Self Driving Cars?

[Rustic23]

Earlier post commented on aircraft autopilot. I flew the F-111. This was the earlyaircraft with auto terrain following radar. After takeoff we would climb to 20,000 ft, check out the system, and then engage the TFR. It would nose over and head for the deck. Passing through 10,000 ft it would increase it pitch down as the ground radar kicked in. It would level off at 200 ft, 480 kts, and bounce around on course. In the day, not bad, at night no moon, sporty!

This system was triple redundant, except for the actual radar units, there were two of these in the nose of the A/C. Three computers took the same inputs from both radars, compared them. A mismatch would cause an automatic two g pullup.

It would be interesting to see what kind of backup and cross checking the auto delfdrive systems use.

 

[ERD50]

It's not flashing if it's done sparingly. But when it's repeated dozens of times in a thread it strongly suggests another purpose.

Well, different people will see things differently. I only saw it as a note to reinforce that he has a certain level of experience with the info being posted, it's not being posted with just a "I don't know, but wouldn't it be neat if" kind of mentality.

As far as repetition, this is a loooooong thread 500+ posts at this point? I actually like an added level of redundancy, as it's easy to get lost with all the various input. A few times, posters have mentioned "I posted that earlier" - OK, it's a long thread, and we have a couple other related threads going on, and I don't have total recall. Excuse me!

-ERD50

 

[eroscott]

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.

The GPS is only one of several things these cars use to understand lanes. Certainly it is not even the primary thing. Teslas have "high definition" GPSs. Certainly to help send the individual lanes up to the cloud for mapping and to help in identifying "normal" (common) curve trajectories. Teslas also automatically slow down on some sharp curves.

Mobileye uses the term 'Lane Fusion' which is several things to pinpoint the best estimate of the lane. They use camera still images! Mobileye is in my first version/generation of Tesla AutoPilot.

In Teslas case they use different shades of a grey road on your drivers info center display to let you know what they are seeing. Then as lanes are identified they are dark grey lane lines. These may only show your lane or if they see neighboring lanes they they will show up too (hint when you see these you can turn on your blinker in the Tesla and it will automatically change lanes - diff topic but I use this all the time AND it works WHILE going around a curve ... let that sink in). The lane lines will go blue when they VERY confident in seeing/identifying them and the road paint is solid or dashed.

Other cars on the road are also a major help to confirming lanes. These show up on the Tesla display and in multiple shades/colors to give you a hint on their importance or reliance. Light grey, white, blue in the main lane you are in. It shows cars in other lanes. Tesla is very careful to fade these surrounding cars out as well when they do not become important. ie. stoplight or non-busy road where they are watching for cars coming in your lane.

If any of you have never seen what Mobileye (and now Nvidia) "can do" (deep learning) with single images of camera snapshots you will be blown way watching this old video. I watch it many months ago but thought it may be useful here.

I suggest watching it on a PC where you can use the gear icon to change the speed to x1.25 or x1.50.

The attached is the graphic Mobileye video above.

 

[eroscott]

Additional supplemental video 1432 seconds in tho. Good examples.

Title: Autonomous Car - what goes into sensing for autonomous driving?
https://youtu.be/GCMXXXmxG-I?t=1432

/update -- Also in this at about 32:20 he talks about Tesla AutoPilot. Again my HW1+AP1.5 uses Mobileye but HW2/AP2 uses NVidea ( https://electrek.co/2016/10/21/all-...-new-drive-px-2-ai-platform-for-self-driving/ )

 

[NW-Bound]

Thanks for sharing. Some info from Mobileye, I have seen. The others, I have not. I will watch them, as I have watched and read everything people shared on this thread.

I have enough of this discussion for a while. And I will refrain from making any comments, good or bad. I have spent too much energy on this.

I will wait for a few years, when that Honda is coming out, and do more research when I am about ready to buy a new car.

Thanks again.

 

[JoeWras]

There is one bit of new gps technology that could hasten the development of SDC's. The gps block III satellites are scheduled to begin launch in 2018. They have at least one extra civilian signal and have submeter accuracy. Certainly an improvement over current gps accuracy.

I am concerned about dependence on single points of failure. GPS is one thing, but "the cloud" is quite another. I won't bore you with talk of HA and "5 or 6 9s" talk. Suffice it to say I hope they are thinking about this. Also, the ability of government or bad actors to create a travel apocalypse is concerning, especially if they rip out the wheel and controls.

I know there will be many discussions about this, though. Smart people are thinking about it. I'm still concerned though.

 

[Midpack]

I am concerned about dependence on single points of failure. GPS is one thing, but "the cloud" is quite another. I won't bore you with talk of HA and "5 or 6 9s" talk. Suffice it to say I hope they are thinking about this. Also, the ability of government or bad actors to create a travel apocalypse is concerning, especially if they rip out the wheel and controls.

I know there will be many discussions about this, though. Smart people are thinking about it. I'm still concerned though.

