Buyer Beware: AP2 Extremely Dangerous

[NerdUno]

I just sent the following note to Tesla, but wanted to alert others to some of the problems we have experienced with AP2...

We have 17.3.2 and have used AutoSteer extensively on a number of interstate highways since its initial release. Primitive would be an understatement. Here's what we have encountered. On a straight road with no cars, AP2 appears to work satisfactorily at 45 MPH. On a road which curves leaving the vehicle driving into the sun even with human-readable lane markings that remain easy to decipher with the naked eye, the car is worse than dangerous. The car immediately zig zags from lane to lane with no realization that vehicles may be in the other lanes, even beside the car. On highways with exit ramps, AP2 always follows the exit ramp if you are in the right lane even without a right turn signal and even with clearly marked dotted lines on left side of the lane and even if following a vehicle proceeding straight ahead. AP2 still slams on the brakes when approaching almost any (stationary) overhead sign on an interstate.

I appreciate that AP1 code needed tweaking for the new hardware, but AP2 really leaves the impression that the developers started over... and have not progressed even to the level of what most would consider safe driving. There really should be more warnings and alerts regarding documented problems for those that are expecting a safer driving experience. A traffic fatality would be a huge setback for Tesla, not to mention the poor driver.

 

[Drone Flyer]

I just sent the following note to Tesla, but wanted to alert others to some of the problems we have experienced with AP2...

We have 17.3.2 and have used AutoSteer extensively on a number of interstate highways since its initial release. Primitive would be an understatement. Here's what we have encountered. On a straight road with no cars, AP2 appears to work satisfactorily at 45 MPH. On a road which curves leaving the vehicle driving into the sun even with human-readable lane markings that remain easy to decipher with the naked eye, the car is worse than dangerous. The car immediately zig zags from lane to lane with no realization that vehicles may be in the other lanes, even beside the car. On highways with exit ramps, AP2 always follows the exit ramp if you are in the right lane even without a right turn signal and even with clearly marked dotted lines on left side of the lane and even if following a vehicle proceeding straight ahead. AP2 still slams on the brakes when approaching almost any (stationary) overhead sign on an interstate.

I appreciate that AP1 code needed tweaking for the new hardware, but AP2 really leaves the impression that the developers started over... and have not progressed even to the level of what most would consider safe driving. There really should be more warnings and alerts regarding documented problems for those that are expecting a safer driving experience. A traffic fatality would be a huge setback for Tesla, not to mention the poor driver.

As an AP owner, I get a chuckle whenever I see postings concerning AP 2.
If and when it comes out in it's fullest capacity, say in 3 years, you will still have to grab the steering wheel every three minutes anyway because of forever legislation. You will still be limited to the speed limits on rural roads give or take a 10 km/hr leeway. Everyone knows we all really want it to be faster especially in autonomous mode but that won't happen.
My point is, if you're going to spend this much on a Tesla, wouldn't you want to drive the darn thing yourself?
I can see full autonomy on a mini van or truck, but not a fun to drive Tesla Model S.
Might as well hop in a cab if you don't want to drive!
The true benefit of AP to me, isn't the auto steer, but the collision avoidance features. To me, that is worth every penny.

 

[disagree]

As an AP owner, I get a chuckle whenever I see postings concerning AP 2.
If and when it comes out in it's fullest capacity, say in 3 years, you will still have to grab the steering wheel every three minutes anyway because of forever legislation. You will still be limited to the speed limits on rural roads give or take a 10 km/hr leeway. Everyone knows we all really want it to be faster especially in autonomous mode but that won't happen.
My point is, if you're going to spend this much on a Tesla, wouldn't you want to drive the darn thing yourself?
I can see full autonomy on a mini van or truck, but not a fun to drive Tesla Model S.
Might as well hop in a cab if you don't want to drive!
The true benefit of AP to me, isn't the auto steer, but the collision avoidance features. To me, that is worth every penny.

I get a chuckle when I read posts like yours. I really want to be able to drive in urban heavy traffic environments with no stress. I don't have a care at all about going fast on wide-open highways of the Great White North.

Maybe --- just maybe --- different people want different things from their vehicles and the capabilities of those vehicles.

 

[Matias]

.

but AP2 really leaves the impression that the developers started over...

