Autonomy Investor Day - April 22 at 2pm ET

[thegruf]

i would have liked to hear if Tesla considered the stopped vehicle half in lane (aka Firetruck) scenario sorted in current software build, or whether this requires HW3.
It's a bit challenging to consider this a 0.0001 edge case, although this scenario often also involves an intermediate vehicle moving out of path at short notice presenting the system with a response time to reacquire the threat and act.

 

[Lift]

Here's the mapped route btw: Google Maps

 

[Bladerskb]

Here is the FSD demo:

This is a joke right?

 

[S4WRXTTCS]

This is a joke right?

Nope.

That's it.

There is absolutely no content in that as they sped it up to a ridiculous speed.

All we really have is written/recorded testimonials from the test drives. From what I've heard so far I do find it encouraging at least in a sense of growth from what it is now.

It's also important to point out that this is all Tesla code. This isn't a repeat of 2016 when they cheated by using NVidia code. So I don't think anything shown will take more than 9-12 months to be on my car.

 

[buttershrimp]

This is a joke right?

If you slow down the video Blader, you can see your Acura trying to ram them off the road

 

[S4WRXTTCS]

Here's the mapped route btw: Google Maps

Does this mean they have a route planner now?

 

[Pale_Rider]

different cameras too iirc

You could tell from their presentations though that there was a mix of video, coming from both AP2.0 and 2.5.

 

[dckiwi]

This is a joke right?

I've read a lot of your posts, Bladerskb.

I've also experienced the improvement in AP first-hand over the last six months. My experience, combined with this video, some of the impressions from the test drives, and the presentations today, all point in one direction - and it ain't yours.

I wouldn't want to be the one betting against Tesla FSD, but to each his own.

 

[buttershrimp]

This is either a Con or a massive leap in tech.....

Removing the steering wheel and pedals out of the car on purpose? Man, the Unabomber is gonna be pissed.

 

[Soda Popinski]

There are probably twenty companies that have good FSD by that metric. It’s all about reliability and edge cases.

Maybe, but how many do you think will be in production cars and running FSD in 2 years?

 

[ord3r]

That new viz of the environment is pretty nifty

 

[thegruf]

@Bladerskb - even by your standards you are sounding a bit desperate today.

Perhaps the most telling thing away from the obvious strides in AP/FSD performance was that Tesla have created their own entire hardware platform dedicated solely to their installation and already have a next gen product in development.

This single aspect alone frees them of being shackled to other providers be they NVidia or Intel/Mobileye or whomever.

As FSD technology matures in coming years, irrespective of whether you prefer GPS/HD maps/LIDAR or just plain old vision, all the other auto companies are going to have to pay big NVidia or Intel/Mobileye or the like. There is a very good reason Intel paid $22Bn (iirc) for Mobileye - it's a long term bet and they fully expect a return on that investment. Tesla in the meantime have shown how to profitably manufacture EVs and will ever more obviously show a cost advantage in FSD as well.

It is increasingly clear yet again that Tesla/Elon Musk have a farsightedness that most others struggle to comprehend and problems day to day, even month to month or quarter to quarter are just bumps in the road.

 

[Daniel in SD]

Maube, but how many do you think will be in production cars and running FSD in 2 years?

Zero (including Tesla). I think there might be some production cars doing FSD demos since there are MobilEye EyeQ4 FSD demos.

All we really have is written/recorded testimonials from the test drives. From what I've heard so far I do find it encouraging at least in a sense of growth from what it is now

There were exactly the same testimonials when Google gave journalists rides in their self-driving cars in 2010.
Really it's about believing that Tesla will make exponential progress due to their larger dataset. When people talk about betting against it are they talking even odds?

 

[S4WRXTTCS]

There were exactly the same testimonials when Google gave journalists rides in their self-driving cars in 2010.
Really it's about believing that Tesla will make exponential progress due to their larger dataset. When people talk about betting against it are they talking even odds?

The difference is these people are familiar with NoA, and have prior experiences to pull from.