Supposedly Waymo test cars have a kill switch, and I would think that would exist for many years if not indefinitely, after steering wheels and pedals go away. But I don't know how it would work, obviously it can't just stop the car in the middle of a road. Like you say, I am sure they're thinking about it.

 

[JoeWras]

Supposedly Waymo test cars have a kill switch, and I would think that would exist for many years if not indefinitely, after steering wheels and pedals go away. But I don't know how it would work, obviously it can't just stop the car in the middle of a road. Like you say, I am sure they're thinking about it.

Less concern about a kill switch than bad actors (say, a country we are at war with) takes over and ruins the entire infrastructure. This is a black swan event and probably too far off the radar.

Closer to home would be emergency situations where something went down, and all the cars went to safe mode. Gotta make sure police, ambulances, etc. can still go. I suppose there will be a day that driving is a skilled profession.

 

[Midpack]

Closer to home would be emergency situations where something went down, and all the cars went to safe mode. Gotta make sure police, ambulances, etc. can still go. I suppose there will be a day that driving is a skilled profession.

Not sure if you meant it this way, but I'll bet police, fire, ambulance, military, etc. will be the last to lose manual controls, if ever. Hadn't occurred to me until just now.

 

[JoeWras]

Not sure if you meant it this way, but I'll bet police, fire, ambulance, military, etc. will be the last to lose manual controls, if ever. Hadn't occurred to me until just now.

That's exactly how I meant it.

 

[Music Lover]

Not sure if you meant it this way, but I'll bet police, fire, ambulance, military, etc. will be the last to lose manual controls, if ever. Hadn't occurred to me until just now.

Likely there will eventually be some sort of programming that would move all the vehicles to clear a path for emergency vehicles..no more relying on idiot humans who don't want to or are too stupid to get out of the way.

 

[Totoro]

Not sure if you meant it this way, but I'll bet police, fire, ambulance, military, etc. will be the last to lose manual controls, if ever. Hadn't occurred to me until just now.

And the other way around? A self-driving ambulance fleet might have better coverage and be on scene faster? If I recall correctly, ambulances are also very frequently in accidents themselves, especially intersections.

I wonder if police would get the capability to commandeer vehicles from a distance.

 

[explanade]

Looks like another milestone coming up this month in CA.

But check out the note about $26 billion market by 2025 and millions in a few years:

But this month, California set the stage for the next phase of innovation that could dramatically alter transportation and mobility across the globe. The state has proposed regulations to allow fully autonomous vehicles to drive on public roads – meaning empty cars with no steering wheels and no backup driver inside.


Empty cars with no steering wheel could soon be driving in California
Read more
The new rules are a game-changer for the nascent industry, opening the doors to a host of complex questions about legality, ethics and safety. The regulations, which could go into effect this year, pave the road for a deployment that could revolutionize modern society.

“This is like the smartphone transition times 10 as far as the potential to change our existence on the planet,” said Karl Brauer, senior analyst with Kelley Blue Book, an automotive research company. “There’s a sense of almost panic and certainly a frantic pace that all these industries are going through to try to position themselves in this new world.”

The race to dominate the market is rapidly accelerating in California, where major technology corporations, traditional automakers and artificial intelligence startups are engaged in aggressive competition. In an industry that could be worth $26bn by 2025, with potentially millions of vehicles on the road in just a few years, there’s a lot at stake.

https://www.theguardian.com/technol...ation-google-uber-tesla?CMP=oth_b-aplnews_d-1

 

[meierlde]

Note that for example I would like to see someone try a self driving jeep over black bear pass in Telluride (view the ride in a regular jeep on youtube). There are switchbacks that take backing to make for example.

Note the apparent need for human takeover in Wamo is about once every 5k miles.

BTW having worked in tech you see lots of things get hyped to the sky and then disillusionment occurs before the tech becomes a true sucess.
But we do see that non traditional automakers have a problem scaling production up consider how long it has taken Tesla to expand production.

 

[eroscott]

Just to give you guys an idea of what cameras and algorithms can do...

I was taking a back road to get around traffic yesterday but was still able to use Tesla AutoPilot (see attached picture of road and dash). Impressive for such a bad road, right? That road changes a fair amount as there is some more visible paint in other areas. It also uses changes is color and texture to understand lanes. Video in my 03-15-2017 09:40 PM post show some of this.

The blue steering wheel tells driver AP is on and blue (vs dark grey) lane lines on the Driver Info Center indicate AP has a "lock" on the "lane lines".

 

[explanade]

Ford issued a warning about revising their forecasts downwards.

MSNBC was talking about how the company is heavily-dependent on robust US sales.

And one area where expenditures could affect finances is their SDC R&D.

They also noted that Waymo has way better driver-intervention numbers than anyone else and that Waymo had partnerships with some manufacturers such as Honda or Toyota.

 

[Midpack]

They also noted that Waymo has way better driver-intervention numbers than anyone else...

