What will happen within the next 6 1/2 weeks?

[_jal_]

Nit-pick: Bread is made with yeast, not bacteria.

Carry on.

Blah! I thought yeast was bacteria but it’s fungus. I guess that makes more sense.

Now excuse me. I have to patent ringworm bread.

 

[_jal_]

I think it's possible that by the end of the year we could see a small number of vehicles with a clunky version of street driving. My car will almost take turns from a left turn lane. So, for instance, I think it is possible that some cars will be able to actually make some of those turns. (I didn't get FSD, so mine probably won't) I also think it is possible that we will see stop light and stop sign stops taking place. My previous AP2 car would stop at the end of exit ramps. I think that's standard behavior on a number of exits - I don't remember and I didn't have NoA for long before turning it in.

It is going to be very clunky and hardly actually like the "demo" from 2016. My 2016 MS had EAP so I had Enhanced Summon for a few weeks. It is not actually a viable product that normal people could use at all. It's not something that just anyone could use with confidence that everything is going to be okay. That being said, I did use it in a parking lot to bring the car to my son and I, and he thought it was totally cool. I also used it in two different parking lots that had no traffic while standing no farther away than 30 yards, and the bystanders were totally wowed. The future is here now! But it actually wasn't. It's brittle beyond brittle, but the expectations are shattered because the car is driving with no one in it. That's the goal of releasing FSD. It will be a publicity win if even it sucks and is dangerous and only works in 75% of the situations. It is going to be progress at some level to some people. Heck, I remember with Simple Summon people were amazed my car could back out of a spot without a driver at the wheel.

I do worry immensely about the safety of the system though. Like I wrote, I had AP2 on my previous car. I was one of the guys whose headlights were manual for the first month or so because they didn't even have automatic headlights. Not high beams - just headlights. I had to turn on my headlights when it was dark out. There was a "button" on the display I had to "push." Insane. Well, when auto steer came out, I remember using it on an exit ramp and basically being terrified that it couldn't take the ramp. If I hadn't intervened, the car would have driven off the ramp. At the time I remember being incredulous because the way I saw the problem was that Tesla knew the average radius of the curvature of the ramp I was on. I know they did because they drew the ramp on the display. They had to know how to draw the line. They also knew the minimum turn radius auto steer could take. It simply wouldn't - no matter how slow the car was going - take turns tighter than whatever radius. So, why the hell would they even let auto steer stay engaged while traveling along a road with a tighter radius than that???

And there's all the "well, they weren't using maps" and the tiles weren't done and all that, but frankly I've read those arguments and I think they're just apologetic. TMC is full of people who used to "explain" why it's impossible to turn on seat warmers from the app (people need to be sitting in seat warmers for them to work was the primary explanation) or why it's impossible to initiate software updates via the app (what if a spouse initiates the update while the other spouse is away with the car?). The car knew the route I was on and it had a map with the exit ramp and it knew enough to draw a to-scale representation of the exit ramp. It doesn't take Einstein to calculate the average curvature of the ramp and Bill Gates to formulate an algorithm that disengages auto steer when the radius is tighter than the minimum radius auto steer is capable of taking. But, Teslas the world over were barreling down exit ramps toward certain catastrophe with no warning to the driver. That's crazy, so I think the same thing is going to happen here.

The value of such a product is highly debatable. I think you could make an argument that even if it doesn't stop at every stop sign, at least it will stop at some. Maybe it'll stop at one that the driver wouldn't have and prevent an accident. Kind of like autobraking systems out there today. They won't stop every time before hitting a pedestrian, but if they save just one pedestrian, it's an improvement. "But those systems are drivers assistance features not autonomous driving systems," people will say. And that's exactly what comes out of one side of Tesla's mouth about their system. The sizzle though is what comes out of the other side which is that it's a Full SELF Driving system that drives ITSELF so the wink wink nudge nudge is to take your hands off the wheel and let it drive itself.

I hope no one gets hurt.

 

[daniel]

Blah! I thought yeast was bacteria but it’s fungus.

Sorry. Wrong again. Yeast is not bacteria and it's not fungus. It's its own thing.

 

[daniel]

Blah! I thought yeast was bacteria but it’s fungus.

Sorry. Wrong again. Yeast is not bacteria and it's not fungus. It's its own thing.

Oops. I was wrong! Score: daniel: 1; _jal_: 1.

