Self Driving Cars?

[NW-Bound]

You people need to be patient.
2040 - autonomous vehicles are here and work well
2060 - human drivers are illegal (in most cases)
2070 - mass transit systems in the USA collapse

We are patient. When the technology is here, we will know it.

On the other hand, there are people who think the technology is here already. Some have killed themselves with immature technology.

 

[Chuckanut]

Mr. Musk certainly is fast an loose with some of his predictions and expectations.

OTOH, we can buy functional Tesla electric automobiles today. Companies can put a satellite on the world's most powerful rocket for less than 1/2 the price of its less powerful competitor. We have a functional spacecraft that can take cargo into space and return it to Earth.

Mr. Musk routinely stretches the limits of what others think can be done. From what I have read about the man, failure is just another learning opportunity. If there was no Mr. Musk, my bet is that the latest advances in EV's would probably be the equivalent of a 5 year old Nissan Leaf, and access to space would be far more expensive than it is today.

But, I must ask, what has Waymo delivered to the consumer? Well, I can buy its LiDar sensors and make my own self driving car. I guess that's something to keep me busy in my spare time.

https://techcrunch.com/2019/03/06/waymo-to-start-selling-standalone-lidar-sensors/

I own no Tesla stock or car, and have no financial interest it his companies other than what may be in my index funds. I certainly have no crystal ball. But, I do recognize that the man in the arena is the one who gets dirty, gets bloodied, and, perhaps all to often fails.

Teddy Roosevelt:

“It is not the critic who counts; not the man who points out how the strong man stumbles, or where the doer of deeds could have done them better. The credit belongs to the man who is actually in the arena, whose face is marred by dust and sweat and blood; who strives valiantly; who errs, who comes short again and again, because there is no effort without error and shortcoming; but who does actually strive to do the deeds

 

[NW-Bound]

But, I must ask, what has Waymo delivered to the consumer? Well, I can buy its LiDar sensors and make my own self driving car. I guess that's something to keep me busy in my spare time...

Waymo (Google) has stressed time and time again that they will not release unsafe self-driving cars. They do not want to hurt people.

Is it altruism, or Google simply wants to protect themselves from lawsuits? No matter what the answer, when Waymo says they are ready, that is more believable than Musk. And that's what I have been waiting for.

I guess I can wait because I am not so desperate for this technology now. If I were incapacitated, or had no access to Uber, I guess I might take a chance with a half-baked product. And then, maybe not. I could just stay home until I got a ride.

 

[Midpack]

Waymo (Google) has stressed time and time again that they will not release unsafe self-driving cars. They do not want to hurt people.

Is it altruism, or Google simply wants to protect themselves from lawsuits? No matter what the answer, when Waymo says they are ready, that is more believable than Musk. And that's what I have been waiting for.

I guess I can wait because I am not so desperate for this technology now. If I were incapacitated, or had no access to Uber, I guess I might take a chance with a half-baked product. And then, maybe not. I could just stay home until I got a ride.

And it’s my understanding when they’re ready, Waymo wants to license their tech to carmakers, they aren’t interested in making cars themselves. IOW you’ll never be able to buy a Waymo model car. Makes sense, and consistent with much of Google’s revenue generation, getting paid well indirectly. If/when SDCs really take off, there will be some carmakers who aren’t technically ready, and licensing Waymo may be their only bridge to SDC (vs irrelevance in the auto industry) - many if not most of the smaller carmakers who don’t have huge R&D budgets or personnel (e.g. Subaru, they trail on most technology).

And for all we know, Waymo may be licensing some of their tech already?

 

[Blueskies123]

In the year 2525?

Based on how long it took the automobile to match the number of non-farm horses I would say that 15 years after fully autonomous cars are available there will be as many autonomous cars as gas-powered cars.

 

[pj.mask]

Traded my 3/4 ton for my buddies Dual Motor 3 yesterday. After a day of cruising in the Tesla, I am happy to say it's not for me.



...



