Firmware 8.0

[number12]

question to all: why wouldn't test drive cars have 8.0 already? Or do they?

 

[BertL]

question to all: why wouldn't test drive cars have 8.0 already? Or do they?

"Pre-release" code has only been made available to Non-Disclosed testers thus far. I don't see any reason why Tesla or any company would want to open up pre-GA code that may have more bugs, where features may not be finalized, and put that sort of mess on vehicles that are in-place to help set a tone and gain a sale with prospective buyers.

 

[AlMc]

Sometime in the next couple of weeks.

EM indicated at the press conference that they still may need to do another firmware update for the beta group before release.

I would have bet it would have been released last week but if another round of beta is needed, then probably the end of September.

My WAG

 

[JohnnyG]

I think there may be some confusion as to the timing origin of the changes in version 8.x. For those involved in manufacturing, software design, and product development which affects the well-being of a human, we can be pretty sure that the RADAR enhancements were well underway prior to the fatal accident in Florida. However, the new nag algorithm was most likely indeed put in place due to the recent media frenzy on recent accidents involving AP. The thing to remember is; just because all of the changes are being released at once, in one major version release, doesn't mean that they were all initiated at the same time. It's quite easy to adjust the nag algorithms in a short time and confidently test them. But not so for a major development in 3D mapping and then using this newly refined system as a primary data point for the safety of the occupants. These two change points, although related to the same system, were most likely initiated at vastly different points in the development. Likewise for the other 200+ enhancements / fixes. Just my 2 cents, based on automotive engineering experience.

 

[Todd Burch]

After reading various articles and the blog post, one question comes to mind: under 8.0, will my car activate AEB when there is a coke can rolling laterally across the highway, with the bottom facing the radar? All the conditions would be met, no?

- large object (at least to the radar)
- object is moving
- object is not whitelisted

For background, I work in the field of metrology, and write software that does metrology analysis. What Tesla is doing with the radar--essentially taking radar returns from various positions in time and space, using photogrammetry (really "radargrammetry") to correlate returns and develop a coarse 3D point cloud--is almost identical to the sorts of instruments we write code for--they take pictures from several positions in space, correlate common points from different pictures (using triangulation/bisection and resection), and use that to determine where the points are in 3D space, just as your eyes are able to perceive depth since your brain is seeing two images from slightly different positions and using that to get depth information.

The reason a coke can rolling across the road wouldn't cause this problem is because the radar would be looking at the signature of that can over time and space. Even if the concave portion is perfectly facing down the road, as the car moves the radar signature returned by that can will vary in intensity. The closer the radar is to the center axis of the can, the stronger the return will be.

As the car moves, it will be moving toward or away from the axis of the can (almost certainly away). As a result, the intensity of the return will drop significantly as the car moves, and can therefore be filtered out.

The scenario you describe would require the can to roll across the highway while the central axis of the can almost perfectly points to the radar, which would be a nearly impossible course of events.

To put it more succictly, if a given radar return (such as the can) is identified as changing intensity dramatically over time and space, you can filter it out.

 

[HookBill]

"Pre-release" code has only been made available to Non-Disclosed testers thus far. I don't see any reason why Tesla or any company would want to open up pre-GA code that may have more bugs, where features may not be finalized, and put that sort of mess on vehicles that are in-place to help set a tone and gain a sale with prospective buyers.

We took our first test drive right after 7.1 was released. I asked the Tesla rep if the car we were driving was running it. He said Tesla usually does not upgrade show vehicles until later in the release period so that actual users received theirs in a timely manner. Can't say that is an actual fact but is what we were told.

 

[HookBill]

We took our first test drive right after 7.1 was released. I asked the Tesla rep if the car we were driving was running it. He said Tesla usually does not upgrade show vehicles until later in the release period so that actual users received theirs in a timely manner. Can't say that is an actual fact but is what we were told.

Edit: oops, misunderstood. It wasn't the droid I was looking for.

 

[msnow]

EM indicated at the press conference that they still may need to do another firmware update for the beta group before release.

I would have bet it would have been released last week but if another round of beta is needed, then probably the end of September.

