8.0 (2.50.185) caution using TACC/Autosteer features

[S100D]

Probably because it's not fully tested. There is one thing worse than not stopping for stationary vehicles, and that is randomly stopping for stationary vehicles. The system has to work flawlessly and predictably. If it stops for stationary objects most of the time, then drivers become complacent and it's a bigger problem then.

Please do not tell people that the Tesla will stop for stationary objects most of the time. The only time a Tesla will stop for stationary objects is after you run into the stationary object and have destroyed your car. The current version of the software does not have Automatic Emergency Braking and will never stop for stationary object.

As Tesla sends out updates for AP2 car, it will eventually have Automatic Emergency Braking, and an Auto Pilot that will stop for stationary objects. Hopefully this future version of software will allow the AP1 cars to have the same capability.

 

[dhanson865]

What happens when I have enabled TACC, driving under 35MPH and eventual run into stop go traffic. Would it break then?

It's all about the time of acquisition. If the object in front of the car is moving when AP first detects it then it gets flagged as something that can move. Those objects it will stop for because it has high confidence that it needs to. If it was at a stop when the AP first detects it then it gets flagged as stationary and AP won't stop for it. For those objects AP is depending on the human behind the wheel to confirm or deny that they are in the path and are an issue. You have to stop for those because AP isn't sure, just too many possible false positives for the system to stop every time it heads towards a stationary object.

 

[vandacca]

Please do not tell people that the Tesla will stop for stationary objects most of the time. The only time a Tesla will stop for stationary objects is after you run into the stationary object and have destroyed your car. The current version of the software does not have Automatic Emergency Braking and will never stop for stationary object.

As Tesla sends out updates for AP2 car, it will eventually have Automatic Emergency Braking, and an Auto Pilot that will stop for stationary objects. Hopefully this future version of software will allow the AP1 cars to have the same capability.

I'm not saying anything of the sort (please re-read my post). The point of my last post was that the software in the full-self-driving video wasn't ready for general release.

 

[NerdUno]

So the core problem of Tesla not breaking into stationary objects with TACC is that it can't deterministically identity if there a stationary vehicle in front of it.

Am guessing that is where having the system learn to identify cars from the tri focal cameras and Radar data processing comes in and eventually training the model to learn identifying the cars would help and perhaps that is why few million miles are needed before it happens.

But how did the AP2 demo video stop for stopped vehicles in front of it, clearly the code and data model exists, then why not put that in beta with this update.

Alternatively they could provide a visual cue notifying the driver that the car is not going to break and one should apply the breaks as an object is detected in front. Or is that FCW?

Just curious. We've had license plate readers for more than a decade. Almost every vehicle has a license plate on the back. Can the Tesla not decipher the difference in an overpass and a vehicle by looking for a license plate? They're all the same size and generally within a defined area on the back of a vehicle. I'm not suggesting it would be perfect, but it certainly would handle 99% of the stationary vehicles on the road.

 

[MXWing]

Just curious. We've had license plate readers for more than a decade. Almost every vehicle has a license plate on the back. Can the Tesla not decipher the difference in an overpass and a vehicle by looking for a license plate? They're all the same size and generally within a defined area on the back of a vehicle. I'm not suggesting it would be perfect, but it certainly would handle 99% of the stationary vehicles on the road.

Yes, it seems so obvious, that license plates could be used to identify cars. If the cameras can read light spectrums, brake lights in conjunction with license plate is like 99.99 confidence?

 

[Hmach]

I have had less than 50 miles with the new software.

The TACC appears to be working okay for me so far without any safety issues. I agree with others that the distance feels closer (in either time and/or distance) than my 7 car length setting. I drive in stop and go downtown traffic. Constant merging at low speeds. TACC slows and stops for merging cars but the merging car does not show on the dash until they are more than 50% into my lane. Also does not keep as much space as I would like for the merging car. New to AP so I don't know if this last issue is normal relative to AP1.

I also have forward collision setting on medium. I have not yet gotten a forward collision warning. Thank gosh no red cars.

