Hey moderator, where is the 'Hey Elon, where is FSD?' thread?

[AnxietyRanger]

You mean like the 100D having 102 kWh? Or 70D that was higher than 70kwh? Or having 0-60 scores better than advertised? Well that's inconvenient for your argument. A few off the 85 is small timer

Yes, Tesla cleaned up its act on the 100 kWh. They also cleaned up their act (probably) on P90DL performance after a court was about to order them to pay damages... We all hope these are signs of Tesla changing for the better after being called out on it enough times.

As for your 70 kWh example, that actually helps prove the point. Not dismiss it. The point is: Tesla was realistically portraying their low-end or small battery capacity (or even under-reporting it in instances), while until 100 kWh they were inflating their high-end or large battery capacity (85 kWh and 90 kWh). People speculated this was done to inflate the difference between the more expensive large battery vs. the less expensive small one. The explanation makes sense, this policy was very consistently making the higher-margin product look better than it in reality was. Why were just the large battery figures inflated? Never the small battery ones (those were even at times deflated)...

Same was happening with the way Tesla was reporting their large battery non-Performance vs. Performance range (in miles/km), the former was in certain clear and repeated instances reported artificially lower than the rest, even though math and experience would have given non-Performance better figures. Arguably, the speculation goes, this was to keep the more expensive Performance looking better. For some reason only the large battery non-Performance version (the one competing with the more expensive, higher-margin Performance version) had deflated range figures whereas all others followed a clear and same mathematical pattern...

@Magus, people are not stupid. They can see when they are being played and unfortunately since a more benign start, since late 2014 Tesla has been arguably playing people in several instances (which exact instances, of course, matter of speculation).

 

[Magus]

I disagree with the idea that being people are being played with. I think that's one interpretation. Outside of EAP if these are seen as Tesla playing people boy this is a very painful world. I completely understand people being upset by EAP,l and quality control issues, but I don't see it for most of these other topics people feel upset about. I also like the fact that Tesla responded to these and other complaints. Sure it may be argued by force, but they chose to regardless.

 

[dhanson865]

Are you implying that the falcon rockets use AP2 hardware? If not I don't understand your point.

From 1 minute before liftoff aka T-1:00 the on board computers have full control of the Falcon 9. Call them flight computers, call them autopilot, call them guidance systems, call them whatever you want but they

* do a full self test of the entire vehicle before launch
* control the throttle and direction of travel from launch to landing
* are able to land the first stage at sea or land

Getting stage 1 to return to launch site or land on a ship at sea were both considered impossible a decade ago or at least financially irresponsible just a few years ago. Now it's common place enough that all the competitors have to create similar programs or face a cost disadvantage so high it would bankrupt a private company.

The launch on Thursday, Formosat 5, was the 40th Falcon 9.
This was the 38th successful launch.
This was the 15th successful booster recovery. Of those, 9 were ASDS (at sea) and 6 were RTLS (on land).
This was the 11 straight recovery in a row where recovery was attempted.
There were 2 successful re-flights of a previously flown booster so far.

I'm guessing the person you replied to thinks this is a sign that fully self driving AP in a Tesla can be done as well.

If you want a next level rebuttal you could go with something like:

For $50,000,000 a launch you can solve problems that you can't for $35,000 a car.

But then cars don't do vertical take offs and landings so I'm not going to say it's impossible to get AP to the fully self driving stage just because of the price difference.

 

[verygreen]

For $50,000,000 a launch you can solve problems that you can't for $35,000 a car.

But then cars don't do vertical take offs and landings so I'm not going to say it's impossible to get AP to the fully self driving stage just because of the price difference.

On the other hand they do clear the skies for those launches and landings.
Sure, it's relatively easy to do FSD when every time you file a plan for the upcoming trip, they clear all the roads of all other traffic for the declared time window + some wiggle room.

 

[dhanson865]

On the other hand they do clear the skies for those launches and landings.
Sure, it's relatively easy to do FSD when every time you file a plan for the upcoming trip, they clear all the roads of all other traffic for the declared time window + some wiggle room.

You make a good point such that I couldn't decide between like, funny, or informative. But I wouldn't call it easy.

The winds are worse than traffic in that they are invisible, unpredictable, and can literally slam your ship into the ground hard enough to make it explode on contact. It is nice to not have planes in the way but the Falcon 9 computers are making thousands of corrections a second (both throttle and angle) to keep things as stable as possible. Even then with the sky equivalent of traction control going full time it's a bumpy ride.

I'd say it's easier as a technical problem for a car to stay in your lane, and follow or avoid other cars than for a rocket to fight invisible wind enough to have a good course with no accidents.

 

[verygreen]

You make a good point such that I couldn't decide between like, funny, or informative. But I wouldn't call it easy.

