NikeWings
Active Member
Yeah, I asked the same yesterday on the other thread he posted. No response.
Could be a more interesting ending to a successful auto park.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Yeah, I asked the same yesterday on the other thread he posted. No response.
Could be a more interesting ending to a successful auto park.
That's possibly, but not necessarily(unless you have some insider knowledge) true. The network itself isn't likely going to change on the fly/retrain itself in the car, but there are strategies for updating parameters the network(or other modules above it) uses to change its behavior.
All of you new folks to Tesla with ap 2 really need to know, honestly, that ap1 is very, very good. I know you bought into a promise, but you should know - like ap1, eventually it will come. I've had ap1 for over a year and it is exactly as advertised and is fantastic. The only way to perfect this is by the fleet - so realize you are part of the solution.
WannabeOwner
Well-Known Member
I think that is because of a perceived need to deliver on promises - which were made too optimistically, as usual. Back in AP1 days 99.999% [pick a number!!] of owners were Tesla Fans, and more tolerant of delays etc. Today I guess that percentage is lower ... maybe a lot lower??
I know the hardware has changed, but what I'm not understanding is why the millions of miles of AP1 driving could not be used as a raw data source for AP2 tuning (OK ... so figure out HOW the data needs to be different and CONVERT the old data into that format; even if it is less than 100% good it surely?? has to be useful in getting a good basic algorithm tuned?)
If AP2 is relying on only cars built since mid December then its a tiny fleet (compared to all AP1 cars, and ALL the miles they have EVER driven)
I think there are two possibilities:
1. Tesla never thought of that.
Or
2. It isn't possible.
My guess is #2.
I just came back from my first attempt to use autosteering on local roads. The first event was the car failing to fully follow a curve and threatening to crash into a tree before I grabbed the wheel. On another street the car was caroming along, seeming to bounce off the right and left lane markers alternately. If a cop had seen me I'm sure he would have considered it probable cause for a sobriety test.
Yes, I was annoyed by the broken promise of December delivery, but I'm disgusted with Tesla's choice to release dangerous software into our cars. If you do try autosteering, keep your hands on the wheel. Not because of nanny warnings, but because you may need it to save your life.
Yes, I was annoyed by the broken promise of December delivery, but I'm disgusted with Tesla's choice to release dangerous software into our cars. If you do try autosteering, keep your hands on the wheel. Not because of nanny warnings, but because you may need it to save your life.
GlmnAlyAirCar
Active Member
If you are in a situation where a fraction of a second makes a difference, you better have you hands on the wheel. You can only really relax on open, straight highway. I guess you AP2 drivers have to learn what is proper and improper use of AP.
Not trying to pick on you @Tangible, but this sounds so familiar, ike something that many of us would have written when the AP1 software was first released over a year ago. There's a reason they put a message on the IC to keep your hands on the steering wheel, every time you engage autosteering...it's not the same as autonomous driving. You need to monitor what AP is doing, and be ready to take over for it.
The safe way to experiment with autosteer is not to trust it at all when you start. Learn its limitations (which admittedly change with different software releases) and gradually relax as you grow more comfortable with it. After over a year and a few thousand miles of AP experience (some people have many more miles than me), I still keep a hand loosely on the wheel. Even more importantly, I always maintain situational awareness for those times when I do need to take over. (If it sounds like this is more work than driving, it's not. I have found AP1 to be a great help for long drives.)
AP2 owners, if you haven't read the following thread, I highly recommend reading it. The details might have changed a bit since these posts were written, but the mindset and your mental approach is still highly relevant.
A flight instructor teaches Tesla Autopilot
Bruce.
GlmnAlyAirCar
Active Member
But Tesla has said that almost all of the AP1 data was usable for AP2.
In reply to Phone pics of Japan
Sterling Anderson @sterling_a 22 Oct 2016
@japanfonpix nearly all of it
That doesn't mean that only AP1 fleet data is required for a well behaved AP2. The more fundamental issue with AP2 right now seems to be the accuracy/reliability of the neural nets for the camera. I think Sterling was probably right that almost all of the other fleet-learned data (road curvature, lane widths, etc) are valid between AP1 and AP2, but given that the computer vision architecture is totally different between the two platforms and MobileEye specifically designed EyeQ3 not to fleet-learn on image recognition to reduce bandwidth requirements, it's likely a misinterpretation of the above statement.
I've counted at least a dozen "I didn't pay attention for a moment and my Tesla hit a wall/fence post/guard rail" AP1 threads here. The reason that AP1 Autosteer asked you to hold the wheel was never because Tesla loves to put in nanny warnings. It was always because these driver-assistance features can give out at any given moment, without much advance notice, and despite it possibly working on the same stretch of road a while ago.
You should always always be prepared to take over at any moment, even with AP1. After 15k miles with AP1, I would not be willing to let go of the wheel if there's anything within a 1-2 second collision radius of me.
Thank you for a much more patient and measured response than mine
Ironically I saw your post go up as I was writing mine and I was thinking "dammit I have to write faster!"
Bruce.
Thanks all for your responses. My frustration with AP is based not on naivety, but on what I think are realistic expectations based on experience with other cars. I currently have a 2015 Acura MDX and recently traded a 2013 Audi A6 for my current D90. Both of these cars are equipped with adaptive cruise control and lane-keeping assistance, which are their manufacturers' equivalents of TACC and autosteering.
