FSD rewrite will go out on Oct 20 to limited beta

[zecar]

This is why Elon keeps saying "Lidar is doomed." He thinks they can get vision-based system working faster than Lidar and other sensor suites can come down in price.

Didn't Elon say FSD would be worth something like $100K? You are looking at the wrong end of the equation. Net value is what matters, not cost.

Also, Lidar wasn't an option for Elon. The reason I'm not dating Halle Berre is not because I have better options. Elon literally used what he had available to put on cars. Cameras, ultrasonic, and the forward radar everyone uses.

 

[arghx7]

Didn't Elon say FSD would be worth something like $100K? You are looking at the wrong end of the equation. Net value is what matters, not cost.

Also, Lidar wasn't an option for Elon. The reason I'm not dating Halle Berre is not because I have better options. Elon literally used what he had available to put on cars. Cameras, ultrasonic, and the forward radar everyone uses.

Cost matters because if it's too expensive, no OEM will green light volume production.

 

[Mardak]

I'd guess differently - let's assume they get the stoplight piece working without human intervention. Then they could employ the same techniques (i.e. pull stalk) to ensure driver is awake just before the required turn, or maybe just 100' before the turn. At least I'd prefer this approach compared to a nag.

The latest traffic light stopping / automatic following behavior also uses a nag, and it's kinda confusing. The stalk is needed to confirm when you're the lead vehicle to continue and hands-on-wheels nag is needed to follow.

This resulted in some strange behavior for me yesterday when Autopilot incorrectly believed the light was green (sun was shining into the traffic light) and was nagging for me to confirm and started flashing the screen blue. I eventually tugged on the steering wheel and the rendered line turned green ahead of the lead vehicle, but the light was still red. Not sure how Tesla would process this data if it did get sent back, but on the surface, it seems like the driver confirmed Autopilot to follow on a green light.

Yeah, it'll be interesting how Tesla manages the confirmation/nag for intersections or even sharp curves as the steering wheel will be turning much more than Autopilot has needed / been allowed to turn. Will it be like the traffic light behavior of without confirmation, the car will come to a complete stop?

 

[Bladerskb]

Which is not a FSD system.

Systems actually delivered at scale will need to be $10K to $50K. The DEVELOPEMENT systems are much more expensive. Tesla developers have had to work with very inexpensive in car systems. Meanwhile all the other developers did not have to work around these cost limitations.

Waymo isn't driving their production system yet because it took a very expensive development system to actually make a robotaxi.

FSD system and Development System (53 mins)

Mobileye beat Waymo to a commercialized level 4/5 system? At the price you quote all car manufacturers must be rushing to integrate this system into their cars. Tesla may as well drop their development efforts as they have lost badly.

NIO will implement the system in the car they will announce Jan 2021 for $10k.

 

[DDotJ]

I imagine they were hand picked, and not a randomized selection of EA members.

So I'd say this is the one that likely held the most value, and one of the reasons I kind of laugh at the early access.

I do because it's basically being a free intern for Tesla.

I say this as someone who would write frequent and detailed emails If I was picked for the early access program.

Elon did mention they were selecting "safe drivers", although I have no idea how they would know who is a safe driver unless they were collect driving heuristics on the Early Access members.

I used to be in Early Access until I sold my Model S for a 3. it looks like the program got a lot more sophisticated from the olden days

If anyone from Tesla is reading, I'd love to be let back into the club *hint hint*

 

[DDotJ]

Interesting, so this isn’t just a private beta. It’s actually going out to public EAP members. Just to clarify, you don’t work for Tesla right?

My theory is that the "Public beta" is the group of people who bought FSD early, the ones Tesla kind of hosed and dropped the price of FSD on. Tesla promised everyone who bought FSD early that they would enter the "Early Access" program but it never happened (probably for logistical reasons).

Then when v10 came out, Tesla rolled out the update a few weeks early to that group of FSD buyers that had the Software Update Preference set to Advanced.

I'm assuming the "Private Beta" refers to the actual Early Access program in its original definition, the group of owners with the NDA and the special test builds. And the Public Beta is the "faux Early Access" that includes the FSD buyers with the Software Update Setting set to advanced that gets the release a week or two before the general public.

I'm still a little bitter that Tesla promised early FSD buyers entry into the Early Access program and delivered by technicality by changing the definition of "Early Access". They haven't been very transparent about how they delivered on their promise.

