Seems like FSD is a complete crock

[EVNow]

It depends on how fast they iterate. Adding more Tesla's will improve the rate at which they validate changes. Lets say you came up with a new build and want to validate some .001% scenario. Having 10 million cars on the road will help you test faster, thus iterate faster compared to 1 million cars.

Correct in theory. But in practice today, the fact that they ask triggers to send back data at such low rates means for the edge cases they are looking at now, they have plenty of cars.

In other words, the limitation is the whole cycle of training, optimizing and validating, creating a signed off build finally. Not lack of data - currently.

No doubt far into the future when they are chasing the 5th 9 or 6th 9, it will come in handy.

green on Twitter

 

[powertoold]

Tesla's approach is the only one that makes sense right now.

Others without a fleet have engineers who are driving the cars around unnaturally. They aren't driving like normal humans but testers, which doesn't produce a natural dataset.

 

[Daniel in SD]

Nope. Other companies are operating in geo-fenced areas and developing thousands of "scenarios" the car has to handle. Thus they will neatly go through the L2 to L5 progression since their approach is to find local solutions to the FSD problem while Tesla's approach is to find a general solution.

I'm not sure I understand. They're developing rules just like Tesla does in their current implementation of Autopilot. Now, there are some people who seem to think that Tesla will use neural nets to drive the car. That would indeed be a radically different approach but there isn't really any evidence of that.
Anyway I'm not sure what any of this has to do with SAE autonomy levels. Other companies are planning on skipping Level 3 entirely and don't have any immediate plans for Level 5 (since that would mean operating everywhere in all weather conditions!).

 

[Daniel in SD]

Well, according to the website enhanced summon is released today. And for whatever reason, they haven't removed that statement which wouldn't take that much time to remove. No "coming soon" or "only to the private beta" or anything.

Seems like a great claim for false advertising, but oh well.

There's also the fact that it won't "really" find you anywhere in the parking lot since it's only designed to be used from less than 150 feet away and only when you are in a position to make sure it doesn't run into anything.

 

[EVNow]

There's also the fact that it won't "really" find you anywhere in the parking lot since it's only designed to be used from less than 150 feet away and only when you are in a position to make sure it doesn't run into anything.

Yes, they need a page or two of disclaimers on that page.

The user manual will have all the right disclaimers, since they are vetted by legal. They need the lawyers to be vetting the website as well.

 

[diplomat33]

Ultimately, I want Tesla to deliver on their FSD promises. I prefer Tesla to be late on their FSD promises but deliver than not deliver at all. I've loved all the AP updates so far and I can't wait to get Enhanced Summon, Traffic Light Response and Automatic City Driving whenever they are available. It's exciting to see Tesla make progress on these features and watch AP become more and more capable over time.

 

[Octo]

Correct in theory. But in practice today, the fact that they ask triggers to send back data at such low rates means for the edge cases they are looking at now, they have plenty of cars.

It can also mean that there simply aren’t enough people in Tesla’s autonomy team who would look at that data.

Given the recent news how unstable that team is, that’s a real possibility.

We’ll see in two weeks.

 

[EVNow]

It can also mean that there simply aren’t enough people in Tesla’s autonomy team who would look at that data.

Given the recent news how unstable that team is, that’s a real possibility.

We’ll see in two weeks.

I don't see that team as anymore unstable than any other team in a usual silicon valley company.

Apart from dev team size, there are other limitations too. Labelers for eg - and they may have a 1,000 contract labelers. It becomes difficult to scale dev team / labelers after sometime.

 

[Octo]

I don't see that team as anymore unstable than any other team in a usual silicon valley company.

I disagree. Tesla’s autonomy team is an outlier, even in SV. It’s highly unusual to have so many high level departures in a short time.

This is a product that is extremely difficult to develop and where the CEO is running a borderline fraudulent strategy.

I would have bailed too. Frankly, I’d be worried that “Tesla Autonomy Engineer” on my resume is becoming a black mark.

But maybe I’m totally wrong and you’ll be enhancedly summoning your car from a busy parking lot in 14 days. And surfing TMC while your car navigates through busy city traffic, negotiating traffic lights and stop signs.

 

[EVNow]

... the CEO is running a borderline fraudulent strategy.

You are now into conspiracy theories.

If he wasn't optimistic (and into "borderline fraudulent strategy") he wouldn't have started a new auto company, let alone an EV company. There hasn't been a successful auto company in US since 1941.

BTW, Tesla autonomy engineers are heavily sought after.

 

[Octo]

You are now into conspiracy theories.

If he wasn't optimistic (and into "borderline fraudulent strategy") he wouldn't have started a new auto company, let alone an EV company. There hasn't been a successful auto company in US since 1941.

BTW, Tesla autonomy engineers are heavily sought after.

What conspiracy?

I just think Musk is a BSer extraordinaire who managed to pull off some successes but is currently failing spectacularly with autonomous driving.

Because it’s in different galaxy, difficulty wise, than talking people into using PayPal or getting a team to engineer a good EV.

Shooting for the moon is admirable if you use your own rocket. If you bamboozle your fans out of their $$$ with lofty promises, then it’s not so admirable anymore.

It’s a futile discussion anyways. You seem to think just willing something into existence is going to work. I hope you’ll get your fully self driving EV in the lifetime of your car. Don’t be disappointed if it doesn’t happen.

 

[jebinc]

What conspiracy?

I just think Musk is a BSer extraordinaire who managed to pull off some successes but is currently failing spectacularly with autonomous driving.

