actually working on it as a research project right now. Basically a neural network, and taking data from the cameras to detect obstactles irl as fast and accurately as possible. We'll see what happens... haha
I'm a mech e student with a serious lack of I T coding experience.

I'll try and tag the videos. One of the guys actually reached out to me on a forum after we had talk about his video. Crazy how small the world is now with the internet. Also, what is crazy is a year ago when I looked at all of this there wasn't many results or hits, but now youtube is flooded with videos on this topic and diy'ers.

Really? Of course divided highways in rush hour traffic are easy, but I'd think that city streets would be much easier than rural roads - and that's before you take weather and critters into account.

I find city streets to lack basic road markings, hide stop signs behind trees or dumpsters, and generally be chaotic. In Las Vegas, there was a continuous stream of people crossing a driveway entrance, and I was blocking the road waiting to make my turn. With a police cruiser watching, I slowly crept through the crowd, parting the people around the vehicle. AI would never decide to drive through a crowd of people.

Good Book...
 

For self driving cars (including current airplanes) read the book Hello World: Being Human in the Age of Algorithms by Hannah Fry (2018). She has some interesting insights for the long term use of self driving cars as well as auto pilots in airplanes. Lots more stuff on algorithms currently in use that was also very interesting.

https://commaai.blogspot.com/

Just as they're working on robots to play futball, someone needs to make an adversarial generative AI to rock-crawl boulders.

Automated machines are useful in situations that are repetitive in nature. Many people will benefit from a self driving vehicles. The caution is that you should never let a machine do your thinking for you.

Even in the aviation, not every airfield has support for those nearly-autonomous "flight directors" already widespread in major airliners.

This sounds fairly realistic, unlike much of the hype that's been pushed of late.

I remember my son telling me "this is probably the last gasoline powered car I will ever buy; the next time I'm ready to purchase there will be so many electric choices." That was 10 years ago.

Yeah its one of those things that they have to wait to map out the rural roads, and until then it will be poor. But things are typically calmer and there are less factors to code for. This is why highway super cruise is such an easy feature to produce, also the grid lock mode of follow the car in front of you. But, In the city you have the chaos of city drivers being poor at following the rules, the pedestrians are cut throat as well, and there is a lot of environmental factors that are ever changing. It will get done, but thats where the hobbiest level probably stops. Its easy for a hobbiest to code out super cruise control for the highway, but trying to navigate the city takes a lot of miles to be logged.

I was talking with my neighbor and we agreed that the first conquest of the autonomous vehicle will probably be the autonomous race car. Two examples:

https://i.kinja-img.com/gawker-media...jempw1gl1t.jpg
https://jalopnik.com/why-driverless-...t-s-1784844768

https://www.youtube.com/watch?v=bX4TXWO7dA0

Here's an interesting feature of this thread: you're responding to a bot. davis54 is the result of running M_a_t_t's post (#4) through a madlibs-style program and spitting it back out. Coming to an American interstate near you!

Very interesting indeed. It didn't trigger my botdar at all, though it's clearly a bot. I think I just passed it off as being the same post I had read previously.

freebeard gave a thanks to the bot, perhaps as a gesture of being very sophisticated.

I've still never understood the point of bots (shills) on a forum though, other than as an academic test.

I did? Not in this thread.

https://ecomodder.com/forum/attachme...1&d=1545949220
https://ecomodder.com/forum/showthre...tml#post587183

Are you saying this didn't go to auction this year?

Its subsequent posts have a signature link to some moon lamp site or something. But who actually clicks on those? I can't imagine this actually being profitable for whoever's behind it.

I'm going to have to stop caring about this. Currently the last post I see in that thread is https://ecomodder.com/forum/showthre...tml#post587191. And a search for davis54 returns three results not including a Profile page.

Not so much as a 'User Removed'.

What's worrisome is that one of my posts is gone. Maybe it's the bug where vBulletin won't display the last pages of a thread, but it says I'm on the last page.

Will autonomous cars share this 'feature' from web pages?

I sent a mod alert to axe that account. Looks like they did, along with all of the posts, so my permalink no longer points to the correct message.

The message you thanked said something about attending some show. Towards the top of the thread, another member had mentioned attending that show.

... I'm going to have to rewatch "Her". We're probably not too many years out before AI and humans will be indistinguishable, especially in written form. My assumption when I see odd word usage is that the poster is not a native English speaker. AllDark uses the word "polemic" frequently, yet I've never seen or heard it used before then. AllDark is a bot!:p

I'm sure he means "controversial". It's so easy for someone who speaks Portuguese to make a lot of mistakes and get lost in the translations while not being so used to write in English.

I though he meant a Simple Desultory Phillipic.