Firmware 8.0

[jgs]

Here's a new "feature" that I just discovered in 8.0 that limits AP even further:

On undivided highways where AP limits TACC speed to the speed limit + 5mph, engaging Autosteer now only allows you to set TACC's max speed to 5mph over the limit.

In 7.1, you could set a higher speed (e.g. 50mph in a 25mph zone) and engaging Autosteer limited the car's speed to 30mph, but this didn't change the 50mph max speed. Then once the speed limit was lifted to a higher number, the car would automatically accelerate to the new limit (5mph over, or 50mph, whichever was lower).

This was useful in situations such as temporarily driving through a town with a low speed limit. However if you now engage full AP (TACC + autosteer) while the limit is low (say 25mph), the max speed TACC will get to is 30mph. Pulling the speed stall up won't allow you to go higher than 30.

So after the speed limit changes to a higher number, the car will only go up to 30mph even though the new limit might be much higher, such as 55mph.

I haven't tested the situation where AP is engaged with a higher speed and a lower limit is encountered to see if the max TACC speed automatically lowers to the new limit + 5mph, or stays at the original TACC speed. I would assume (hope) it doesn't change down, but as it is this new "feature" seems like another step in the lawyer-mandated CYA direction..

From my point of view, this is a feature. Actually, it's a bug fix. In 7.1 I had a few situations where the car started to accelerate wildly in a situation such as you describe. Nothing bad happened but I was NOT well pleased.

In my view this just brings it into the general paradigm that if the speed limit changes, you have to change your TACC setpoint to match. If they ever rejigger TACC to have it autonomously raise its setpoint, that would be a different story.

 

[sillydriver]

Ok I feel l need to log my AP 8.0 experience here. Apologies in advance for the long post.

I have had my MS90D for 3 months..and logged just under 3,000 miles. I have absolutely raved about this car and (at least until 8.0AP) have loved nearly every minute in it...coming off back-to-back leased 750is and a number of other similar vehicles prior to those.

Last week I did a ~1,000 mile road trip...something I would generally not do...especially as a 20 year instrument pilot (yes, I know how real autopilots work). But I love driving the car and AP takes the overall workload and related stress of the drive down a lot, so I decided to drive it.

I did the outbound leg on the latest 7.1. It was AWESOME. Pretty solid tracking, NO scares whatsoever...and very reasonable 'nags' (in fact only a couple the whole trip). Absolute best long distance drive I've ever experienced.

I was pleased to get 8.0 during my stay and was looking forward to putting it through it's paces on the way back.

Well... Let's just say that I thought of two new bumper stickers on the way home: "My Tesla Loves Semi's (so much it wants to kiss them)" and "I'm not drunk, I'm on Autopilot 8.0." I literally thought that if it didn't throw itself (and me) under a semi, it was going to at least get me pulled over. Really very disappointing.

(I'm not going to go into the 'nag' thing. I get that it's 'beta'...and that you'r'e supposed to have your hands on the wheel at all times, etc. etc. I WILL argue the point that you should not be penalized for looking at the road (and not the dash) and leave it at that. Overall my experience on that front hasn't been 'terrible', but it is a decided step backward for me.)

The big issue for me is tracking, especially passing semis. Unlike many people experiencing 'better lane tracking', I have experienced quite the opposite, and it sounds like others on this thread have as well. And I have talked to other friends, including one on his second MS, who have had similar scary issues.

Before 8.0AP I only had a couple (maybe 'several') instances where I needed to quickly take over from a situation where I felt like the car was about to get me into serious trouble. On the way home it happened numerous times. Not tracking well, weaving back and forth/'seeking', hard swerves...especially cresting hills...if you can even call them that (this trip was between florida and atlanta). I did get one minor 'phantom' brake incident, but I didn't register it as a big deal.

The real dangerous thing was it pulling hard into many (at least half) of the semis I passed. Some trucks induced the behavior and some didn't...I have to guess it is related to aerodynamics/buffeting of the particular truck and how AP is trying to deal with that. I only pass semis on the left...so this was always on that side. I assume the behavior would be the same on the right.

Sometimes it felt like an increasing oscillation, others it just felt like an abrupt pull to the right. What was even scarier is that it really felt like the side sonar wasn't doing (or going to do) anything about it. Perhaps it would have, but I didn't feel like letting it go far enough to find out.

