Autonomous Car Progress

[boonedocks]

Autonomous cars will adhere to the posted speed limits and it won't even need to see the signs because that will be coded into the maps. With high def 3D maps a car could drive for some time completely blind just based on the maps and telemetry data. This is by no means safe but would allow a car to limp to the shoulder in a safe way that would not cause other drivers to lose their minds and go careening off the road. Remember, they got breaks too. Safe zones where cars could park in an emergency could even be coded into the maps.

You don't even have to take my word for it. Tesla does not have redundancy in every system or piece of hardware and they spent billions on these types of issues. You can trust forum trolls or the experts. I choose to trust Elon and Tesla as they have yet to let me down, aside from the time line.

" it won't even need to see the signs because that will be coded into the map" I hope you weren't being serious with that part of your statement. Take my daily drive were the database is 70% wrong and THEN add in the dynamic speed limit signs that change within minutes to control traffic on our expressways....I don't think coded speed limit signs will ever work"

 

[Reciprocity]

" it won't even need to see the signs because that will be coded into the map" I hope you weren't being serious with that part of your statement. Take my daily drive were the database is 70% wrong and THEN add in the dynamic speed limit signs that change within minutes to control traffic on our expressways....I don't think coded speed limit signs will ever work"

LOL, not the Nav Maps. The high def maps.

This:

lvl5, Inc. | HD Maps and Computer Vision Software for Autonomy

and This

lvl5, Inc. | HD Maps and Computer Vision Software for Autonomy

Level 5 is made up of ex Tesla employees btw.

 

[Reciprocity]

Just FYI Bleederskab, I dont read your responses very often because I dont really care what you think and I have better things to do.

 

[Compton]

Just FYI Bleederskab, I dont read your responses very often because I dont really care what you think and I have better things to do.

Yup, I wish there was a way to block certain people. Not that they couldn't write but so that I could spare myself from seeing their messages.

 

[boonedocks]

LOL, not the Nav Maps. The high def maps.

This:

lvl5, Inc. | HD Maps and Computer Vision Software for Autonomy

and This

lvl5, Inc. | HD Maps and Computer Vision Software for Autonomy

Level 5 is made up of ex Tesla employees btw.

Still don't see how a database of signs will help dynamically changing signs:







https://youtu.be/XI1NlyszyYI

 

[ZeApelido]

1) There's the actual sensor hardware failure (which also be the result of increment whether such as rain,snow, dust, mud, sand, etc)

2) then there's the software detection failure / inaccuracies (which also be the result of increment whether..etc),
3) then there's the computer hardware failure,
4) then there's software actuation algorithm failure (which also be the result of increment whether..etc),

Just curious, what is your native language?

 

[mongo]

Still don't see how a database of signs will help dynamically changing signs:







https://youtu.be/XI1NlyszyYI

In the original context of a sensor failure and limping to the shoulder/ median: dynamic speed signs are irrelevant and it is further unlikely you would want to accelerate during a fault mode.

From the post you replied to in #201:

This is by no means safe but would allow a car to limp to the shoulder in a safe way that would not cause other drivers to lose their minds and go careening off the road.

 

[boonedocks]

In the original context of a sensor failure and limping to the shoulder/ median: dynamic speed signs are irrelevant and it is further unlikely you would want to accelerate during a fault mode.

From the post you replied to in #201:

Sorry....didn't read the post thoroughly

 

[Reciprocity]

Still don't see how a database of signs will help dynamically changing signs:







https://youtu.be/XI1NlyszyYI

The sign database is as much about location as it is about the what the sign is trying to communicate. If a sign is as you noted, one that changes, the vision system will have to "see" and recognize the sign. The high def map will have coded in its meta data the fact that a sign exists and its exact position. This is used for the car to determine its position when it "sees" the sign but also for the content of the sign if it is in fact something that changes, like a stop light.

You keep acting like every system is failing simultaneous and that I am saying the car will drive in that situation. If that happens, its similar to the driver dying. Only in this case, the car will stop immediately. Like a kill switch, pardon the pun. If the car cannot specific see that sign, it will defer to the 3d map. That is the same case as if there was a semi blocking the view. This is a type of redundancy that does not require a camera that can go on an extend-o arm to get a view above the semi. I get that this stuff is complicated to conceptualize but you keep coming up with things you think will be an issue without first trying to figure if they really are an issue.