The limited time was probably the biggest drawback which didn't allow for a lot of situation.

With that being said I do expect to see a ton more videos/demo's/etc as various features get released.

It should be an exciting year.

Filled with excitement, and drama due to mishaps.

 

[VT_EE]

Nope.

That's it.

There is absolutely no content in that as they sped it up to a ridiculous speed.

All we really have is written/recorded testimonials from the test drives. From what I've heard so far I do find it encouraging at least in a sense of growth from what it is now.

It's also important to point out that this is all Tesla code. This isn't a repeat of 2016 when they cheated by using NVidia code. So I don't think anything shown will take more than 9-12 months to be on my car.

Slow it down to 25% and it's far more informative and impressive. The new UI is pretty slick.

 

[eli_]

I might be weird but I'm more excited about the visualization UI than the actual driving.

 

[conman]

It's not my imagination, but I can clearly see signs being displayed there as well. Seems obvious that it would be a requirement but you can see them pretty clearly in the demo.

 

[EVNow]

Everyone is using NN for their vision systems, what was interesting was that Tesla is exploring using NN for driving policy. Google has images of every single road in the US so I'm betting that their vision system has plenty of data and is working pretty well. Using a NN for driving policy is certainly worth a try and we'll see if it works! I have a hard time wrapping my head around how one would debug such a system.

Not talking just about street view data. That is what Google has. But, edge case data on the road & traffic. Also, not sure whether the street view data is enough for them - I'm think they want completely 3D mapped with Lidar.

 

[Xminus6]

@Bladerskb - even by your standards you are sounding a bit desperate today.

Perhaps the most telling thing away from the obvious strides in AP/FSD performance was that Tesla have created their own entire hardware platform dedicated solely to their installation and already have a next gen product in development.

This single aspect alone frees them of being shackled to other providers be they NVidia or Intel/Mobileye or whomever.

As FSD technology matures in coming years, irrespective of whether you prefer GPS/HD maps/LIDAR or just plain old vision, all the other auto companies are going to have to pay big NVidia or Intel/Mobileye or the like. There is a very good reason Intel paid $22Bn (iirc) for Mobileye - it's a long term bet and they fully expect a return on that investment. Tesla in the meantime have shown how to profitably manufacture EVs and will ever more obviously show a cost advantage in FSD as well.

It is increasingly clear yet again that Tesla/Elon Musk have a farsightedness that most others struggle to comprehend and problems day to day, even month to month or quarter to quarter are just bumps in the road.

I think an equally, if not greater, factor for Tesla is that they can iterate their hardware and software as quickly as they like. Being reliant on a third party keeps you beholden to their schedule. This was the publicly used excuse for why Tesla discontinued their relationship with MobilEye. It was likely a number of reasons from both Tesla and ME but the fact remains that Tesla is in a unique position on, not only their vertical integration for the system, but their ability to change and re-optimize for their hardware/software suite like no other company. Intel/ME and Nvidia can’t deploy and test their systems on even a fraction of the cars that Tesla can on a short timeline.

 

[CarlK]

Few things revealed/confirmed on the technical side are confidence building (not if you're a Lidar fan).

-- Many algorithms can be used to obtain depth info from vision.
-- Vision also has the ability to figure out distance with perspective cues. Tesla uses radar distance to automatically train vision neural net without needing human intervention. This the first time I've heard people doing that and it's super smart.
-- Tesla NN is trained to use small car/pedestrian movements (body languages) to tell their intentions. This is also news to me and probably the remaining missing link imo. Only vision has resolution to do that. Lidar can not no matter how much training you give it.
-- Lidar + HD GPS mapping many use will not work. Small change of road feature will collapse the whole system.

Our understandings went full circle from vision can be as good as Lidar to vision is better to it's needed. Elon made a point why put the expensive and power hungry Lidar there if vision is still needed? All those points are not bullet proof, nothing is, but they are all very good points. We'll see if there will be Lidar converts but I'm sure there will be when Tesla turns on the FSD and proves it works.