The winds are worse than traffic in that they are invisible, unpredictable, and can literally slam your ship into the ground hard enough to make it explode on contact. It is nice to not have planes in the way but the Falcon 9 computers are making thousands of corrections a second (both throttle and angle) to keep things as stable as possible. Even then with the sky equivalent of traction control going full time it's a bumpy ride.

I'd say it's easier as a technical problem for a car to stay in your lane, and follow or avoid other cars than for a rocket to fight invisible wind enough to have a good course with no accidents.

I have to disagree with you on that last point.

You certainly can tell wind right away by changes in measurements from accelerometers and speed at which you approach the land (via radar - easy). This is very similar to the way car makes sure your wheels don't slip. Sure, you can hit black ice or whatever, but you know exactly something like that happened and apply brakes/whatever as needed. That's why Teslas don't ever spin their wheels on start too (unless you enable slip-start?). In my view this VTOL of a rocket is not too much harder than a similar thing with a multicopter - sure you need bigger cpu and forces are greater so you need to react faster, but that's about it. Just keep your vehicle upright and use gps/whatever to keep aiming at your destination.

Now in real traffic you also need to have a huge visual component that's not there in rockets. You need to read signs, traffic signals, recognize objects not all of which could be confirmed by radar (either because the radar beam is narrow and the object comes sideways or because object is RF-nonreflective), ....

 

[u00mem9]

Most of us are fans, that wish Tesla would get there head on straight and stop contributing to their own troubles.

Where would AP2 be if they hadn't let the mobileye deal fall apart?

 

[Magus]

I agree, let's get back to talking Tesla/FSD and away from hyperbole.

 

[conman]

IWe can keep discusisng Elon's FSD promise, given it remains timely at least as long as first FSD features start appearing.

In that case here is my prediction: We will see one, just one, feature of FSD rolled out by the end of this year. That way it can appear that FSD is making progress and on some kind of "target". I don't know what it will be but I doubt they'll leave us with no sign of its progress.

 

[efusco]

I just snipped around 20 posts in this thread...you know, the one that was started because the other thread on the same topic was closed because grown adults couldn't be polite and keep it on topic...yeah, that same.

So, I don't want to spend my time doing this. I'll start putting those who make ad hominum attacks, false accusations, or intentionally derail the topic either on a ban or on moderation and I don't really care if you think it's unfair. It's unfair to the moderators that we have to keep doing this over and over.

Keep it polite and topical. If you can't, find a new thread to post in. If there are people who just annoy you so much that you can't control yourself, put them on "ignore". Either way, stop it.

 

[AnxietyRanger]

In that case here is my prediction: We will see one, just one, feature of FSD rolled out by the end of this year. That way it can appear that FSD is making progress and on some kind of "target". I don't know what it will be but I doubt they'll leave us with no sign of its progress.

That seems plausible. I wonder if you have any thoughts what that FSD only feature might be?

My speculation hovered around expanded traffic light and sign recognition, as that would require using a third front camera. That would seem like an obvious candidate for that "one feature"?

Any other ideas?

 

[TaoJones]

That seems plausible. I wonder if you have any thoughts what that FSD only feature might be?

My speculation hovered around expanded traffic light and sign recognition, as that would require using a third front camera. That would seem like an obvious candidate for that "one feature"?

Any other ideas?

Hear hear. If ever we get what was mentioned during the 10/2014 event (traffic light and stop sign *reaction* - as opposed to mere recognition), I'll be a happy guy. And as an AP1/AP2 owner who fell for the 12/2016 video and the supercharging/100kW pack head games earlier this year, happy would be a very high bar at this juncture.

Alright, so that's 2 features. But the point remains: If you find the greyscale argument plausible (and I do), and if you had AP1 and have already experienced near field sign recognition, then how nice it would be by year end to have the car *react to* traffic lights and stop signs.

Given that it's almost September, well, heh.

 

[dhanson865]

I have to disagree with you on that last point.

You certainly can tell wind right away by changes in measurements from accelerometers and speed at which you approach the land (via radar - easy). This is very similar to the way car makes sure your wheels don't slip. Sure, you can hit black ice or whatever, but you know exactly something like that happened and apply brakes/whatever as needed. That's why Teslas don't ever spin their wheels on start too (unless you enable slip-start?). In my view this VTOL of a rocket is not too much harder than a similar thing with a multicopter - sure you need bigger cpu and forces are greater so you need to react faster, but that's about it. Just keep your vehicle upright and use gps/whatever to keep aiming at your destination.

You are talking about reacting to wind and you act like its no big deal.

After it hits your instrument it may be too late. A strong shear or downdraft will crash a flight before controls can react. It doesn't matter how high it is above the ground if it starts pointing down more than up you've lost control and its time for the self destruct sequence. There is no recovery from that, this isn't a cruise missile it's a cargo rocket. Even if the winds are light enough it takes a lot of work to react vs being proactive.