Curb Weight: Dual Motor - 4,072 lb

Curb Weight: Diesel Duranamax - 7,620 lb

...

Question: what was the MSRP on your truck? Wondering if it should be compared to model 3 or model S/X

I know trucks can get pretty pricy. Mode X is ~6,700lbs

 

[ERD50]

In the year 2525?

That's optimistic based on the rest of that song:

In the year 5555
Your arms hangin' limp at your sides
Your legs got nothin' to do
Some machine's doin' that for you

5555!

-ERD50

 

[ERD50]

Based on how long it took the automobile to match the number of non-farm horses I would say that 15 years after fully autonomous cars are available there will be as many autonomous cars as gas-powered cars. ...

Two problems with that comparison is that we have millions of cars on the road, and their average life is ~ 15 years. And self-driving capability will be a progression over many years.

Note that on your graph, during those first 15 years, autos increased very quickly, horses barely declined. So it isn't a 1:1 replacement, which I think would mostly be the case with cars, though a true SDC might open up some new markets. But I'd expect it mostly to be current drivers buying the tech to replace their current car.

OTOH, I also won't rule out that by the time true, full SDC is available, that we won't even need/want it. Just like the horse breeders of 1880 didn't envision that people would not want a better horse in 30 years, they'd want something almost unknown to them at the time.

Brings to mine that old line about predictions...

-ERD50

 

[NW-Bound]

While I am very interested in the development of this technology for the technical aspects, I am not trying to make any prediction on any timeline. It is harder than trying to figure out where the market is going, and the latter impacts me more than SDC.

If self-driving cars happen in my lifetime, what's left of it, I will enjoy it. If not, I will drive myself until I no longer can, then I will take Uber. No matter how the technology progresses, I want to be sure I have the money to afford the available option.

 

[NW-Bound]

Yesterday, I ran across a Youtube video from Tesla made during an event called "Tesla Autonomy Investor Day". This was in April 2019, when Musk announced the use of the Model 3 with FSD ("Full Self Driving") for a taxi fleet. I have read about this announcement, plus the claim that Tesla3s are built for 1,000,000 miles to last as long as semi trucks.

Having spent some time in the past to watch past Waymo presentations, I also spent 2 hours to watch the above Tesla video. I scanned through the first 1 hour of demo car driving, and skipped to the later 2 hours of technical presentation of the hardware and the software.

While Tesla has made good progress with its autopilot, I believe it is highly unlikely that Musk will deliver his promise of a taxi fleet by next year.

One technical thing that I will note here that Musk still claims that lidar is not necessary for a self-driving car, and that humans can drive cars with just vision, and his cars have more cameras than humans have eyes. It is all very convincing, but people forget that computer AI is still no match for human's brain.

Bashers of lidar conveniently forget that Waymo has vision cameras too on top of multiple lidars and radar sensors. What Musk has is a subset of Waymo sensors. Waymo already used vision cameras to interpret traffic lights and signs long ago, something Tesla is still working on. Waymo's lidar is a thing on top of the vision cameras to prevent driving the car into semi trailers and red fire trucks, something Tesla cars have done several times.

It is of course much less expensive and more elegant to use just vision cameras, but nobody has demonstrated an AI system that is robust enough to eliminate the lidars. Musk thought he had enough computer power in the original Tesla 3, then found out that he needed more CPU power to do the job, hence had his own chip designed. This is definitely good progress, but the chance is high that he will find out that the new hardware is still not enough, and that he will need a deeper neuron network to solve the problem.

I believe Waymo knows that the current technology is not reliable enough without lidar. When AI is sufficiently powerful to eliminate the lidar, Waymo will do so. It is not that I am not impressed with AI technology at this point. It's that while good, it is still not good enough, nor reliable enough.

Here's an analogy. The aerospace technology that allows us to explore space is advanced and impressive. However, it does not mean that we can each have our own rocket or flying car anytime soon. The public tends to quickly make the jump, such as to expect stem cell research to allow us to grow new organs any time now, and to allow us to live as long as Methuselah. Nope. It's not that easy, nor that quick.