Yeast is in the fungus kingdom. They first developed hundreds of millions of years ago, but as recently as half a century ago when I was in school I don't think fungi were even recognized as their own kingdom.

OTOH, you wouldn't want to make bread out of just any fungus, since some are quite toxic.

Sorry for the digression. I'll stop now.

 

[electronblue]

My main point is that this fear that some of you have that Tesla will release traffic light response that relies on bad map data so that the car will plow through red lights, is completely unfounded IMO. Tesla is not going to release traffic light response with such an obviously dangerous flaw. If traffic light response does rely on map data, Tesla would need to make sure the map data was accurate. But we know that Tesla is working to get camera vision perfect where the car will correctly see red lights without needing map data.

Why not?

In late 2016 Tesla released a rudimentary version of Autopilot 2 that was so bad that it would totally have done the sort of dangerous things you say in this thread Tesla would never release to the public.

The whole launch strategy for Autopilot 2+ — back then and still today — from Tesla relies on the observant driver to catch it doing dangerous things, which it has done many times over the iterations.

Just because you came along at a point in history when the basic Autopilot had become quite mature (relatively speaking anyway), does not somehow make this history or Tesla’s obvious approach go away.

It just means you have been ignorant of it.

 

[diplomat33]

Why not?

In late 2016 Tesla released a rudimentary version of Autopilot 2 that was so bad that it would totally have done the sort of dangerous things you say in this thread Tesla would never release to the public.

The whole launch strategy for Autopilot 2+ — back then and still today — from Tesla relies on the observant driver to catch it doing dangerous things, which it has done many times over the iterations.

Just because you came along at a point in history when the basic Autopilot had become quite mature (relatively speaking anyway), does not somehow make this history or Tesla’s obvious approach go away.

It just means you have been ignorant of it.

I am well aware of how AP was in 2016 and that Tesla requires the driver to pay attention and intervene if AP is going to make a mistake. I am not ignorant of anything. My point is that the driver being there to catch things, does not exclude Tesla also trying to make traffic light response as good as possible before releasing.

 

[diplomat33]

We get another new feature rolling out in 2019.40.2: Autosteer stop sign warning. One small step closer to the FSD "traffic light and stop sign response" feature.

 

[electronblue]

@diplomat33 this is the point I responded to:

My main point is that this fear that some of you have that Tesla will release traffic light response that relies on bad map data so that the car will plow through red lights, is completely unfounded IMO.

Not this which is completely different:

My point is that the driver being there to catch things, does not exclude Tesla also trying to make traffic light response as good as possible before releasing.

You claimed Tesla would not release a traffic light response that relies on bad map data so that the car will plow through red lights — and I said Tesla had done an absolutely comparable thing with ”EAP” many times already. IMO ”fears” that this would happen are not only completely founded, based on history and knowledge of the map data Tesla is using, they are also very likely the real way things will play out.

You were unrealistic and quite likely wrong about Tesla earlier in this thread when you disagreed with those who feared Teslas will miss red lights once the feature is released. They very likely will — if history is any guide — and it will be up to the driver to catch them.

 

[diplomat33]

@diplomat33
You were unrealistic and quite likely wrong about Tesla earlier in this thread when you disagreed with those who feared Teslas will miss red lights once the feature is released. They very likely will — if history is any guide — and it will be up to the driver to catch them.

So before we even have the traffic light response feature, you've already made up your mind that it will totally suck.

 

[electronblue]

So before we even have the traffic light response feature, you've already made up your mind that it will totally suck.

That is not what I said.

I said it will likely miss stops and that it is realistic, based both on history of Tesla Autopilot and what little we know of the feature (e.g. map use, the initial stop recognition already out there etc.), to expect this.

At the very least I consider it unrealistic for you to dismiss the possiblity by opining such fears are ”completely unfounded”. Why did you claim fears of it plowing through red lights are completely unfounded?

 

[am_dmd]

Just like every other AP feature Tesla has covered their **s here. AS stop sign warning says "in some cases" so it may or may not work. Just got the update on my AP2/MCU1 but as usual Release notes are blank and rebooted 3x so have no idea what my particular configuration got.

 

[electronblue]

Sometimes I just don’t understand @diplomat33.