 

[R.S]

I get your point, but at some level everything stops lacking redundancy. Current cars have (rare) failure modes that render the vehicle directionless, powerless, or just pain unmovable. It is going to be interesting how they rate fault occurrence probability, especially since all electronic device fail given sufficient time (at a minimum due to to migration of dissimilar metals at junctions causing increased resistance). Will FSD HW have a mandatory replacement period (hours of operation or fixed time)?

That's why you need redundancy, to get the failure rate down to a minimum. The probability that one system fails is a lot higher, than the probability of both systems failing at the same time. That's why full electric steering is fully redundant right now. Because in a regular assisted steering system you can still do it manually, if the assistance fails. And as soon as there is SW involved, things will fail without notice (other than mechanical parts where you can see wear).

I am not saying the car will continue on for miles and miles.. only meters and meters. Again.. Think a bit. If you are saying that all 2 forward facing cameras and sonar and radar failed? Then the car just comes to a quick stop and turns of the hazards. I am confused as to why you dont understand that a car can safely get out of the way after losing a camera or a radar or a sonar or all sonar. Who cars if it limps into another car anyway? Cars have blowouts all the time and tap other cars. Cars will also make every attempt to not be run into. I know I do it all the time, its called defensive driving and you are actually at fault in an accident in some cases if you do not use defensive driving.

How would you handle a hypothetical system failure in the middle lane of a highway? Hard on the brakes? Slowly brake and then drive to the right in hope that you will find a shoulder lane at some point? What if the next one is a mile ahead? Those really aren't solutions. That doesn't mean it can't ever happen, but it has to be an ultra rare error. That's why you need redundancy. And the Tesla vehicles have redundancy already, as you mentioned, two forward facing cameras and radar, that should be good enough IMO.

I was just arguing that you need much more redundancy. At some point definitely a second AP chip, if the first fails. The system doesn't need to be 100% capable, but capable enough to retain some safety features, something telemetry and 3D maps can't do.

I am not sure if the current cars are redundant enough, but I think they are (besides the computer). And if not Tesla will have to retrofit, dual radar, or cameras come to mind.

I'm also not sure why some people think it's different if Tesla changes a AP chip (generally seen as totally not a big deal), or adds lidar to the cars (seen as worst case scenario). In a couple of years you will be able to build lidar into the car's head and brake lights. And they will probably be less expensive than the two supercomputers needed for the current sensor suite to get to level 5. And the LED light is connected to the CAN BUS anyways, so super simple and cheap retrofit, if needed.

 

[Reciprocity]

How would you handle a hypothetical system failure in the middle lane of a highway? Hard on the brakes? Slowly brake and then drive to the right in hope that you will find a shoulder lane at some point?

If the system fails completely, how is it going to do anything? There would be a fail safe to full stop. Exactly how fast, I couldnt say. But they will test, simulate and figure out what works best. That should exist today for test vehicles. That big red button. You are talking about a complete system failure? No cameras, no computer, no sonar, no nothing.. What is Waymo or anyone else going to do? Have to complete second systems? This is the same if the driver dies, except there will be a sub system that can bring the car to rapid stop and put on the blinkers and phone for help. Though I would imagine that these cars will be actively monitored.

 

[mongo]

Philosophical question:
Does an AP car have to be better than drivers who run out of gas?

 

[Bladerskb]

Philosophical question:
Does an AP car have to be better than drivers who run out of gas?

Philosophical question:
Do drivers who run out of gas all of a sudden go blind? or have blurred vision leading them to miscalculate the presence or absence, distance and trajectory of other road users and objects?

 

[mongo]

Philosophical question:
Do drivers who run out of gas all of a sudden go blind? or have blurred vision leading them to miscalculate the presence or absence, distance and trajectory of other road users and objects?

Philosophical question:
Do drivers who run out of gas all of a sudden go blind? or have blurred vision leading them to miscalculate the presence or absence, distance and trajectory of other road users and objects?

Actual question:
What does any of that have to do with coasting to a stop (potentially without normal brakes or steering)?

 

[R.S]

Philosophical question:
Does an AP car have to be better than drivers who run out of gas?

A driver out of gas can still steer and brake. So sure, a traction power failure must be handled by AP.

It has to be better than a drunk driver, or an unconscious driver, also.

 

[R.S]

If the system fails completely, how is it going to do anything? There would be a fail safe to full stop. Exactly how fast, I couldnt say. But they will test, simulate and figure out what works best. That should exist today for test vehicles. That big red button. You are talking about a complete system failure? No cameras, no computer, no sonar, no nothing.. What is Waymo or anyone else going to do? Have to complete second systems? This is the same if the driver dies, except there will be a sub system that can bring the car to rapid stop and put on the blinkers and phone for help. Though I would imagine that these cars will be actively monitored.