My WAG

In one of the back and forth tweets between a reporter and Elon he said "two more beta code drops" so I presume one this week and one more in the next week or two do your guess is pretty good.

 

[chillaban]

For background, I work in the field of metrology, and write software that does metrology analysis. What Tesla is doing with the radar--essentially taking radar returns from various positions in time and space, using photogrammetry (really "radargrammetry") to correlate returns and develop a coarse 3D point cloud--is almost identical to the sorts of instruments we write code for--they take pictures from several positions in space, correlate common points from different pictures (using triangulation/bisection and resection), and use that to determine where the points are in 3D space, just as your eyes are able to perceive depth since your brain is seeing two images from slightly different positions and using that to get depth information.

The reason a coke can rolling across the road wouldn't cause this problem is because the radar would be looking at the signature of that can over time and space. Even if the concave portion is perfectly facing down the road, as the car moves the radar signature returned by that can will vary in intensity. The closer the radar is to the center axis of the can, the stronger the return will be.

As the car moves, it will be moving toward or away from the axis of the can (almost certainly away). As a result, the intensity of the return will drop significantly as the car moves, and can therefore be filtered out.

The scenario you describe would require the can to roll across the highway while the central axis of the can almost perfectly points to the radar, which would be a nearly impossible course of events.

To put it more succictly, if a given radar return (such as the can) is identified as changing intensity dramatically over time and space, you can filter it out.

Agreed. For the coke can example, the main point the blog was trying to make is that 10Hz sampling and unlocking more radar signatures per sample via the new Bosch driver is what helps the car tell that apart. The coke can would only briefly be able to amplify a return signal at a focal point. The effect would not be long lasting and would not paint a continuous picture of imminent doom.

 

[rog]

For background, I work in the field of metrology, and write software that does metrology analysis. What Tesla is doing with the radar--essentially taking radar returns from various positions in time and space, using photogrammetry (really "radargrammetry") to correlate returns and develop a coarse 3D point cloud--is almost identical to the sorts of instruments we write code for--they take pictures from several positions in space, correlate common points from different pictures (using triangulation/bisection and resection), and use that to determine where the points are in 3D space, just as your eyes are able to perceive depth since your brain is seeing two images from slightly different positions and using that to get depth information.

The reason a coke can rolling across the road wouldn't cause this problem is because the radar would be looking at the signature of that can over time and space. Even if the concave portion is perfectly facing down the road, as the car moves the radar signature returned by that can will vary in intensity. The closer the radar is to the center axis of the can, the stronger the return will be.

As the car moves, it will be moving toward or away from the axis of the can (almost certainly away). As a result, the intensity of the return will drop significantly as the car moves, and can therefore be filtered out.

The scenario you describe would require the can to roll across the highway while the central axis of the can almost perfectly points to the radar, which would be a nearly impossible course of events.

To put it more succictly, if a given radar return (such as the can) is identified as changing intensity dramatically over time and space, you can filter it out.

Thanks a lot for that detailed explanation! What you are saying makes a lot of sense.

 

[sandpiper]

Okay, so just to lay out what the other side of the argument is, Electrek's transcript of Elon's presentation has him saying:


"Autopilot accidents are far more likely for expert users. It is not the neophytes. It’s the experts. ... They get very comfortable with it and repeatedly ignore the car’s warnings. It’s like a reflex. The car will beep at them, they tug the wheel, the car will beep at them, they tug the wheel, and it becomes an unconscious reflex action. So we will see half a dozen or more, sometimes as many as 10 warnings in one hour continuously ignored by the driver. We really want to avoid that situation."


So if they really want to avoid the situation of "beep and tug the wheel," it won't be accomplished by allowing people to get away with simply tugging after the beeps. That's exactly what he just said they want to stop. And he equates "the car will beep at them, they tug the wheel" with "So we will see ... as many as 10 warnings in one hour continuously ignored by the driver." In other words, he clearly considers tugging followed by not continuing to leave one's hands on the wheel, which would result in no further beeps, to be an "ignore." So I fear that three tugs followed by removing one's hands from the wheel after each tug, which will lead to more beeps, is considered to be an ignore. Three such ignores and you're out. Three beeps and you're out.