 

[willow_hiller]

Elon did mention they were selecting "safe drivers", although I have no idea how they would know who is a safe driver unless they were collect driving heuristics on the Early Access members.

I believe Karpathy actually talked about this in one of his presentations. Tesla is trying to quantify good/bad driving. And I think it will ultimately tie into Tesla insurance offerings. But he said something around the lines of "Good drivers are all good in the same way, but bad drivers are all bad in different ways."

 

[Snuffysasa]

Sure. Provide a link to an inexpensive self driving system. A better mobile mapping system, which is analogous to a portion of what Waymo has on car, is a half million dollars.

You got to believe us on this one, but Waymo's production stack and any other FSD production stack is closer to 10-15k and moving towards 5k

Mobileye's system costs $15,000 with compute, cameras, lidars and radars.
Waymo made their own in-house Lidar of their most powerful lidar and made it 90% cheaper than the velodyne variant. From $75k to $7.5k 3 years ago. Its been several years which means its most likely even more cheaper as other lidar manufacturer prices have dropped as-well.

They also sell their honeycomb which is their short range lidar and we already know that this lidar is way cheaper than their powerful 360 lidar that cost only $7.5k in 2017. Their honeycomb probably costs around $3k to make.

Their system have 4x honeycomb short range lidar and 1x 360 long range lidar.
When you actually do the math of their Lidar system in 2020.
Its most-likely around $20k for their lidar system.
We know cameras and radars are very cheap in contrast to lidar.
Their overall sensor suite with lidars, radars and cameras would cost well under $40k

EDIT: $40k is a conservative number, the true number is probably around $20k. With Waymo's 5th gen they halved the cost of their sensor again. So the $7.5k sensor now costs around $3k

Thanks for the breakdown and estimates. I'll add a datapoint here for what it's worth. I talked to a Waymo sales rep for honeycome at the start of 2020, and he seemed to suggest if a customer wanted to place a large scale order of honeycombs from Waymo they could get the cost to under $1k... so cost for Waymo to make it definitely significantly under $1k. And I agree the top lidar can't cost more than 3k.

I was estimating that Waymo's 4th gen stack used in production now is maybe $30-40k cost to build. And Waymo's 5th gen stack possibly under $10k. Though, I would definitely like to get more confirmation at some point.

Which is not a FSD system.

Systems actually delivered at scale will need to be $10K to $50K. The DEVELOPEMENT systems are much more expensive. Tesla developers have had to work with very inexpensive in car systems. Meanwhile all the other developers did not have to work around these cost limitations.

Waymo isn't driving their production system yet because it took a very expensive development system to actually make a robotaxi.

Like others said, it is a FSD system. And also Waymo is a production system, especially 5th gen.

False. Waymo is driving their production system on the Jaguar I-Pace and the Chrysler Pacifica. What you see on the Chrysler Pacifica robotaxi in Chandler taking passengers is the production system.

Yes. And also keep in mind what is used in the driverless robotaxi service in chandler now is still the 4th gen platform which was designed to achieve small scale (few hundred - few thousand) driverless ops. Waymo 5th gen which is production ready was designed for large scale production driverless ops.

FSD system and Development System (53 mins)
NIO will implement the system in the car they will announce Jan 2021 for $10k.

10k cost to consumer? nice!

 

[zecar]

I believe Karpathy actually talked about this in one of his presentations. Tesla is trying to quantify good/bad driving

That would seem to be a pretty easy task in an EV. People who seldom touch the brake are looking ahead and not crowding other cars.

As far as insurance, I'm sure that Tesla's plan has been to simply provide the actuaries at their partner insurance company driving information. Really this is not much different than the discount offered by insurance company for running their accelerometer app on the drivers phone.

Tesla likely hasn't expanded their insurance offering beyond California because they do not have an upcoming self drive offering that could impact an owner's ability to insure a Tesla. Or Tesla will public beta something like level 3 on interstate with their California customers only.

 

[mikes_fsd]

Also, Lidar wasn't an option for Elon.

That is a false statement.
When he called lidar a crutch, he specifically qualified it and said that at SpaceX they use lidar for the docking with the space station, he said that that is where lidar is needed/useful.

When trying to solve the problem of FSD, they definitely took lidar into consideration and based on experience they decided against it.

 

[zecar]

That is a false statement

Link to the lidar Tesla could have installed.