Sometimes I could hold it off with SOLID left pressure on the wheel (right up to the edge of disengaging). Other times I had to force off AP, a couple times pretty aggressively. On 7.1 I do recall it often hugging the lane next to a semi closer than I liked, but I don't recall it ever pulling into one anywhere close to this aggressively.

Beyond that, I still feel like AP isn't tracking as well as it was before, even in stop and go. My wife commented on it during a drive last night as well. I just engaged it going to lunch and though it was going to sideswipe the car next to me.

I DO feel it handles AP accel/decel/braking better than before.

I love my car and I support Tesla and what they are doing on a number of fronts. I am not bashing at all here..and I am sure as heck not making it up. I love how they push forward and I am willing to fully understand the limitations of my vehicle's systems as 'Pilot in Command'. I am holding out faith in their agile development processes, OTA updates, and fleet-learning capabilities to get this at least back to where it was before…hopefully quickly. I hope that's not a naive position.

At the moment, at least as it relates to AP, I wish I was still on 7.1.

I am sure that others have had different/better experiences, but that is my experience...please don't flame me for sharing it. I'm just trying to be helpful and add to the dialogue and the record.

Yet another good and detailed post -- I've probably seen about 15 in various threads -- about bad or even dangerous autosteer behavior. On the plus side reports show the TACC side of AP in 8.0 seems to be working better. I think some have speculated that the autosteer problem stems from fleet learning having to start over from the beginning, given the software changes.

Here's my theory. While radar has become a primary input for braking and obstacle avoidance in TACC, its role in the auto steer part of AP has also increased: then the car can steer away from large objects or stopped cars in the road. The problem is that incorporating radar data into autosteer requires a significant revision of the code (not just the training set) and because Tesla has curtailed its relationship with MobilEye, Tesla has had to come up with the new autosteer code for the EyeQ3 hardware in existing cars all by themselves, without Mobileye's help. This means that Tesla has essentially had to redevelop the autosteer code in 8.0 from square one, meaning their optical lane-following software is the company's first in-house attempt, and is software that does not build on MobilEye's experience or code base, or at least not to the extent that a new release done in partnership with MobilEye would. Thus the quality of 8.0 optical lane-following is unfortunately worse than in the original 7.0 release of AP. This is bad news for users but is the kind of reset that is necessary if Tesla is going to do all of its future AP code in-house.

My decision to buy an MS predated the introduction of AP, I love my car for all the qualities that this truly great BEV has, and I wouldn't trade it for any other car. However, the state of autosteer seems to be pretty bad in release 8.0.

 

[⚡️ELECTROMAN⚡️]

It's kind of funny they raised the price and then lowered the quality of AP.

 

[Stoneymonster]

Yet another good and detailed post -- I've probably seen about 15 in various threads -- about bad or even dangerous autosteer behavior. On the plus side reports show the TACC side of AP in 8.0 seems to be working better. I think some have speculated that the autosteer problem stems from fleet learning having to start over from the beginning, given the software changes.

Here's my theory. While radar has become a primary input for braking and obstacle avoidance in TACC, its role in the auto steer part of AP has also increased: then the car can steer away from large objects or stopped cars in the road. The problem is that incorporating radar data into autosteer requires a significant revision of the code (not just the training set) and because Tesla has curtailed its relationship with MobilEye, Tesla has had to come up with the new autosteer code for the EyeQ3 hardware in existing cars all by themselves, without Mobileye's help. This means that Tesla has essentially had to redevelop the autosteer code in 8.0 from square one, meaning their optical lane-following software is the company's first in-house attempt, and is software that does not build on MobilEye's experience or code base, or at least not to the extent that a new release done in partnership with MobilEye would. Thus the quality of 8.0 optical lane-following is unfortunately worse than in the original 7.0 release of AP. This is bad news for users but is the kind of reset that is necessary if Tesla is going to do all of its future AP code in-house.

My decision to buy an MS predated the introduction of AP, I love my car for all the qualities that this truly great BEV has, and I wouldn't trade it for any other car. However, the state of autosteer seems to be pretty bad in release 8.0.

Except that Tesla has made it clear for a long time that AP software is theirs, running on MobilEye HW. This is not the explanation for the perceived (which is NOT universal) change in quality.