I hope I'm wrong, but I think his statement makes it clear I'm not.

That's exactly how I interpret it. I'm not clear on how it could be otherwise.

 

[brooklynrab]

Amped, as a fellow longtime upset MS Classic owner, on one hand I share your cynicism from one of your earlier posts in this thread about whether v8 will do anything for us, and about all the constant opportunities to make the software better, but instead seeing Tesla make our interface worse, or not doing anything, while Musk was busy chasing publicity and shiny objects.
However, if Model 3 is going to be a mass market success, Musk must get core user interface features like NAV, media and voice control to be competitive at least with a Hyundai, Chevy Impala or a 2005 Lexus. Even Musk must realize this by now, and they have had a long time to work on V8 since he acknowledged these shortcoming with his V7-7.1 tweet, and later with the early V8 leaks.
On this sunny, dry, NYC morning, I will let the weather make me an optimist (but preparing for the punch in the stomach that I know may come).

 

[GlmnAlyAirCar]

It seems it's all about AP. My hint was the comment about Nav capabilities with AP won't happen until 8.1. Sound familiar? I hope I'm wrong.

You don't have to look down. You just have to keep your hands on the wheel. That's what Tesla (?) wants you to do. They won't implement something that will allow you to not constantly have your hands on the wheel.

No, you are wrong. Hands on the wheel isn't good enough. You need to keep applying a torque to the wheel to avoid the nag.

 

[tomas]

Amped, as a fellow longtime upset MS Classic owner, on one hand I share your cynicism from one of your earlier posts in this thread about whether v8 will do anything for us, and about all the constant opportunities to make the software better, but instead seeing Tesla make our interface worse, or not doing anything, while Musk was busy chasing publicity and shiny objects.
However, if Model 3 is going to be a mass market success, Musk must get core user interface features like NAV, media and voice control to be competitive at least with a Hyundai, Chevy Impala or a 2005 Lexus. Even Musk must realize this by now, and they have had a long time to work on V8 since he acknowledged these shortcoming with his V7-7.1 tweet, and later with the early V8 leaks.
On this sunny, dry, NYC morning, I will let the weather make me an optimist (but preparing for the punch in the stomach that I know may come).

Coming from another classic owner who wants a better media player, I wonder if you think Tesla would be positioned to deliver the 3 (or even survive) if they had focused on polishing the mundane vs chasing shiny objects, like: dual motor, ludicrous, AP, and gigafactory? Could they have continued sales momentum with a continually refined S? IMO, the only 'misfire' was the X, which is in itself a very cool vehicle I'd like to own, but not necessarily helpful to the company's path to volume or profitability.

Its a pretty tough path... they have to keep both eyes on the bigger target: maintain sales momentum, raise enough money to fund the huge capital requirements, and demonstrate the ability to be profitable. As the newcomer, they have to do that with outrageous acts... parity on the basics is not enough.

Having said that, like you, I hope they deliver a few more parity items in 8.0. From early reports, I suspect they will. If they don't I'll still enjoy the fact that I've been driving - on balance - the world's best production car since 2012, even if I was too early for some of the shiny objects!

 

[jdjeff88]

I'm late to the party having just got my car but every rumor I read about 8.0 during the last 45 days when I started paying attention was that 8.0 is meant to bring about a significant user interface change. The recent update/press conference focused on AP for obvious reasons but it doesn't mean to me that the other changes that have been leaked/rumored won't come about as well.

 

[brooklynrab]

Coming from another classic owner who wants a better media player, I wonder if you think Tesla would be positioned to deliver the 3 (or even survive) if they had focused on polishing the mundane vs chasing shiny objects, like: dual motor, ludicrous, AP, and gigafactory? Could they have continued sales momentum with a continually refined S? IMO, the only 'misfire' was the X, which is in itself a very cool vehicle I'd like to own, but not necessarily helpful to the company's path to volume or profitability.