The LKA, in both cases, does not have a maximum speed. It works only when there are clearly visible lane markings. When it is working, it steers steadily down the center of the lane. I've never seen the meandering behavior that I experienced today, or the failure to fully track a curve.
Tesla's TACC, on the other hand, is superior. In my other cars the ACC will often over-react, braking sharply when it isn't necessary. They also require an action to resume travel after coming to a stop in heavy traffic.
I'm sure that AP will improve. In not too many years we'll all be amused and horrified by what passed for automation in 2017. I just wish Tesla had been more forthcoming about how long this will take. When they charged me $10k for EAP it wasn't characterized as a GoFundMe campaign for some future system, but that's what it appears to be.
The LKA, in both cases, does not have a maximum speed. It works only when there are clearly visible lane markings. When it is working, it steers steadily down the center of the lane. I've never seen the meandering behavior that I experienced today, or the failure to fully track a curve.
Tesla's TACC, on the other hand, is superior. In my other cars the ACC will often over-react, braking sharply when it isn't necessary. They also require an action to resume travel after coming to a stop in heavy traffic.
I'm sure that AP will improve. In not too many years we'll all be amused and horrified by what passed for automation in 2017. I just wish Tesla had been more forthcoming about how long this will take. When they charged me $10k for EAP it wasn't characterized as a GoFundMe campaign for some future system, but that's what it appears to be.
I've had far worse experiences in my 2014 Audi A6 that I traded in for a Tesla with AP1.
The Audi lane keeping assist is completely inactive below 37mph, and above 37mph, will only nudge you towards the center of the lane with mild resistance. It regularly misreads black pavement grooves as lane lines particularly in construction zones, and will nudge me in the wrong direction all the time on my commute where there's a lot of slabs of concrete that have black lines going straight forward even around a curve.
Even with the steering assist set to Early, it cannot maintain a lane if you let go of the steering wheel. The most it does is nudge when you've nearly departed out of your lane, which bounces you the other way, and then hopefully it nudges again.
Don't get me wrong, it's really helpful, but it's the human that's doing 90% of the lane keeping work. By itself, the system is unreliable even compared to AP2.
(And Audi's lane keeping max speed is around 100mph, and that's also the limit for adaptive cruise control setpoint too, unfortunately)
P.S. DISCLAIMER: I've yet to try AP2 personally, but I've been very happy with my decision to upgrade from a '14 fully loaded A6 to a AP1 Model S. Better semiautonomous functionality was my primary goal in switching away from Audi, and I've cross-tested against Audi Traffic Jam Assist on the 2017 Q7 Prestige and it still doesn't compare to AP1.
I will stand by my previous post and agree that we are transitioning from a Tesla enthusiast group to a regular consumer model where folks want what they want right now - beta warnings be damned. This is to be expected, and even more so when model 3 comes out. Tesla is becoming more mass and mainstream and many of the new folks to the club are less interested in the "mission" and more interested in the "me." This is a normal progression in the mission, but triggers a rather whistful and sad remembrance of the original attraction to it for the "me".
GeekfromFRL
Member
I Think that is part of being a tesla owner. You believe in the vision of Tesla and are willing to invest in it and be patient. If you want a car that keeps distance and follow some lines, buy Audi, BMW etc. If you want to commit to a innovative company that works on a better future for man kind, buy tesla. Just my opinion.
The most potent force working against that sense of mission that the early wave of owners are nostalgic for is Tesla itself. Over-promising, under-delivering, and most of all under-communicating are all ways of alienating the people who should be your most fervent supporters.
"Beta" is a concept borrowed from the software industry and never before seen in the automotive world. What Tesla has failed to bring across, though, is the crucial element of choice. For many software systems the user can choose between a stable production version and a beta version with exciting new features and higher risk of failure. Which version is the right one depends on mission-criticality, user risk tolerance, and other factors best decided by the user himself.
If Tesla were to adopt such an approach the benefits would be enormous. Those who crave novelty and love innovation per se could join the beta group. Others, who may be fervent about the ecological benefits of a high-performing EV but more interested in getting comfortably and safely from point A to point B than in the thrill of the journey, would be grateful to the Beta users for blazing the trail.
"Beta" is a concept borrowed from the software industry and never before seen in the automotive world. What Tesla has failed to bring across, though, is the crucial element of choice. For many software systems the user can choose between a stable production version and a beta version with exciting new features and higher risk of failure. Which version is the right one depends on mission-criticality, user risk tolerance, and other factors best decided by the user himself.
If Tesla were to adopt such an approach the benefits would be enormous. Those who crave novelty and love innovation per se could join the beta group. Others, who may be fervent about the ecological benefits of a high-performing EV but more interested in getting comfortably and safely from point A to point B than in the thrill of the journey, would be grateful to the Beta users for blazing the trail.
But this choice exists already. Do not turn it on. Safe, stable, beautiful EV. Turn it on - beta test. Pretty simple really at this point.
wesley888
Member
Good point.
On the other hand, Auto-Pilot functions require the user to explicitly go into the setting menu to enable it. So, for people who don't want beta functions, they can simply not enable Auto-Pilot functions.