"Disengagements" FWIW http://www.early-retirement.org/forums/f27/self-driving-cars-85740-18.html#post1851106

And one source re: Waymo vs Tesla disengagements? http://www.cbronline.com/4th-revolution/google-waymo-overtakes-tesla-self-driving-car-race/ . I suspect Elon Musk might tell a different story, and I wouldn't presume to know better.

In 2016, Waymo logged over 600,000 autonomous miles in California and reported just 124 disengagements whereas the Tesla Autonomous Vehicle Program drove only 550 miles and reported 182 disengagements. This means that for every 1,000 miles Waymo disengaged 0.2 times compared to 330 times by Tesla.

Gartner analyst Mike Ramsey told CNBC: “This is just more evidence of the gulf between Waymo and all other companies in terms of testing and execution of self-driving systems.”

 

[Chuckanut]

Here's how you trap a self driving car.

Trapping a Self-Driving Car is Surprisingly Easy | Nerdist

The car will drive through the dotted line, but by the time it processes that solid lines aren’t to be messed with, it finds itself surrounded by solid white on all four sides.

 

[Midpack]

Here's how you trap a self driving car.

Trapping a Self-Driving Car is Surprisingly Easy | Nerdist

Cute, the site must be like The Onion. The SDC would never put itself in that circle.

 

[JoeWras]

Here's how you trap a self driving car.

Trapping a Self-Driving Car is Surprisingly Easy | Nerdist

Well, when cars depend on GPS, this won't be a problem.

The weirdest thing about this video is that I've actually been in that parking lot! I haven't been in many places in the world, but I've been there. Recognized it immediately. It is near Delphi, Greece.

 

[Ronstar]

An Uber self driving car involved in crash in Tempe, Az

https://www.bloomberg.com/news/articles/2017-03-25/uber-autonomous-vehicle-gets-in-accident-in-tempe-arizona

 

[travelover]

An Uber self driving car involved in crash in Tempe, Az

https://www.bloomberg.com/news/arti...ous-vehicle-gets-in-accident-in-tempe-arizona

Hard to tell much from the article about this accident. For self driving cars at this point, it is kind of like when a car runs into a motorcyclist, it is termed " a motorcycle accident".

 

[Midpack]

Hard to tell much from the article about this accident. For self driving cars at this point, it is kind of like when a car runs into a motorcyclist, it is termed " a motorcycle accident".

And the irony is, some people use these SDC accidents where a manually driven car is at fault, to show why SDC's are unsafe. Kind of a twisted Catch-22.

Almost all the Waymo accidents have been another car rear ending them while the Waymo vehicle was stopped at an intersection. It's not like a driver could do anything if he/she was in the same situation, they'd get rear ended too. There's either a car in front of you, or most likely crossing traffic. Yet I've had a friend use this exact situation as his leading argument against SDC with "I don't want to be a sitting duck." Hello?

IMO it's going to be great when the roads are filled with SDC's. But getting there could be messy, and manually driven cars will be at fault more often than not...

 

[ERD50]

And the irony is, some people use these SDC accidents where a manually driven car is at fault, to show why SDC's are unsafe. Kind of a twisted Catch-22.

Almost all the Waymo accidents have been another car rear ending them while the Waymo vehicle was stopped at an intersection. It's not like a driver could do anything if he/she was in the same situation, they'd get rear ended too. There's either a car in front of you, or most likely crossing traffic. ...

Not necessarily.

The question would be (and I don't think we know in this case), could the SDC be smart enough to make an evasive maneuver? Perhaps a real driver could have avoided the accident?

I've avoided being rear-ended, so you can't say there is nothing a driver could do. In my case, I was stopped at a light, and saw a car coming way too fast in my rear-view mirror. I looked left/right, it was clear, so I floored it and got on the horn, and yelled to my passengers so they knew what was going on. At that point, I could hear his breaks squeal, and he ended up mid-way through the intersection. Sure, if I was trapped to the front/sides, not much I could do. And maybe SDC have been programmed to handle this as I did - but do we know that?

I also avoided a front-end collision. I came up behind a guy in a parking garage that was stopped to make a right turn. He must have realized he actually wanted to turn left, and just started backing up, seemingly unaware that I had pulled up behind him after he had stopped. I simultaneously got on the horn, and hit reverse. Between the two actions, I got out of his way, and he stopped when he heard the horn, so I didn't need to back up very far.

Maybe instead of accusing 'some people' of twisted logic, perhaps you should consider they may have a point. SDC's aren't all sunshine and puppies either.

-ERD50.

 

[Music Lover]

And the irony is, some people use these SDC accidents where a manually driven car is at fault, to show why SDC's are unsafe. Kind of a twisted Catch-22.

It works both ways...the actions of a small percentage of drivers is used to promote self-driving cars as being safer than all drivers in all situations.

It probably won't take much for a self-driving car to be safer than a frail 86-year old person with slow reflexes and poor vision, but how much more work is needed to make one safer than all the good drivers?