On the emergency brake discussion, it is anticipated that the fully released AP2 will allow for this functionality, right? Another way, is whether the vehicle stationary or matched to the TACC prior to becoming a stationary object only a testing concern which will be finished at some point or an inherent limitation?

 

[S100D]

I have had less than 50 miles with the new software.

On the emergency brake discussion, it is anticipated that the fully released AP2 will allow for this functionality, right? Another way, is whether the vehicle stationary or matched to the TACC prior to becoming a stationary object only a testing concern which will be finished at some point or an inherent limitation?

Hmach, yes the AP2 software when fully release will be able to stop for stationary objects, people, animals, etc. It has to be able to do this for full autonomous driving. If you watch the Tesla autonomous driving video, you see that it stops for stop lights, stop signs, people, etc. This includes full automatic emergency braking.

 

[thimel]

I have had less than 50 miles with the new software.

The TACC appears to be working okay for me so far without any safety issues. I agree with others that the distance feels closer (in either time and/or distance) than my 7 car length setting. I drive in stop and go downtown traffic. Constant merging at low speeds. TACC slows and stops for merging cars but the merging car does not show on the dash until they are more than 50% into my lane. Also does not keep as much space as I would like for the merging car. New to AP so I don't know if this last issue is normal relative to AP1.

I also have forward collision setting on medium. I have not yet gotten a forward collision warning. Thank gosh no red cars.

On the emergency brake discussion, it is anticipated that the fully released AP2 will allow for this functionality, right? Another way, is whether the vehicle stationary or matched to the TACC prior to becoming a stationary object only a testing concern which will be finished at some point or an inherent limitation?

In AP1, cars in adjacent lanes are shown. When one merges into my lane, TACC starts slowing very shortly after the car crosses the line. This is still a bit later than I would do it, but is is pretty good. TACC used to respond much more slowly, it has improved over time.

 

[LastGas]

And here's one of a Volvo that didn't do as well:

Was that Korean guy driving?

 

[LastGas]

LastGas, if you try the following scenario I think you will find out that you will destroy your car.

Turn on the autopilot on AP1 and find a car at a stoplight that is about 300 feet away. Your car cannot see the car stopped at the stoplight since it is too far away. Since you were not following the car when it stopped at the stoplight, you Tesla will not stop and it will run into the back of the car at the stop light. ....

This has not been my experience. My car stopped correctly in this scenario at least twice this morning. As I said in my original comment, it doesn't work 100% of the time, and the driver must be ready to intervene in any Autopilot scenario, but in my experience it stops most of the time.

 

[Hmach]

@S85D and @thimel thanks for the info. I couldn't imagine driving with AP that won't stop for you in an emergency (yes, even though it is only a driver assistance device).

Based on AP1, and perhaps obviously, AP2 lane markings and car identification are still basic. The cars I do see on the console are hopping all over the place, in and out of lanes, and disappearing and reappearing. It also does not show the cars in front or next to me consistently or accurately. I assume we are just data gathering for the HW2 and this has nothing to do with actual use by the driver.

 

[S4WRXTTCS]

So the core problem of Tesla not breaking into stationary objects with TACC is that it can't deterministically identity if there a stationary vehicle in front of it.

Am guessing that is where having the system learn to identify cars from the tri focal cameras and Radar data processing comes in and eventually training the model to learn identifying the cars would help and perhaps that is why few million miles are needed before it happens.

But how did the AP2 demo video stop for stopped vehicles in front of it, clearly the code and data model exists, then why not put that in beta with this update.

Alternatively they could provide a visual cue notifying the driver that the car is not going to break and one should apply the breaks as an object is detected in front. Or is that FCW?

To me it all comes down to the acceptance of false positives. On the video demonstration the code being used probably generates a crap ton of false positives. Where it couldn't be used with this update due to how unacceptable that would be.

They probably figured the FCW would act as a bit of a stand in. The FCW isn't supposed to stop the car, but just generate a warning.

