I've seen dozens of VINs with the new codes. There was an X in Fremont last week with the new autopilot but nobody went by to try to check it out when I mentioned it.Can we wait until it's here before we start claiming it's here?
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Why AP 2.0 Won't Be Here Soon, and It Won't Be What You Think It Is
- Thread starter Todd Burch
- Start date
-
- Tags
- Model S
I think Elon predicted full autonomy (though didn't say what level) would be possible within a year in the fall of 2015. He also said that regulatory agency approval was coming some time after that.
I too am an engineer and worked in the commercial aircraft industry. Car tech tends to move faster than aircraft and the regulators aren't quite as conservative, but regulators in general are conservative about this sort of thing. Most humans are wary of blindly trusting a machine, even when the machine is safer.
Commercial aircraft autopilots have been good enough for some time to fly the plane from runway to runway without any human interaction, but the minimum crew is still 2 pilots and take-offs and landings are still done on manual for the most part, even though humans error or acts of malice are responsible for almost all commercial aircraft crashes in the last 20 years.
Back in the early 90s there was a joke at Boeing that the next gen flight deck was going to be one pilots and a dog. The pilot's job was to feed the dog and the dog's job was to bite the pilot if he (or she) touched anything. Being a commercial pilot today must be stultifyingly dull.
Flying via autopilot is a simpler job for a computer to handle 99% of the time than for a car. Cars routinely get much closer to one another than planes do. The only time aircraft get very close to one another at all is around airports and there ground control is telling every aircraft for 50 miles exactly where it should be. The only time problems crop up is when someone breaks the rules, which is uncommon because to be a pilot requires a lot more training than to drive a car.
Autonomous cars need to deal with other cars, pedestrians, bicycles, motorcycles, trucks, animals, etc. All of which can be difficult to predict. The only animals planes need to deal with are birds that usually get mulched. On top of that autonomous cars need to deal with a lot of conditions planes don't. They need to obey all signs, traffic laws, traffic lights, and do so in a wide range of environmental conditions. On top of all that, they need to deal with varying road conditions. Sometimes lines on the pavement may be invisible due to lighting conditions, construction, or just the paint failing with age. Road surfaces vary in quality from brand new paving to pot holed messes.
An aircraft autopilot may only have to deal with an edge condition once or twice a year, an autonomous car may have to deal with them several times a day.
Regulators are getting pushed hard to approve autonomous car tech, but they are going to drag their feet until it can be proven. I suspect Tesla is recording tons of data and will be recording even more with the AP 2 cars and they will be running their offline simulations against that data to prove the autonomous software could handle real world situation. It will take millions of hours of data to prove the system can handle enough situations to get approved.
It just seems a little early for the I told you so speech. If it takes them 2 years to get it working, the OP is basically correct.
⚡️ELECTROMAN⚡️
Village Idiot
That's a really big if. Maybe if the software version 8.0 team is working on it. Heaven forbid.
Enabling full self driving will require regulatory approval AND resolving liability.
Of the two, the regulatory hurdles may be the easiest.
With full self driving software, Tesla is essentially providing the driver for the car - so from one perspective, an argument could be made that Tesla should be held accountable for any accidents that happen when the car is in self driving mode, which includes when the car is driving completely by itself without any humans in the car (which was shown in the demonstration video, and is a basis for Tesla's ride service).
Though, will Tesla accept liability for every car operating in self driving mode?
And will drivers and their insurance companies accept the liability for a car operating in self driving mode, when there isn't anyone even in the car?
Accidents will happen - no matter how good the software and hardware are - and when they do, who will be responsible?
It may take much less time to get the software operational - than getting over the regulatory and liability hurdles...
Of the two, the regulatory hurdles may be the easiest.
With full self driving software, Tesla is essentially providing the driver for the car - so from one perspective, an argument could be made that Tesla should be held accountable for any accidents that happen when the car is in self driving mode, which includes when the car is driving completely by itself without any humans in the car (which was shown in the demonstration video, and is a basis for Tesla's ride service).
Though, will Tesla accept liability for every car operating in self driving mode?
And will drivers and their insurance companies accept the liability for a car operating in self driving mode, when there isn't anyone even in the car?
Accidents will happen - no matter how good the software and hardware are - and when they do, who will be responsible?
It may take much less time to get the software operational - than getting over the regulatory and liability hurdles...
Maybe you folks should read the original post again. If you think Level 5, fully autonomous driving has arrived in a Tesla, I want what you're smoking. Again, the OP is about autonomous software, which is 98% of the problem.
I'll say it again. It's about software. Hardware does not mean it's here.
It's a software AND a regulatory issue.
Just throwing a processor and a bunch of cameras in a car does not make it fully autonomous, nor does an edited video showing selected scenes from a single drive in light traffic. (And I'd hardly call Tesla's video demo a "complex urban environment.")
Not sure what is hard to understand about that, but I don't see a need to argue it any further. If you believe it's here, I'll be looking forward to all of the videos people post of their empty Teslas driving around town.