Driving a car allows you to plan ahead vs solid objects. Wind isn't a factor on the average drive. Wind is a factor on the average rocket launch and is one of the leading causes of scrubbing a flight because so many flights failed due to wind in the past.

The car AP/Driver has the easier job because they don't deal with winds as a primary concern. Rockets don't deal with street signs or stop lights but then short of a hurricane or tornado street signs aren't attacking your car.

 

[AnxietyRanger]

Wind isn't a factor on the average drive. Wind is a factor on the average rocket launch and is one of the leading causes of scrubbing a flight because so many flights failed due to wind in the past.

While I don't disagree, it must be said that abundance of caution in flight is also proportional to the damage if/when realized. Delaying a launch of a very expensive rocket due to wind makes sense, especially since the frequency of launch is so infrequent that a small relative delay is very much worth it...

Driving in bad weather may also make sense, but delaying it may not always be as realistic as delaying a rocket launch... hence cars drive when planes don't and planes fly still when rockets don't... it isn't just about it being hard, it is also about the risk vs. reward. That may also be why driving a car is actually more risky than boarding a plane.

So, some of that rocket launch scrubbing is just managing that risk...

Back to AP:

Rockets don't deal with street signs or stop lights but then short of a hurricane or tornado street signs aren't attacking your car.

Many unexpected things can hit cars, though. The forces of nature on average may be milder towards a car given that it doesn't have to fight gravity, but the ground is also a busier place due to this same reason. For example, an animal might jump in front of a car at any time.

I have no comment on (or interest in) which of these autonomous features are/would be easier to implement. Just putting some thoughts out there on the nature of self-flying vs. self-driving...

 

[verygreen]

You are talking about reacting to wind and you act like its no big deal.

After it hits your instrument it may be too late. A strong shear or downdraft will crash a flight before controls can react. It doesn't matter how high it is above the ground if it starts pointing down more than up you've lost control and its time for the self destruct sequence. There is no recovery from that, this isn't a cruise missile it's a cargo rocket. Even if the winds are light enough it takes a lot of work to react vs being proactive.

I feel like if you sample your vertical angle frequently enough you can always catch it. This was done even before computers with simple inertial mechanical systems I am sure (for take off anyway).
I am not a rocket scientist, but that's my understanding of it anyway, it might be wrong too of course

(Also are you saying that if I want to do a nuclear strike and win, I just need to choose a windy day at the target location? )

 

[conman]

That seems plausible. I wonder if you have any thoughts what that FSD only feature might be?

My speculation hovered around expanded traffic light and sign recognition, as that would require using a third front camera. That would seem like an obvious candidate for that "one feature"?

Any other ideas?

I'm guessing they will not be rolling out FSD features until the alleged EAP features for AP1/2 parity all show up, at least in beta form. This means we should be seeing speed sign recognition for EAP and... auto wipers. I have my own reservations about auto wipers - it isn't that they can't detect rain drops with the cameras. I'm quite sure they can detect them. The problem I envision is that the cameras aren't behind the portion of windscreen that the wipers clean. This means that they can't dynamically see raindrops returning after the wipers have cleaned them. Of course one can make an algorithm based on new drops appearing but that is inherently more complex than comparing wiping to new drops.

Anyway back to the question at hand, my prediction for an FSD feature is one of the ones you've listed - traffic lights, and with a big fat beta warning (perhaps it should be an alpha warning.) Traffic lights have the luxury of telling the car when it can go again whereas stop signs do not. So it's fine for the car to detect when to stop based on stop signs but without much more smarts it won't know when it can go again.

 

[AnxietyRanger]

The problem I envision is that the cameras aren't behind the portion of windscreen that the wipers clean.

But they are... No?

Anyway back to the question at hand, my prediction for an FSD feature is one of the ones you've listed - traffic lights, and with a big fat beta warning (perhaps it should be an alpha warning.) Traffic lights have the luxury of telling the car when it can go again whereas stop signs do not. So it's fine for the car to detect when to stop based on stop signs but without much more smarts it won't know when it can go again.

Yep, you make a good point about the danger... My idea is that they'd be more informative at this time than something the car would strongly react to... There is no image for green in the Tesla firmware:

AP2.0 Cameras: Capabilities and Limitations?

AP2.0 Cameras: Capabilities and Limitations?

 

[_jal_]

The wipers on my S wipe the windshield in front of the cameras. I don't get why this feature isn't done, but then again I don't get any of it anymore

 

[AnxietyRanger]

The wipers on my S wipe the windshield in front of the cameras. I don't get why this feature isn't done, but then again I don't get any of it anymore

Model X as well.

 

[conman]

But they are... No?

The wipers on my S wipe the windshield in front of the cameras. I don't get why this feature isn't done, but then again I don't get any of it anymore

Are they? I must be remembering wrong, sorry.