 

[NW-Bound]

I am reviving this thread because there's a lot of interest on the current state of this technology of self-driving on another thread about EVs. Yet, SDC (self-driving car) does not have anything to do with EV.

When people get excited about SDC and I point out that it is not ready, the argument invariably brought up is that it does not have to be perfect, only better than humans drivers.

Yes, I do not disagree with that. But as an AAA test of the recent pedestrian detection and avoidance system on 4 different cars shows, the current system is so dismal, it falls way short of what is needed. Some cars repeatedly ran over the mannequin.

The cars tested are: 2019 Chevy Malibu, 2019 Honda Accord, 2019 Tesla Model 3 and 2019 Toyota Camry.

The Camry performed the best, beating Tesla. In one particular scenario, all cars showed no indication of detecting the test mannequin.

The tests were done at 20 mph and 30mph. At 20 mph and broad daylight, the systems were already flaky. At 30 mph, all systems were ineffective. At night, they all fail, all the time.

The video I linked in the other thread is shown below.

Back to the argument that SDC does not have to be perfect, am I too demanding to want a self-driving car to be able to detect a pedestrian in broad daylight, in plain view and not obscured by any object, and to be able to brake to a complete stop when traveling at 20mph? And to be able to do that repeatedly and flawlessly?

If an SDC cannot do the simple task above, how is it better than a human driver? What kind of human driver runs down a pedestrian at every encounter, in broad daylight? Do you drive like that?

We will eventually have SDC, but the current technology in production falls so short.

 

[ls99]

In the year 2525?

Zager and Evans

In the year 2525, if man is still alive
If woman can survive, they may find
In the year 3535
Ain't gonna need to tell the truth, tell no lie
Everything you think, do and say
Is in the pill you took today ...........

Will Tesla still be a car company?

 

[Sunset]

Because giving up control is so scary to most drivers, I don't see self driving cars being a selling point until they are proven to be about 90% safer than a human driver.

Removal of the steering wheel and brake will take much longer, unless automakers install ejection seats

 

[NW-Bound]

Because giving up control is so scary to most drivers, I don't see self driving cars being a selling point until they are proven to be about 90% safer than a human driver.

Removal of the steering wheel and brake will take much longer, unless automakers install ejection seats

Pedestrians and bicyclists do not have ejection seats.

 

[Sunset]

Because giving up control is so scary to most drivers, I don't see self driving cars being a selling point until they are proven to be about 90% safer than a human driver.

Removal of the steering wheel and brake will take much longer, unless automakers install ejection seats

Actually i thinking of safety of the driver. Even if self driving cars kill the same number of pedestrians, this will be acceptable to the buying public, as long as the driver and passengers get a big increase in safety.

After all, many vehicle buyers will buy a big vehicle like a truck/suv, instead of a smart car as a collision feels safer in a big vehicle.

 

[NW-Bound]

OK, I see your point. People are selfish, and want to protect themselves even if it is at the expense of the public.

Then, we have to allow pedestrians and bicyclists to arm themselves with bazookas or RPGs.

 

[aja8888]

Pedestrians and bicyclists do not have ejection seats.

No need to eject. It happens automatically when bashed by a car. driverless or not.

 

[NW-Bound]

No need to eject. It happens automatically when bashed by a car. driverless or not.

A few posters have expressed delight at seeing the mannequin being repeatedly knocked into pieces when hit by the test cars.

The sound of the impact however was "crunchy", not like that made by real human bodies.

 

[The Cosmic Avenger]

But while SDC pedestrian avoidance needs to improve, this is only one aspect of automobile-related accidents, injuries, and deaths. I don't think I have ever needed to brake sharply to avoid a pedestrian, but I have had to brake sharply to avoid a serious accident with another vehicle. And while I realize that that is anecdotal, not data, according to the NHTSA, pedestrian fatalities are only 12% of all traffic fatalities. If SDCs can make a huge difference in the other 88% (https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/810968, p.11), I think we need to consider how many occupant lives we're basically willing to trade in order to save one pedestrian. I do think you are making a very important and often ignored point when it comes to SDC safety, of course, but it's not that simple.