 

[chillaban]

To me it all comes down to the acceptance of false positives. On the video demonstration the code being used probably generates a crap ton of false positives. Where it couldn't be used with this update due to how unacceptable that would be.

They probably figured the FCW would act as a bit of a stand in. The FCW isn't supposed to stop the car, but just generate a warning.

While visiting family, I was driving a rental low-end GM vehicle with a camera based FCW system.... Over the course of 200 miles I had around 10 FCW alerts in snowy weather. Maybe 1 of them was remotely legitimate -- I was approaching a much slower car 5 car distances away and was planning on taking action later. The rest of them had nothing remotely threatening when the car suddenly started beeping.

An hour into the drive I was ignoring most of the alerts. Harass a user too many times with false FCW alerts and they will not pay attention that one time when a split second reaction will count.

 

[3Victoria]

This is my educated guess at how things stand, take them with a (large) grain of salt:

AP1 uses MobilEye to see objects, a simple linear radar, and processing that integrates and processes the sensor data, and drives the car. The Florida crash was caused by not detecting the truck, it being a stationary object in the path of the car. The MobilEye chip failed to see the truck likely because of contrast problems. The radar also failed to 'see' it. This is because the linear radar sends a pulse and reads the returning reflected signal. The late part of the reflected signal represents all stationary things in front of the car, including: signs, lamp posts, potholes, fences, and, unfortunately large trucks crossing perpendicularly to the direction of the car. If the car braked every time it saw a stationary item, you wouldn't get very far. So, in the truck crash, the radar 'saw' the truck but ignored it, just like it ignores all stationary objects.

Tesla negotiated with the radar manufacturer to obtain the raw feed from the radar, and this lets them 'see' in a more 3D manner, since the radar scans the space in front, and Tesla can compare serial frames of data and so build a 'point cloud' (this is similar to what Lidar does). This likely lets then detect large stationary objects (like a truck). There can still be false positives for stationary objects. However, Tesla plans to map these, so they can ignore persistent stationary items (like signs, overpasses, etc.), but still detect intermittent stationary items (like people, cars, etc.). However, this requires them to build a map of persistent items. They do this by mapping will cars repeatedly pass by the persistent stationary items to confirm they are really persistent.

AP2 is a completely new system that has the potential to be much more. However, since it uses wholly new hardware/sensors, and it uses a new technique for processing (deep neural nets), it needs to be developed from scratch. However, Tesla is now releasing the AI model for functionality close to AP1. The system still has to learn to calibrate the cameras, map the world, etc. so it can eliminate false positives, similarly to AP1. The system will be periodically extended in functionality.

The demo system shown in the video shows the *potential* and is *not* promised yet. Its internal model is likely not complete, and is not validated, and so cannot be released, yet. An analogy: consider AP2 to be a primer book, written, edited, well vetted, and close to being ready to be published. The demo is a university level text-book - much more detailed and complex, but still in the writing and editing phase, and not ready to be published. [[ Caveat: all analogies fail to convey the whole story. ]]

 

[Skalinator]

Yes, it seems so obvious, that license plates could be used to identify cars. If the cameras can read light spectrums, brake lights in conjunction with license plate is like 99.99 confidence?

I created a thread for AP2.0 owners to log experiences, in somewhat of a format. Come log yours

AP 2.0 - Lets log our experiences

 

[Andyw2100]

I created a thread for AP2.0 owners to log experiences, in somewhat of a format. Come log yours

AP 2.0 - Lets log our experiences

And while you're at it, it would be great if you could also log your firmware version at @HankLloydRight's excellent Firmware Upgrade Tracker (http://ev-fw.com.)

As of right now, there are only six cars that have logged their update to version 8.0.2.50.185. There must be a lot more of you than that!

 

[Skalinator]

And while you're at it, it would be great if you could also log your firmware version at @HankLloydRight's excellent Firmware Upgrade Tracker (http://ev-fw.com.)

As of right now, there are only six cars that have logged their update to version 8.0.2.50.185. There must be a lot more of you than that!

Will do!