 

[Duma]

View attachment 197434

Back to 8.0 navigation issues. Here's a shot of my nav screen when I'm actually 3miles, not 10, from home and less than 10 minutes, not 29, away. I just need to go one mile and take a right. All of the route shown that looks like the head a T-Rex in not needed. In 7.1 it also refused to take the shortest or quickest route. It picked a different, but equally flawed route. Is there anything I can do?

The whole navigation system is a bit of a kludge. There are two pieces. One is the Map app that appears on the console and uses map data from Google. While this is called "Navigation," the only input to the navigation is selecting the end point (start point is current location) and any intermediate super charging stops. The other part is embedded navigation software from Navigon (Garmin subsidiary) which uses onboard maps. The actual route is computed by Navigon so that this works even if there is no Internet connectivity from the car. The Navigon selected route is what gets displayed back in the Map app.

The good part of doing endpoint selection using the Google data in the Map app is that it takes full advantage of the context sensitive address entry, including landmarks, that Google has perfected over the years. When you do not have Internet connectivity the interface falls back to the native Navigon function where you have to enter all the details (street number, name, city, state). The bad part is that you expect Google quality routing but what you get is Navigon, which may be even a cut worse than Garmin. (Speculation: since Navigon is the embedded system by multiple car manufacturers, this may be Garmin's way of preserving its market for stand alone GPS units.) Moreover, updates to the onboard map data are limited to the approximately once a year update via Tesla from Navigon. (Unlike Garmin, which is quarterly and Google which is continuous.)

I also suspect that the traffic aware routing is based on connecting to Garmin's traffic data rather than Googles. Ideally both should reflect the same reality, but this could be affected by the data sources available to each of them.

Finally, I think that the updated "Navigation software" in 8.0 refers only to Map App that appears on the console. I suspect that the embedded Navigon software has not been updated and so the underlying navigation flaws will remain. Seems consistent with the evidence.

While we can and should complain to Tesla about navigation problems, but realistically all Tesla can do is lean on Navigon for improvements or switch to another embedded navigation vendor. Not sure who that might be or if they would be any better. Unlikely that Tesla would pull this in house since is it does not (yet) seem part of the core mission (EV, batteries, and getting to autonomous driving).

 

[cynix]

I received update notification at this morning. Will do it after work.

Wow! Is this the first update outside North America?

 

[Cyclone]

This was required to solve the issue where the car accelerates wildly if you disengaged autosteer only. One could argue about the holistic solution, but this is necessary given what the rest of the interface is.

I agree with you assessment. But they simply could have made disengaging autosteer into TACC simply issue a "set speed" to the current speed. Then the side effects of always limiting wouldn't occur.

 

[msnow]

The whole navigation system is a bit of a kludge. There are two pieces. One is the Map app that appears on the console and uses map data from Google. While this is called "Navigation," the only input to the navigation is selecting the end point (start point is current location) and any intermediate super charging stops. The other part is embedded navigation software from Navigon (Garmin subsidiary) which uses onboard maps. The actual route is computed by Navigon so that this works even if there is no Internet connectivity from the car. The Navigon selected route is what gets displayed back in the Map app.

The good part of doing endpoint selection using the Google data in the Map app is that it takes full advantage of the context sensitive address entry, including landmarks, that Google has perfected over the years. When you do not have Internet connectivity the interface falls back to the native Navigon function where you have to enter all the details (street number, name, city, state). The bad part is that you expect Google quality routing but what you get is Navigon, which may be even a cut worse than Garmin. (Speculation: since Navigon is the embedded system by multiple car manufacturers, this may be Garmin's way of preserving its market for stand alone GPS units.) Moreover, updates to the onboard map data are limited to the approximately once a year update via Tesla from Navigon. (Unlike Garmin, which is quarterly and Google which is continuous.)

I also suspect that the traffic aware routing is based on connecting to Garmin's traffic data rather than Googles. Ideally both should reflect the same reality, but this could be affected by the data sources available to each of them.

Finally, I think that the updated "Navigation software" in 8.0 refers only to Map App that appears on the console. I suspect that the embedded Navigon software has not been updated and so the underlying navigation flaws will remain. Seems consistent with the evidence.

While we can and should complain to Tesla about navigation problems, but realistically all Tesla can do is lean on Navigon for improvements or switch to another embedded navigation vendor. Not sure who that might be or if they would be any better. Unlikely that Tesla would pull this in house since is it does not (yet) seem part of the core mission (EV, batteries, and getting to autonomous driving).