If the prediction was "off by eons", where are the videos?
I'll say it again. It's about software. Hardware does not mean it's here.
It's a software AND a regulatory issue.
Just throwing a processor and a bunch of cameras in a car does not make it fully autonomous, nor does an edited video showing selected scenes from a single drive in light traffic. (And I'd hardly call Tesla's video demo a "complex urban environment.")
Not sure what is hard to understand about that, but I don't see a need to argue it any further. If you believe it's here, I'll be looking forward to all of the videos people post of their empty Teslas driving around town.
If the prediction was "off by eons", where are the videos?
What makes you think Tesla picks up the bill for L4 accidents? If you listen to his press conference just last week, even with L5 Elon says Tesla would only cover issues that are shown to be a technical issue on Tesla's side. (He uses Otis elevators as an example to support his viewpoint).
"You will probably not see AP 2.0 hardware for another few quarters at the very earliest. And that's being optimistic"
That's totally deserving of a joke on being miles off in terms of the hardware release. Where people saw that post, and then felt comfortable about getting AP1 thinking the hardware wouldn't change for at least a few quarters.
Now I didn't mean anything serious about the joke because it's always a guessing game, and sometimes were miles off even if we think we have a good feel for what will happen.
As to software I don't have a conflicting opinion.
In the history of the entire world has anyone ever been right who's predicted the software (any software) will be here sooner than expected?
roblab
Active Member
I suppose you saw what I saw TODAY on the news. An OTTO Autonomous Equipped truck drove from warehouse to warehouse with a semi load of beer, with no driver in the driver's seat (he was watching from the cab sleeping quarters) for the entire 150 miles. Why can Tesla not be able to do this and better, in even another month, if not even already doing it?
Anything you or I think is pure speculation.
Going back and reviewing this thread, I actually side with @Todd Burch's claim that he is not incorrect insofar as his original estimates and the current reality. He was wrong on hardware, and he admits it (though he minimizes the extent to which he was wrong by saying "a quarter or two"), but the overall gist of the original post was that functionality wouldn't be around for an extended period of time. He may still be wrong, but we don't currently have any evidence that contradicts that post outside of the hardware claim.
Given that, there are a good number of very generous "get out of jail free" cards here in this thread, and none were picked up. A little humility can be an appealing trait. It also minimizes pushback when and if misplaced confidence meets reality.
Either way, I think the Electron Pilot was not wrong. His timetable gives him a decent period of time to be wrong, but we haven't reached that, and we may not. So I think the self-righteous smirks from the opposition can probably go.
Just one dude's observation.
Given that, there are a good number of very generous "get out of jail free" cards here in this thread, and none were picked up. A little humility can be an appealing trait. It also minimizes pushback when and if misplaced confidence meets reality.
Either way, I think the Electron Pilot was not wrong. His timetable gives him a decent period of time to be wrong, but we haven't reached that, and we may not. So I think the self-righteous smirks from the opposition can probably go.
Just one dude's observation.
Regarding the truck: while being able to drive from one warehouse to another is a great feat, it does not cover the wide variety of scenarios a true L5 system must handle. It's like Elon's demo driving from NY to LA by the end of next year. If they are lucky and no odd scenarios are encountered along the way, it can be done. But the minute some unusual situation comes up, there will be problems. What do you think would happen if that truck encountered a curve in which some parked cars prevented the semi from navigating the curve safely? Do you think the truck would have backed up and found a different route? Don't think so...
And no government is going to let these cars roam around without a safety backup until those problems are solved.
I use the construction flagger as an example. What would the truck have done were it to encounter a construction flagger? Or a downed tree in the road with a car parked on the shoulder and traffic coming in the other direction? Or a sinkhole? Or an accident blocking all lanes of traffic and forcing a detour?
I've watched a few lectures given by the Google car project's lead engineer. While Google's cars can navigate the majority of scenarios, many of them are handled with extreme hesitation and slowness, angering other drivers. Some scenarios the car just doesn't know what to do. It is these difficult edge cases that prevent full autonomy for now. These very difficult edge cases must be solved before true L5 is a reality.
Only then will it be considered for regulatory approval, and only after that approval will Joe Schmo be able to send his car off to Times Square to pick up a rider for the Tesla Network.
And no government is going to let these cars roam around without a safety backup until those problems are solved.
I use the construction flagger as an example. What would the truck have done were it to encounter a construction flagger? Or a downed tree in the road with a car parked on the shoulder and traffic coming in the other direction? Or a sinkhole? Or an accident blocking all lanes of traffic and forcing a detour?
I've watched a few lectures given by the Google car project's lead engineer. While Google's cars can navigate the majority of scenarios, many of them are handled with extreme hesitation and slowness, angering other drivers. Some scenarios the car just doesn't know what to do. It is these difficult edge cases that prevent full autonomy for now. These very difficult edge cases must be solved before true L5 is a reality.
Only then will it be considered for regulatory approval, and only after that approval will Joe Schmo be able to send his car off to Times Square to pick up a rider for the Tesla Network.