 

[vandacca]

This is my educated guess at how things stand, take them with a (large) grain of salt:

AP1 uses MobilEye to see objects, a simple linear radar, and processing that integrates and processes the sensor data, and drives the car. The Florida crash was caused by not detecting the truck, it being a stationary object in the path of the car. The MobilEye chip failed to see the truck likely because of contrast problems. The radar also failed to 'see' it. This is because the linear radar sends a pulse and reads the returning reflected signal. The late part of the reflected signal represents all stationary things in front of the car, including: signs, lamp posts, potholes, fences, and, unfortunately large trucks crossing perpendicularly to the direction of the car. If the car braked every time it saw a stationary item, you wouldn't get very far. So, in the truck crash, the radar 'saw' the truck but ignored it, just like it ignores all stationary objects.

Tesla negotiated with the radar manufacturer to obtain the raw feed from the radar, and this lets them 'see' in a more 3D manner, since the radar scans the space in front, and Tesla can compare serial frames of data and so build a 'point cloud' (this is similar to what Lidar does). This likely lets then detect large stationary objects (like a truck). There can still be false positives for stationary objects. However, Tesla plans to map these, so they can ignore persistent stationary items (like signs, overpasses, etc.), but still detect intermittent stationary items (like people, cars, etc.). However, this requires them to build a map of persistent items. They do this by mapping will cars repeatedly pass by the persistent stationary items to confirm they are really persistent.

AP2 is a completely new system that has the potential to be much more. However, since it uses wholly new hardware/sensors, and it uses a new technique for processing (deep neural nets), it needs to be developed from scratch. However, Tesla is now releasing the AI model for functionality close to AP1. The system still has to learn to calibrate the cameras, map the world, etc. so it can eliminate false positives, similarly to AP1. The system will be periodically extended in functionality.

The demo system shown in the video shows the *potential* and is *not* promised yet. Its internal model is likely not complete, and is not validated, and so cannot be released, yet. An analogy: consider AP2 to be a primer book, written, edited, well vetted, and close to being ready to be published. The demo is a university level text-book - much more detailed and complex, but still in the writing and editing phase, and not ready to be published. [[ Caveat: all analogies fail to convey the whole story. ]]

Thanks @3Victoria for the post. I'm not trying to be pedantic, but I just wanted to correct a few points that I thought are important in the understanding of AP.

• It was my understanding that the Florida tractor trailer was in fact moving across the path of the Model-S owner. So technically, it was moving in a perpendicular path (truck was on a cross-street)
• Not sure how the MobilEye hardware failed, but it could have been due to the truck's high stance (radar could see under truck) and therefore mistake it for an overhead sign. If the truck had a aerodynamic skirt, it probably would have been registered by the radar.
• When radar is reflected, the early signals represent objects that are close by and later signals represents objects that are further away. Large signals represent large objects and small signals represent small objects.

I'm in agreement with pretty much everything else you wrote. When Tesla has completed the "permanent stationary object" map of the highway system, make sure you don't break down under one of those stationary signs.

 

[DrivingTheFuture]

Since our most recent update I have noticed our X when on autopilot will automatically slow down suddenly from 70 to 50 (or 55?) Whenever it detects a car stopped on the side of the highway, this was very disconcerting the first time and is still annoying however it may be a good safety feature. I believe it may have been put in place because of the real first AP death in China (front corner impact into road cleaning truck) which was never officially confirmed but from video appears to be on AP (Josh Brown death was first publicly acknowledged by Tesla?)

 

[chillaban]

I've noticed the same. If passing a stopped vehicle on the shoulder, the car slows down in 2.50.114, to about 50mph and somewhat briskly. It feels roughly equivalent to full regen.

I've also noticed that it reacts in much the same way if it thinks a car in the adjacent lane veered into your lane, even just barely straddling the lane marker.


So far, I haven't perceived a situation where the slowdown rate is "dangerous", but it does seem like it could be a good safety feature once they fine tune exactly when it's activated. Some of the slowdowns I've experienced seem largely unnecessary, but if deployed properly, it definitely grabs your attention and makes you look at the road (if you were zoning out)