"When you do not have Internet connectivity the interface falls back to the native Navigon function" - no it always uses Navigon for turn by turn not just when you lose connectivity. I have noticed subtle new Navigon features such as an arrow pointing to your destination when you get close plus the graphics of the display so Navigon has been updated as well (not that it's better just newer). Tesla needs to stop leaning on or passing the buck or blaming Navigon and just fix the damn thing. This is an important driving tool and as owners of a premium car we deserve better.

 

[S4WRXTTCS]

The real dangerous thing was it pulling hard into many (at least half) of the semis I passed. Some trucks induced the behavior and some didn't...I have to guess it is related to aerodynamics/buffeting of the particular truck and how AP is trying to deal with that. I only pass semis on the left...so this was always on that side. I assume the behavior would be the same on the right.

Sometimes it felt like an increasing oscillation, others it just felt like an abrupt pull to the right. What was even scarier is that it really felt like the side sonar wasn't doing (or going to do) anything about it. Perhaps it would have, but I didn't feel like letting it go far enough to find out.

I seriously wonder if the AP tiles aren't being deleted during an update. What you're describing is identical to what happened to my car when I took it on a long drive RIGHT after doing one of the 7.1 updates. At the time other people noticed AP acting stupid as well. There was some talk about re-calibration, but nothing came from that discussion. Then after awhile I stopped experiencing the truck lust. I simply assumed it was fairly rare, and that the circumstances just weren't right for it.

But, now I see your experience with 8.0 which is WAY different than my experience. The difference between the two experiences is you updated in the middle of a trip, and I updated approximately a week before my trip. On my trip I had zero truck lust incidences in a 400 mile trip along a freeway heavily traveled by semi's.

FYI - The side monitoring is done by ultrasonic sensors, and they have issues detecting trucks because trucks sit so high. They can detect the wheels, etc. But, not the middle of the trailer when it's right next to you.

 

[Ingineer]

Most definitely. VxWorks embedded system ran for 1087 days in the longest stretch to date. Zero memory leaks. Highly reliable systems can be coded, it's just a matter of putting in the effort. Jet engine FADEC, nuclear plants, space shuttle avionics, Amazon cloud, are all good examples.

I'll agree with you that having thousands of users banging on different software configurations presents a challenge. Code can be instrumented to track down system resets. Tesla already gets data from the cars. I would be surprised if they didn't routinely collect software failure logs.

RT

You can't compare a high level multithreaded OS like Linux running hundreds of processes to a simple embedded system such as in FADEC. When a system has to be critically safe, they almost never use a complex OS, but something simple that's easy to audit and predict. The ABS controller in the car is a good example of this. They typically have 2 MCU's cross-checking each other, and running a simple non-multithreaded codebase.

Yeah, I've had Linux servers with uptimes of close to a year, but they are usually running only a few services that have long history and are coded relatively solid. The Tesla CID is a mess. There are many many services running and a huge amount of GUI stuff using QT all on the same box. In my experience, any modern system running a complex GUI needs rebooting somewhat often to maintain performance.

 

[Apricot]

About to go on a 900 mile road trip, so I may have more thoughts on 8.0 AP.

But given its existing limitations from 7.x -- flaky in intersections and off-ramps, squirrely on hill crests, drifts out on curves, stays too close to cars in next lane -- I've found it to be quite solid. And in general TACC has been 100% accurate, it's just been autosteer that I've had to override.

Today I noticed something that didn't happen in 7.1 for me: when motorcycles passed by closely on the left, the car actually shifted over to the right side of the lane. This happened multiple times and was welcome. I'm eager for the car to also start leaning away from next lane cars in the future...

 

[RBowen]

Just picked up my 75D last week, and I am trying to get used to AP. Please help me understand what happened tonight. I was traveling east on I8 in San Diego going 60-65 in the slow lane. I activated AP and the car started accelerating wildly scaring my wife. I braked to exit autopilot, waited a few seconds, and reactivated AP and did not get any acceleration. However, when the left lane marker widened because of an on ramp to the right, the car started drifting to the right and I had to take over. Not impressed. Am I missing something? Thanks for any feedback.

 

[evp]

I discovered another feature that was stolen from us by 8.0. I've always been irritated that I couldn't instruct voice command to "navigate to <member of contact list>". The workaround that I tolerated was to say "call <friend>"; the contact info for <friend> would appear, and I'd touch their displayed address to begin navigation. In 8.0 the interaction with voice command all happens on the small screen, and we never see the full contact info. So now I have to attempt to use the microscopic A-Z buttons on the contact list, then scroll thru the list looking for the correct contact. While driving. I have close to 1000 contacts, this is just an accident waiting to happen.