I don't mind being wrong. In fact, I'd love to be wrong, in which case I'd be looking for a path to upgrade my car.
But we've only just heard the hen is pregnant, and folks are already counting the chicks. Or something like that.
But we've only just heard the hen is pregnant, and folks are already counting the chicks. Or something like that.
henderrj
Active Member
Can't comment with accuracy on digital cameras but with film (albeit 35 years ago) this was not true in bright sunlight. Also if you look closely when the car takes the exit there is just a moment when it is in shadow - that's when I thought I saw just the effect you mention. But, as I say, daylight makes it very difficult to tell. I also wouldn't be surprised to find a filter (software or physical) that is removed at night, or at least in bright light. Oh well, guess we'll have to wait for Tesla to peal out the data one bit at a time!
henderrj
Active Member
Actually they clearly said "from on ramp to off ramp" with the driver doing all the surface street work. Still, an impressive feat. And I really wonder how they got regulatory approval to have no safety driver in the drivers seat! Those Budweiser people must have a lot of political pull.
That's just Leon's reality distortion field...he wishes someone else was left holding the bag, and he may have been successful fooling internal legal counsel...but that doesn't make it true.
If these cars are released with autonomous function, and they are involved in accidents where the other car is not found to be at fault, the insurance company will do one of 2 things (and likely both)
1) Adjust the cost to insure the car to offset the new expectation of Tesla AP caused accidents (and the accuracy of that charge will improve as more data is available, but I assure you, insurance companies err on the conservative side
)2) Seek compensation from the other responsible party...Tesla
That's just Leon's reality distortion field...he wishes someone else was left holding the bag, and he may have been successful fooling internal legal counsel...but that doesn't make it true.
If these cars are released with autonomous function, and they are involved in accidents where the other car is not found to be at fault, the insurance company will do one of 2 things (and likely both)
1) Adjust the cost to insure the car to offset the new expectation of Tesla AP caused accidents (and the accuracy of that charge will improve as more data is available, but I assure you, insurance companies err on the conservative side
2) Seek compensation from the other responsible party...Tesla
I absolutely agree that one of those two things will happen...and more likely (2). However I'm not sure all of these legal implications have really been hashed out yet, since such autonomous technology is not yet in production.
Will make for some interesting legal battles in the future.
According to Tesla's website: Tesla believes the new AP 2.0 hardware is capable of full self-driving ("in almost all circumstances"). They state that enabling the functionality is "dependent upon extensive software validation and regulatory approval, which may vary widely by jurisdiction."
For $3,000 (or $4,000 after delivery), what are you actually getting? If all of the cars already have all of the AP 2.0 hardware, without the self-driving software, what benefit do you really get from activating all 8 cameras (instead of 4)?
And, what if Tesla discovers, after "software validation" the AP 2.0 sensor suite isn't sufficient to enable self-driving? Will they refund the Full Self-Driving payments? Or will they provide free retrofits to fix any hardware deficiencies?
Unless Tesla is providing a guarantee that for $3,000 you will absolutely be getting self-driving capability (eventually), they should make some changes to how they are advertising this option - and be more realistic that they can't completely predict if the current hardware and future software will be sufficient to ever support fully autonomous driving.
As an "early adopter", when we place the order for our next Model S - and our Model 3 (next year?), we're likely to check off the box for Full Self-Driving. And, my expectation is that while we won't have full self-driving for years (because it's not just the software - Tesla also has to address the regulatory and liability issues), it is very likely we'll see "enhanced" "enhanced autopilot" that will be able to drive the car "in almost all circumstances", with the driver maintaining control (as is the case today with AP 1.0). And while that won't be full self-driving, it will still be an improvement, and could be worth the additional $3,000, to extend AP to surface streets.
For $3,000 (or $4,000 after delivery), what are you actually getting? If all of the cars already have all of the AP 2.0 hardware, without the self-driving software, what benefit do you really get from activating all 8 cameras (instead of 4)?
And, what if Tesla discovers, after "software validation" the AP 2.0 sensor suite isn't sufficient to enable self-driving? Will they refund the Full Self-Driving payments? Or will they provide free retrofits to fix any hardware deficiencies?
Unless Tesla is providing a guarantee that for $3,000 you will absolutely be getting self-driving capability (eventually), they should make some changes to how they are advertising this option - and be more realistic that they can't completely predict if the current hardware and future software will be sufficient to ever support fully autonomous driving.
As an "early adopter", when we place the order for our next Model S - and our Model 3 (next year?), we're likely to check off the box for Full Self-Driving. And, my expectation is that while we won't have full self-driving for years (because it's not just the software - Tesla also has to address the regulatory and liability issues), it is very likely we'll see "enhanced" "enhanced autopilot" that will be able to drive the car "in almost all circumstances", with the driver maintaining control (as is the case today with AP 1.0). And while that won't be full self-driving, it will still be an improvement, and could be worth the additional $3,000, to extend AP to surface streets.