Well, first forget the idea that I am arguing against Tesla here. I am just talking about redundancy in general. So let's imagine a car with just one 360 degree camera for example. If that camera fails, or anything in between that camera and the computer fails, it's game over.

I guess that Waymo will have full redundancy. I think everyone will. Tesla already has some redundancy, radar, multiple cameras, sonar. And maybe you don't need complete redundancy, things happening in back might not be as safety relevant. That big red button actually just stops traction power, not steering and braking.

What you describe is a Level 3 system, where the driver needs to take over, if something goes wrong. A level 4 system needs to take the passengers to a safe spot and the middle of the highway just isn't safe.

 

[mongo]

A driver out of gas can still steer and brake. So sure, a traction power failure must be handled by AP.

From a different post:

A level 4 system needs to take the passengers to a safe spot and the middle of the highway just isn't safe.

This is the type of thinking (potential duality) I am curious about. How many edge cases should an autonomous car be hardened against? And how well should they be forced to handle them?

In one case you have human drivers whose cars are reduced to coast, steer, and stop. On the other hand, it appears a FSD needs to always have sufficient control, sensing, and power to pull off a busy highway (stopped traffic is impossible to pull off of but also less dangerous in the short term).

On the one hand, punishment for human drivers getting in avoidable situations is minimal to none. On the other, OEMs are projected to be raked over the coals (yes, hyoerbole-ish) if their cars are not double/ triple redundant against all failure modes.

Will autonomous cars need fully redundant power sources and wiring also? Perhaps a fully separate camera, battery, gimballed motor/drive wheel to move the car the safety?

Locally, there were people dropping concrete off overpasses which killed a driver. Should self driving cars be required to be hardened against that and the occasional bullet?

A car is driving the 401 between a semi and a lane splitting motorcycle when it runs over an object in the road (for cold weather states change to pothole) that obliterates the front wheel (at a minimum). What is allowable there? Would slamming on the brakes have been acceptable if the object were detected early enough?

I understand the drive for safety and the desire to engineer it in. It is similar to what SpaceX is going through NASA and Crew Dragon. What is the real level of fault handling needed, what outside influences should be considered, what kind of other-driver model should it be expected to protect against (in operating and fault modes)?

Where is the line of the perfect being in the way of the good?

 

[PhantomX]

In the manufacturing world, a lot of automation equipment are designed to predict failure and respond before actual failure occurs, since a failure can be quite expensive if we are talking about high volume production system. Similarly, why can't the system in the autonomous car predict system failures prior to actual failure and react accordingly? Why must wait until the system actually failed, then react?

Another thing to consider, today's road rules are designed for human drivers. I have seen discussion where road rules can evolve with the advancement of autonomous cars, including such things as higher speed and tighter spacing. If those changes do occur, stopping in the middle of the road suddenly may not be the best way to address a failure. Even pulling over to the shoulder may cause a major accident if there is large disruption to speed of traffic.

Also, regarding redundancy, should autonomous car have backup systems like those on the airplane? If not, why not? Would one be okay flying on modern airplanes without full redundancy to the avionics and control systems? why should car be different?

Interesting time ahead...

 

[cwerdna]

Same driver, different vehicle: Bringing Waymo self-driving technology to trucks

 

[cwerdna]

Autonomous Nissan LEAFs Start Easy Ride Trials

Seems pretty limited right now.. From the story

“Easy Ride is envisioned as a mobility service for anyone who wants to travel freely to their destination of choice in a robo-vehicle. During the field test, in the Minatomirai district of Yokohama, in Japan’s Kanagawa Prefecture, the participants will be able to travel in vehicles equipped with autonomous driving technology along a set route. The route spans about 4.5 kilometers between Nissan’s global headquarters and the Yokohama World Porters shopping center.
...
The field test will enable Nissan and DeNA to learn from the experience of operating the Easy Ride service trial with public participation, as both companies look toward future commercial endeavors. Nissan and DeNA will also work to develop service designs for driverless environments, expanded service routes, vehicle distribution logic, pick-up/drop-off processes and multilingual support. The companies aim to launch Easy Ride in a limited environment at first, and then introduce a full service in the early 2020s.

FWIW, I have visited Nissan's Yokohama HQ at least three times in past 3 years or so. And, late last year, I was definitely around the Yokohama World Porters area. I think I went inside that mall.