 

[NW-Bound]

Out here in the west, we do not have many pedestrians, but I encounter pedestrians all the time in parking lots. And I also drive through school zones all the time. In the cities, I am sure there are a lot more pedestrians. NYC, Chigaco, SF, Seattle?

Now, here's why I said the test results as seen by AAA have to be flawless. In our lifetime of driving, we hope not to run down a pedestrian a single time. And how many pedestrian encounters do we get in a lifetime of driving?

Now, suppose you have a really lousy driver who runs down a pedestrian once every 1000 encounters. In order for the SDC to beat him, it has to have the accident rate of less than 1/1000. Did the AAA test each car 1000 times? Each time trying different cloth colors on the mannequin? Varying light conditions, dawn, dusk, twilight? Rain, fog, snow?

What does it take to prove that the system is better than a human driver, and a lousy one at that? I don't know what it would take, but the AAA simplistic test shows that we have a long way to go. The results have to be flawless in this simplistic test, before we even go further.

And that's what I was trying to show. I never claim we will never have SDC, only that people should not put too much trust in current systems, and get themselves or bystanders killed. That has happened already when Tesla cars drove under semi-trailers, plowed into highway barriers, parked red fire trucks, etc...

Back on pedestrian detection braking systems, even imperfect systems as seen on the Toyota Camry are a welcome start. I would not mind having that. But it serves to supplement the human driver, not to replace him.

 

[ERD50]

..., according to the NHTSA, pedestrian fatalities are only 12% of all traffic fatalities. ....

To add to what NW-Bound just posted, if all cars were SDC, and anywhere near as bad as that video shows, pedestrian accidents would go up by some huge factor, and would no longer be 12% (they'd be maybe 12,000% of current!).

... If SDCs can make a huge difference in the other 88% ... I think we need to consider how many occupant lives we're basically willing to trade in order to save one pedestrian. I do think you are making a very important and often ignored point when it comes to SDC safety, of course, but it's not that simple.

But that's the big "if". AFAIK, the current stats are on highway driving (and new cars), which is far safer than average miles, but I think gets compared to overall averages, so I'm not sure we have good data yet? And if it is so bad with pedestrians, in broad daylight, what makes you think it can handle a highway with maybe a car with a flat on the side of the road, a truck that loses some of its load, or any number of oddball occurrences?

Many of these posts seem to be written in circular/tautological logic. If/when SDC is better than average drivers, it will be better than average drivers? Well of course. But when?

And the consequences must be considered. We also hear that the car will ask the driver to take over if conditions are not suitable for SDC (snow covering the lanes, construction, something else?). So now you have a driver who maybe hasn't actually driven for 3 months, and they need to take over? I bet they are lousy drivers, due to inexperience.

Yes, it's not simple.

-ERD50

 

[NW-Bound]

... We also hear that the car will ask the driver to take over if conditions are not suitable for SDC (snow covering the lanes, construction, something else?). So now you have a driver who maybe hasn't actually driven for 3 months, and they need to take over? I bet they are lousy drivers, due to inexperience.

Yes, it's not simple.

-ERD50

Even in the future when we have real SDC, there will be times when the system fails. Nothing works all the time and forever. Computer hardware fails. Your smartphones and laptops fail with time. Airplanes are designed to be a lot safer than cars, and they still crash. We have accepted this risk, as long as it is small and can be quantified or defined.

For cars as well as anything else that can put human life at risk, we design them so that they fail "safe". It's OK for an SDC computer to detect a failure, then try to coast to the side of the road and summon help via a satellite link.

Designed right, an SDC can handle an internal failure in the safest way possible, the same as a human driver handles a flat tire or a blowout event while driving.

Of course, before they worry about how to handle failures, they need to get the system to perform correctly when everything is a go. This has not been demonstrated.

 

[CaptTom]

A few posters have expressed delight at seeing the mannequin being repeatedly knocked into pieces when hit by the test cars.