Because it’s in different galaxy, difficulty wise, than talking people into using PayPal or getting a team to engineer a good EV.

Shooting for the moon is admirable if you use your own rocket. If you bamboozle your fans out of their $$$ with lofty promises, then it’s not so admirable anymore.

It’s a futile discussion anyways. You seem to think just willing something into existence is going to work. I hope you’ll get your fully self driving EV in the lifetime of your car. Don’t be disappointed if it doesn’t happen.

Around here, he's known at "The Charlatan". Doesn't get me wrong, I love the vision, SpaceX, Tesla, etc., but he over sells and under delivers (on time) just about everything. Probably why FTC and investors are constantly after is @$$.

 

[AlanSubie4Life]

Around here, he's known at "The Charlatan". Doesn't get me wrong, I love the vision, SpaceX, Tesla, etc., but he over sells and under delivers (on time) just about everything. Probably why FTC and investors are constantly after is @$$.

I love my car, and I like Elon's receptiveness to fans and comments. I recognize no one is perfect, so I just ignore his salesman tactics and like him for who he is and what he has accomplished. We do have to recognize that Tesla has been very successful, and the Model 3 is an excellent car. Have they lived up to all the promises and hype that Elon has pushed? No. Would someone else have taken Tesla's place without Elon, or would Tesla have been as successful without Elon? No idea.

I don't expect FSD/vehicle autonomy is going to yield results for anyone anytime soon. It's just too difficult. In addition, I have heard that Velodyne's RMA rate on their LIDARs is 100% in the first year!

I guess we'll see. It's good that people are trying, and I hope that all companies involved can be responsible and maintain public trust & safety.

 

[happyzod]

I'm not sure I understand. They're developing rules just like Tesla does in their current implementation of Autopilot. Now, there are some people who seem to think that Tesla will use neural nets to drive the car. That would indeed be a radically different approach but there isn't really any evidence of that.
Anyway I'm not sure what any of this has to do with SAE autonomy levels. Other companies are planning on skipping Level 3 entirely and don't have any immediate plans for Level 5 (since that would mean operating everywhere in all weather conditions!).

If you watch autonomy day, you will see how their approach is different. They use machine learning to figure out intent, and train their models using real life examples. Then they validate the model in the real world before deploying it,

Take the autonomy day example. If a car is merging into your lane, instead of a rules based approach to detect it, they will use machine learning and their data engine to train a model that does quite well in figuring out if a car wants to merge into your lane. Path-finding and driving policy are separate from their vision system, those can be rules based. In this case the rule would be to slow down for the merging car.

 

[Daniel in SD]

If you watch autonomy day, you will see how their approach is different. They use machine learning to figure out intent, and train their models using real life examples. Then they validate the model in the real world before deploying it,

Take the autonomy day example. If a car is merging into your lane, instead of a rules based approach to detect it, they will use machine learning and their data engine to train a model that does quite well in figuring out if a car wants to merge into your lane. Path-finding and driving policy are separate from their vision system, those can be rules based. In this case the rule would be to slow down for the merging car.

If you watch presentations from other AV companies they also try to do path prediction of other actors. It's an absolutely critical part of autonomous vehicle software and nothing new. Whether or not neural nets will work better than heuristics, who knows. I look forward to seeing since right now autopilot is horrible at predicting when people are merging into your lane. Someone can be one third of the way into your lane and the car acts like they're not about to merge...

 

[EVNow]

It’s a futile discussion anyways. You seem to think just willing something into existence is going to work.

LOL. This shows how poorly informed you are. My views are quite clearly documented here and yet you prefer to make up stuff.

 

[Octo]

LOL. This shows how poorly informed you are. My views are quite clearly documented here and yet you prefer to make up stuff.

I was referring to your repeated statement that Musk’s approach “just do it” is a predictor for future success. Just because the can do attitude worked for his ventures more or less, doesn’t mean it will for autonomous driving.

I’m not against any of these efforts. I think they are great. I just cannot stand that he pre-sells it for real money, raising expectations with bombast. The main reason why I dislike this so much is the real danger for Tesla’s reputation. Together with the service issues and my own experience buying a Tesla vehicle, I’m simply concerned I made a rather expensive mistake because if this company goes under, kiss your cars value goodbye.

 

[Daniel in SD]

I was referring to your repeated statement that Musk’s approach “just do it” is a predictor for future success. Just because the can do attitude worked for his ventures more or less, doesn’t mean it will for autonomous driving.

I agree. I don't think it really compares to starting an EV company or a rocket company. The question was never whether or not you could design a $100k EV, it was whether or not you could get people to buy it. With autonomous vehicles the question is whether or not it's even possible to make one. There are tens of billions of dollars being invested in to the industry (and plenty before Tesla even started their program). It's a totally different situation IMHO. It's the difference between pre-selling a special set of Arachnid wheels and a fusion reactor.

 

[diplomat33]

With autonomous vehicles the question is whether or not it's even possible to make one.

I am of the opinion that it is definitely possible, it's just a matter of figuring out how to solve it. Every challenge has seemed impossible until it was solved. Fully autonomous vehicles is undoubtedly the hardest AI problem we've ever faced but it is not impossible.

 

[Daniel in SD]

I am of the opinion that it is definitely possible, it's just a matter of figuring out how to solve it. Every challenge has seemed impossible until it was solved. Fully autonomous vehicles is undoubtedly the hardest AI problem we've ever faced but it is not impossible.

Making a fusion reactor is also "definitely possible" but should someone pre-sell one to consumers?