The DID start over from 0. AP1 lane detection hapens in MobilEyeQ3 on proprietary hardware and software. Tesla can't use any of that.

 

[drshan]

I just sent the following note to Tesla, but wanted to alert others to some of the problems we have experienced with AP2...

We have 17.3.2 and have used AutoSteer extensively on a number of interstate highways since its initial release. Primitive would be an understatement. Here's what we have encountered. On a straight road with no cars, AP2 appears to work satisfactorily at 45 MPH. On a road which curves leaving the vehicle driving into the sun even with human-readable lane markings that remain easy to decipher with the naked eye, the car is worse than dangerous. The car immediately zig zags from lane to lane with no realization that vehicles may be in the other lanes, even beside the car. On highways with exit ramps, AP2 always follows the exit ramp if you are in the right lane even without a right turn signal and even with clearly marked dotted lines on left side of the lane and even if following a vehicle proceeding straight ahead. AP2 still slams on the brakes when approaching almost any (stationary) overhead sign on an interstate.

I appreciate that AP1 code needed tweaking for the new hardware, but AP2 really leaves the impression that the developers started over... and have not progressed even to the level of what most would consider safe driving. There really should be more warnings and alerts regarding documented problems for those that are expecting a safer driving experience. A traffic fatality would be a huge setback for Tesla, not to mention the poor driver.

Toally agree with all of your observations. I bought the vehicle after test driving an HW1 AP1. Very surprised that HW2 AP2 started at this level. I dont not believe from reading prior comments that HW1 was ever performing like this, even day 1.

 

[brucet999]

Toally agree with all of your observations. I bought the vehicle after test driving an HW1 AP1. Very surprised that HW2 AP2 started at this level. I dont not believe from reading prior comments that HW1 was ever performing like this, even day 1.

You are correct. AP1 was pretty good when it was first released, even on two-lane highways. There were, of course, problems with losing control on curves over a hilltop, "truck lust", off ramps, etc, but nothing as bad as what we are hearing from AP2 users now.

 

[Deans]

Robotic Hair Transplant

 

[BestRadar]

What I dont understand is where is the firmware used in this video:

Autopilot

Was this not an HW2 Vehicle? It seems like they are starting from scratch and it makes no sense how they had something performing very well in that video yet it seems we have taken many steps backwards.

 

[Cosmacelf]

What I dont understand is where is the firmware used in this video:

Autopilot

Was this not an HW2 Vehicle? It seems like they are starting from scratch and it makes no sense how they had something performing very well in that video yet it seems we have taken many steps backwards.

Beware Demoware. You can always make a software system look really good in a very controlled or tuned environment. Go somewhere else where the sun angle is different, or the color of the lines is different and watch out...

 

[Cosmacelf]

Elon's companies are always at the bleeding edge. Being a software guy myself, I've always thought his software teams are pushed to the limit to deliver complex software and they frequently don't get it right. I suspect trying to replicate Mobileye is turning into a bigger challenge than he originally thought. There was a good reason Mobileye went with custom silicon. The DrivePX hardware may not even be up to the task - I don't think anyone truly knows if it'll every work correctly. Having said that, Elon does have a history of delivering. Delivering late, but delivering.

This is ever green:

Sandra and Woo » [0811] The Divine Comedy, Page 9 | The comedy webcomic

 

[J1mbo]

What I dont understand is where is the firmware used in this video:

Autopilot

Was this not an HW2 Vehicle? It seems like they are starting from scratch and it makes no sense how they had something performing very well in that video yet it seems we have taken many steps backwards.

What they seem to be doing right now is trying to make the system AP1 equivalent. 1 camera, radar, ultrasonics. No evidence that the other cameras or more advanced software features provided by NVIDIA (which were used in that video) are active in AP2 cars yet.

 

[Blazing Whisper]

As an AP owner, I get a chuckle whenever I see postings concerning AP 2.
If and when it comes out in it's fullest capacity, say in 3 years, you will still have to grab the steering wheel every three minutes anyway because of forever legislation. You will still be limited to the speed limits on rural roads give or take a 10 km/hr leeway. Everyone knows we all really want it to be faster especially in autonomous mode but that won't happen.
My point is, if you're going to spend this much on a Tesla, wouldn't you want to drive the darn thing yourself?
I can see full autonomy on a mini van or truck, but not a fun to drive Tesla Model S.
Might as well hop in a cab if you don't want to drive!
The true benefit of AP to me, isn't the auto steer, but the collision avoidance features. To me, that is worth every penny.