 

[K-MTG]

They should make one universal search menu like Apple spotlight, it will make life so much easier. (Navigation, music, phone, google, etc

 

[Wanderer]

So in summary, AP is more human-like except when it veers right or slams on the brakes. And the media player is more difficult to navigate, and is crippled for some.

Anyone enjoying the new voice commands at least ?? Because I'm really starting to wonder if I want this. My car is damn near perfect right now...

I am with you. I do not really see any reason to upgrade. I am very happy with 7.1 and there seem to be some significant shortcomings to 8.0 at the moment, including some very dangerous bugs. I am going to sit this one out a while longer.

 

[takeshiken]

Wow! Is this the first update outside North America?

Update finishied. It's 7.1 (2.36.31)
Sorry for misunderstanding post.

 

[BertL]

...I'm eager for the car to also start leaning away from next lane cars in the future...

As reported upthread, I had mine do this with 8.0 a couple of times last week. I think you'll be pleasantly surprised with this new automated response.

 

[BertL]

I discovered another feature that was stolen from us by 8.0. I've always been irritated that I couldn't instruct voice command to "navigate to <member of contact list>". The workaround that I tolerated was to say "call <friend>"; the contact info for <friend> would appear, and I'd touch their displayed address to begin navigation. In 8.0 the interaction with voice command all happens on the small screen, and we never see the full contact info. So now I have to attempt to use the microscopic A-Z buttons on the contact list, then scroll thru the list looking for the correct contact. While driving. I have close to 1000 contacts, this is just an accident waiting to happen.

Agreed. I'm dialing more wrong numbers now than I ever have because Voice Commands don't recognize my voice well enough. I thought this was a good idea Tesla had trying to immediately dial when there was only one choice (vs a single-item pick list like we had), but not so much after practical use.

 

[BertL]

Just picked up my 75D last week, and I am trying to get used to AP. Please help me understand what happened tonight. (1) I was traveling east on I8 in San Diego going 60-65 in the slow lane. I activated AP and the car started accelerating wildly scaring my wife. I braked to exit autopilot, waited a few seconds, and reactivated AP and did not get any acceleration. (2) However, when the left lane marker widened because of an on ramp to the right, the car started drifting to the right and I had to take over. Not impressed. Am I missing something? Thanks for any feedback.

Two issues:
1) Perhaps when you reactivated Autosteer, it went back to the (higher) speed shown in the TACC circle to the left of the speed digits (which it is designed to do), or was perhaps Autosteer 8.0 doing the scary "veer towards the right" or "dive into a semi" scenarios discussed upthread? FWIW, I've also found with 8.0, if Autosteer or TACC has slowed-down because of a car in front, it seems to speed-up much faster than 7.1 as MS begins to change lanes using the stalk.

2) This was more prevalent in earlier AP releases, but as I said upthread with a similar example, the problem seems to be back with 8.0 -- be that because of re-learning or logic failures with the new radar-biased code base, no one here knows for a fact and can only speculate. IMHO, Autosteer 8.0 does not remain sufficiently biased to a clearly-identified left lane line when a road begins to widen (it tries to keep seeking the center which becomes a big fail with "Y-like" off-ramps up ahead, and in places where the right lane just naturally expands with no painted right line markers -- I'd much prefer it stay to the left as most drivers would do).

OH! ...and congrats on that new 75D!

 

[GlmnAlyAirCar]

No but a rework of Nav with added features was supposed to be part of 8.0 per the leaked info we read about several weeks ago. Here's hoping it will be updated in 8.1 and we are not here waiting another year for 9.0 to have it.

I continue to hold out (perhaps a bit naively) for a complete overhaul of nav in 8.1.

Ultimately, Tesla's quest to bring about fully autonomous vehicles is going to rely on precise navigation using a highly detailed map database. Information was posted regarding their collection of precision map data some time ago; I am certain their data has only gotten better.

It's obvious that Navigon won't cut it for autonomous navigation. It's also true that Tesla has given very little attention to improving the Navigon system. I have to speculate it is because they have been working on their own, in-house system.

Perhaps parts of it will be available for 8.1. My thought is that this is needed for AP to follow exits in 8.1.