Its a pretty tough path... they have to keep both eyes on the bigger target: maintain sales momentum, raise enough money to fund the huge capital requirements, and demonstrate the ability to be profitable. As the newcomer, they have to do that with outrageous acts... parity on the basics is not enough.

Having said that, like you, I hope they deliver a few more parity items in 8.0. From early reports, I suspect they will. If they don't I'll still enjoy the fact that I've been driving - on balance - the world's best production car since 2012, even if I was too early for some of the shiny objects!

Tomas, you act as if this is an "either/or proposition". If the wording of my statement was unclear, let me be clear -- it has been well documented in these forums by a number of us that there are a number of fixes and "simple" modifications that tesla has virtually ignored/refused with regard to imperfect and incomplete software that would not have resulted in the abandonment of the big projects, or even cost them much in resources (as confirmed vociferously on this forum by people with more technical knowledge than me). What has been missing is management skill and desire. 7.0 and 7.1 for example, reflected amateur QC and UI, UX. Just scan the forum.

 

[tomas]

Tomas, you act as if this is an "either/or proposition". If the wording of my statement was unclear, let me be clear -- it has been well documented in these forums by a number of us that there are a number of fixes and "simple" modifications that tesla has virtually ignored/refused with regard to imperfect and incomplete software that would not have resulted in the abandonment of the big projects, or even cost them much in resources (as confirmed vociferously on this forum by people with more technical knowledge than me). What has been missing is management skill and desire. 7.0 and 7.1 for example, reflected amateur QC and UI, UX. Just scan the forum.

I frequent the forum. I just happen to disagree with you about priorities.

 

[pdxgibby]

From the blog post...

"This is where fleet learning comes in handy. Initially, the vehicle fleet will take no action except to note the position of road signs, bridges and other stationary objects, mapping the world according to radar. The car computer will then silently compare when it would have braked to the driver action and upload that to the Tesla database. If several cars drive safely past a given radar object, whether Autopilot is turned on or off, then that object is added to the geocoded whitelist."

I have a couple concerns.

It sounds like this assumes that the "world" doesn't change.

1) Let's say that hundreds of Teslas have traveled along an interstate and there is no geocoded whitelist entry at some location, X. Suddenly, a Tesla approaches this location and there is a semi-truck present (think Josh Brown situation). Radar detects it and triggers an emergency braking actuation of the braking system. Awesome!

Now, let's say that hundreds of Teslas have traveled along an interstate and there is no geocoded whitelist entry at some location, X. At some point, construction crews add a new overpass at this location. The very next Tesla to reach this location would presumably see a radar obstacle and start to trigger an emergency braking actuation to prevent the impact.

The geocoded whitelist is not sufficient to differentiate here.

2) At what point is the geocoded data provided to the Tesla to validate against? Certainly not in real time....relying on connectivity to make a split-second decision is a terrible idea. Could each Tesla really carry a geocoded database of all whitelist entries at once? It's ever changing...and ever-growing. Would it be region-by-region? What if cellular data suffers an outage and you aren't able to download your whitelist data. Would the Tesla view everything as an emergency braking event because it cannot be validated against the whitelist?

I'm very eager to see how well this works. Let's call me excited...with some slight apprehension.

 

[chillaban]

A blogpost cannot explain the nuances of operation in the level of detail to do the implementation justice. Autopilot already relies on map tiles as a possible input, and it downloads and caches nearby tiles. You would have to go quite a distance with no service in a new area to not have tiles handy and in that case you get pre-8.0 style collision protection, which still is pretty good.

As far as roads changing and new signs showing up, false braking alerts / collision alarms are still subject to the user's approval. There is a lot of reference to "ramp rates", which implies the braking action won't be all in. You may sense your car slowing down with full regen if you are truly the first unlucky soul to encounter a brand new sign that appeared overnight, in which case you can hit the gas and override it, and now the whole fleet learns that this obstacle is a false alarm because you did not collide with it.

I expect the approach will work well in the real world and there will be a lot of mitigations implemented for the various corner cases that don't fit well into a short blog post about a significant amount of work.

 

[3Victoria]

Burp.