The sound of the impact however was "crunchy", not like that made by real human bodies.

Owwww. Yeah, that's taken the discussion to a whole 'nuther level. I don't think I can enjoy the sight of the impacts any more.

...If SDCs can make a huge difference in the other 88% (https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/810968, p.11), I think we need to consider how many occupant lives we're basically willing to trade in order to save one pedestrian.

Out here in the west, we do not have many pedestrians, but I encounter pedestrians all the time in parking lots. And I also drive through school zones all the time. In the cities, I am sure there are a lot more pedestrians. NYC, Chigaco, SF, Seattle?

Now, here's why I said the test results as seen by AAA have to be flawless. In our lifetime of driving, we hope not to run down a pedestrian a single time. And how many pedestrian encounters do we get in a lifetime of driving?

A couple of comments on these discussions about pedestrian accidents.

The hits in the video were unrealistic. The car was often going too fast for the conditions, and the dummy pedestrians darted out from behind other vehicles or timed their crossing to maximize the chance of being hit.

Most pedestrians don't do that.

I doubt that even a good human driver would have done much better. An average driver, not paying close attention, would have done far worse. And remember, half of all drivers are worse than average.

Yes, I know that children who don't know any better and dart out don't deserve to ever be hit. But the fact remains, it happens now.

I almost never see children in a school zone when the "15MPH" lights are flashing. Heck, around here parents drive 100' to the end of their driveways in their SUVs to pick up the kids from the school bus. There are simply fewer kids in the road these days.

I'm 100% convinced that self-driving vehicles are going to kill far, far fewer pedestrians, adult and child, than the current crop of human drivers.

I think a bigger risk is that fear of killing even one idiot pedestrian who might jump out into traffic will lead to such over-caution that the self-driving cars will all be stopped dead any time there's a pedestrian within 100 yards.

 

[NW-Bound]

The hits in the video were unrealistic. The car was often going too fast for the conditions, and the dummy pedestrians darted out from behind other vehicles or timed their crossing to maximize the chance of being hit...

AAA said 20 mph and 30 mph. At 20 mph, there were misses. At 30 mph, it got really bad, even for someone crossing at a crosswalk and not darting.

About the scenario of the pedestrian or kid darting out between cars, that's a tougher test. However, AAA said the cars did not even attempt to brake even if late, or show any indication of detection up to the impact.

At night, no cars saw the pedestrian, even out in clearance (no darting). That would be the same as someone crossing at a crosswalk.

With two pedestrians standing still in the middle of the road, the cars did not see them at all! They seem to require motion.


PS. I find this kind of tests very interesting. No doubt, we will see the systems getting better and better with time. And that's why I have an interest in following this technology.

Same as AAA, I was shocked to see current systems perform so poorly, even with Tesla that claims to have multiple cameras and superfast CPUs. And it did not perform as well as the Camry.

 

[JoeWras]

The hits in the video were unrealistic. The car was often going too fast for the conditions, and the dummy pedestrians darted out from behind other vehicles or timed their crossing to maximize the chance of being hit.

I disagree. But that's OK, we can have our different opinions. For the most part, I thought they were good tests.

What these systems have to do is evaluate the wider environment better.

1) For the adult in view off the line of driving: a good driver is ready to take action should the ped be an idiot. These systems need to do that too. A good driver will slow, evaluate their escape route, look for others in the area (dogs, children, a cane, etc.) It isn't all about applying the brakes.

2) Kids are a special case. Yep, that was a very tough set of kid tests, especially the "out from the cars" version.

Still, situational awareness is necessary. Are you in a residential neighborhood? Do you see balls being thrown ahead or kicked? Are there kids anywhere around?

I didn't see any of the above so none of us would have a situational warning. If anything, the computers saw the cars in an industrial area and turned off the "kid flag." Humans would have the same bias. Still wondering what the SDCs are doing for complex situational awareness. I know they can "see" a football 100 yards ahead. Do they use that information?