I respectfully disagree with you. While it's absolutely fun to drive the car yourself, Autopilot is a huge selling point as well for the 99% of the time that you drive regular routes, e.g. for your work commute. Sometimes you drive for fun, but most of the time you're driving to get somewhere and it's boring as hell. The problems being reported on AP2 are scary... I fear that its only a matter of time before someone gets hurt and Tesla will suffer some serious setbacks. I'm nervous that it was launched prematurely.

 

[NerdUno]

As an AP owner, I get a chuckle whenever I see postings concerning AP 2.

My point is, if you're going to spend this much on a Tesla, wouldn't you want to drive the darn thing yourself?

I fully expect to get the features that I paid an $8,000 premium for. If I hadn't wanted them, I wouldn't have purchased them.

 

[Saghost]

I appreciate that AP1 code needed tweaking for the new hardware, but AP2 really leaves the impression that the developers started over... and have not progressed even to the level of what most would consider safe driving. There really should be more warnings and alerts regarding documented problems for those that are expecting a safer driving experience. A traffic fatality would be a huge setback for Tesla, not to mention the poor driver.

No tweaking possible. AP1 was built on top of Mobileye's proprietary hardware and software, which they no longer sell to Tesla, as Matias noted. In particular, all of the object recognition (going from a black and white (and red) image to a set of lane lines, cars, and pedestrians) was handled by a neural network Mobileye spent a decade training.

That's the part Tesla is working hard to replace on NVidia hardware, and apparently having more trouble doing than they expected. Once they get the object recognition sorted out, the rest of the AP1 code can be (presumably has been) carried over.

 

[drshan]

No tweaking possible. AP1 was built on top of Mobileye's proprietary hardware and software, which they no longer sell to Tesla, as Matias noted. In particular, all of the object recognition (going from a black and white (and red) image to a set of lane lines, cars, and pedestrians) was handled by a neural network Mobileye spent a decade training.

That's the part Tesla is working hard to replace on NVidia hardware, and apparently having more trouble doing than they expected. Once they get the object recognition sorted out, the rest of the AP1 code can be (presumably has been) carried over.

Disappointing. Certainly not what was advertised. My advisor told me that I could drive the car home from Cleveland (pick up location) to Pittsburgh on full autopilot in Late December when I placed the order. Now I am in serious doubt of HW1 parity in the next 6 mo to be honest.

 

[Saghost]

Disappointing. Certainly not what was advertised. My advisor told me that I could drive the car home from Cleveland (pick up location) to Pittsburgh on full autopilot in Late December when I placed the order. Now I am in serious doubt of HW1 parity in the next 6 mo to be honest.

I dunno. I'm confident Tesla will get there, and they seem confident that they are close, but it may take them a while.

 

[MXWing]

Droneflyer got a slap on the wrist with only one dislike!

I may consider the fastest M3 without autonomy features just to bring the cost down. We can't discount other peoples expectations though, especially if they slapped down 8k expecting something.

Good thing is Tesla I'm sure will make good on their promise -eventually-. My Volt 2017 "autonomy" was broken from day 1 and will remain forever broken.

The best thing for Tesla to do here is to offer discounts for those who buy in early and charge more for later adopters. They already do this in a way, but changing the message will make a lot of people happier and set expectations accordingly.

 

[182RG]

and they seem confident that they are close

EM thinks he is always close, or just lies about it. It's how he keeps the faithful fed.....

 

[182RG]

handled by a neural network Mobileye spent a decade training.

Meanwhile, Audi, Volvo, MB, and others appear to be at, or better than AP2 in its current state. Tesla is going to quickly lose their edge. I'd like to test drive the '17 A4 with Traffic Jam Assist.

 

[Cosmacelf]

Meanwhile, Audi, Volvo, MB, and others appear to be at, or better than AP2 in its current state. Tesla is going to quickly lose their edge. I'd like to test drive the '17 A4 with Traffic Jam Assist.

Yes! When you think about it, Tesla went from being years ahead of competitors to being behind them for the last 5 months.