I would consider it absolutely the normal response to expect intial stop sign/red light recognition from Tesla to miss some signs/lights (based on bad maps, bad vision, what have you), as @am_dmd notes above as well... to the extent that it is very surprising to see someone strongly opining otherwise. I consider my view on this to be so normal and common, so much the baseline truth in the Tesla world that I would say even people with wildly different views of Tesla usually agree on this. The greatest fans and haters usually agree on this. We agree Tesla is approaching releasing stuff with the driver resonsibility first, quality only as second and sometimes this means driver catching things Autopilot 2+ missed.

I mean expecting something completely contrary to everything we know of how Tesla has developed and ramped up Autopilot... what the!?!

My main point is that this fear that some of you have that Tesla will release traffic light response that relies on bad map data so that the car will plow through red lights, is completely unfounded IMO.

 

[diplomat33]

That is not what I said.

I said it will likely miss stops and that it is realistic, based both on history of Tesla Autopilot and what little we know of the feature (e.g. map use, the initial stop recognition already out there etc.), to expect this.

At the very least I consider it unrealistic for you to dismiss the possiblity by opining such fears are ”completely unfounded”. Why did you claim fears of it plowing through red lights are completely unfounded?

Sometimes I just don’t understand @diplomat33.

I would consider it absolutely the normal response to expect intial stop sign/red light recognition from Tesla to miss some signs/lights (based on bad maps, bad vision, what have you), as @am_dmd notes above as well... to the extent that it is very surprising to see someone strongly opining otherwise. I consider my view on this to be so normal and common, so much the baseline truth in the Tesla world that I would say even people with wildly different views of Tesla usually agree on this. The greatest fans and haters usually agree on this. We agree Tesla is approaching releasing stuff with the driver resonsibility first, quality only as second.

I mean expecting something completely contrary to everything we know of how Tesla has developed and ramped up Autopilot... what the!?!

I think we are arguing over degrees. Nobody disputes that Traffic Light Response will miss a few cases. I readily admit that the feature will miss some cases. I've never said that the feature will be 100% perfect right out of the gate. The question is how many cases will it miss. I simply took issue with saying it will "plow through red lights" because I found the language to be overly sensationalist.

 

[electronblue]

If history is any judge, it will initially miss a lot of cases, but that’s okay because the driver will catch them.

 

[diplomat33]

If history is any judge, it will initially miss a lot of cases, but that’s okay because the driver will catch them.

Well, we don't know how many cases it will miss since the feature is not out yet. But the fact that the feature has not been released to the public yet even though it has been in internal development for many months now, shows that Tesla is working to make the missed cases as low as possible before they release it.

 

[electronblue]

Well, we don't know how many cases it will miss since the feature is not out yet. But the fact that the feature has not been released to the public yet even though it has been in internal development for many months now, shows that Tesla is working to make the missed cases as low as possible before they release it.

Not at all. It just shows it has not been released yet for whatever reason. We don’t know what the reason is. Smart Summon was delayed by a year and still misses all sorts of stuff all the time.

 

[diplomat33]

Not at all. It just shows it has not been released yet for whatever reason. We don’t know what the reason is. Smart Summon was delayed by a year and still misses all sorts of stuff all the time.

Sure. But if Tesla did not care about missed cases since the driver is expected to catch them and, feedback from the fleet supposedly helps Tesla solve the missed cases faster, then why not release the feature 5 months ago?

 

[electronblue]

Sure. But if Tesla did not care about missed cases since the driver is expected to catch them and, feedback from the fleet supposedly helps Tesla solve the missed cases faster, then why not release the feature 5 months ago?

It could require HW3 to work sufficiently well for anything more than a warning (and launching HW3 features is a can of worms retrofit-wise), for example. Maybe they are waiting to integrate and validate more mapping data for it work at all.

Look, I’m not saying Tesla isn’t working on improving features, I just think you are once again cherry picking an extremely sunny scenario when reality to me seems a lot more messy. The reality is a probably a combination of many things and the eventual release probably just an evolutionary step that is probably just about as messy as every other Autopilot 2+ software release before it.

Had you just said, sure it will miss red lights, but the driver will be there to catch it before catastrophy, I would have agreed history so far would support that. That to me seems a likely scenario. But left on its own devices, history would say the initial Autopilot 2+ red light detection will absolutely plow through many red lights.

 

[arghx7]

This dude will never be satisfied. Give him a million dollars and he’ll find a way to complain.

 

[electronblue]

This dude will never be satisfied. Give him a million dollars and he’ll find a way to complain.

Thank you for that useful argumentation.