 

[diplomat33]

When trying to solve the problem of FSD, they definitely took lidar into consideration and based on experience they decided against it.

What experience are you referring to? I don't think it was really about experience. After all, when Tesla made the decision not to use lidar, other companies were already showing from experience that lidar was a good idea and was useful for FSD. So Tesla actually rejected current experience when they decided against lidar. And, Tesla did not have any significant experience with FSD when they made the decision. Tesla was still on AP1.

No. I think business was a key factor. in 2016, when Tesla launched AP2, lidar was too expensive so it was not a viable option if Tesla was going to equip all their cars with it. Remember that Tesla's profit margins per car were razor thin. And I think that Elon's belief that camera vision would be solved soon also played a big role. Elon said at a conference back in 2014 or 2015 that he thought FSD was a solved problem. So the decision not to use lidar came from business considerations and from Elon's personal convictions about FSD, not experience. Basically, Elon said "I don't want to do FSD the way others are doing. I think I know how to solve FSD better so this is what we are going to do." It had little to do with experience.

 

[powertoold]

Link to the lidar Tesla could have installed.

You said lidar wasn't an option for Elon. That's wrong. Elon specifically considered it with his experience from SpaceX, thought it was stupid for use in cars, and decided against it. Elon never said we can't use it or can't install it.

Anyway, here are some of the lidar statements from autonomy day (starts at 2:33:15):

 

[zecar]

You said lidar wasn't an option for Elon. That's wrong. Elon specifically considered it with his experience from SpaceX, thought it was stupid for use in cars, and decided against it. Elon never said we can't use it or can't install it.

Anyway, here are some of the lidar statements from autonomy day (starts at 2:33:15):

How would Tesla put LIDAR on a 2015 Model S?

I have had a LIDAR based vacuum since about 2016 (Neato and now Roborock). The availability of these products has nothing to do with LIDAR being practical in both a cost and size for Tesla.

 

[powertoold]

How would Tesla put LIDAR on a 2015 Model S?

I have had a LIDAR based vacuum since about 2016 (Neato and now Roborock). The availability of these products has nothing to do with LIDAR being practical in both a cost and size for Tesla.

I don't know, you're asking if Tesla can design cars?

Anyway, the point is that lidar was an option, was considered, and rejected.

 

[zecar]

I don't know, you're asking if Tesla can design cars?

Anyway, the point is that lidar was an option, was considered, and rejected.

Here's some 2020 pricing on the class of LIDAR Waymo is using for its lower speed taxi:

Ouster Releases New Ultra-Wide View OS0 Lidar Sensor and Full Lineup of 128-Channel Resolution Lidar Sensors

 

[mikes_fsd]

What experience are you referring to?

The experience of being the first private company in the world (not just US) to launch the first private spacecraft that docked with the International Space Station for the very first time autonomously.

That experience.

Oh, also the experience that of the first privately designed+built human rated spacecraft that has docked with the ISS!

That experience as well.

In both instances (and every Dragon docking) they use lidar!

 

[diplomat33]

The experience of being the first private company in the world (not just US) to launch the first private spacecraft that docked with the International Space Station for the very first time autonomously.

That experience.

Oh, also the experience that of the first privately designed+built human rated spacecraft that has docked with the ISS!

That experience as well.

In both instances (and every Dragon docking) they use lidar!

Yes, and based on how good lidar is with docking spacecrafts, Elon should know that lidar would work great for autonomous driving. Lidar can classify objects and measure precise position, distance and velocity of objects, which are very important in autonomous driving. Many companies are using lidar successfully for autonomous driving. So we know that lidar does work for autonomous driving.

The fact is that Elon did not reject lidar because it is a bad sensor for FSD. It's clearly not since so many companies use it successfully. He just prefers camera vision because he believes that with the right machine learning, camera vision can "solve" FSD. That's why he rejects lidar.

 

[ItsNotAboutTheMoney]

Yes, and based on how good lidar is with docking spacecrafts, Elon should know that lidar would work great for autonomous driving. Lidar can classify objects and measure precise position, distance and velocity of objects, which are very important in autonomous driving. Many companies are using lidar successfully for autonomous driving. So we know that lidar does work for autonomous driving.

The fact is that Elon did not reject lidar because it is a bad sensor for FSD. It's clearly not since so many companies use it successfully. He just prefers camera vision because he believes that with the right machine learning, camera vision can "solve" FSD. That's why he rejects lidar.

More accurately he says that he rejects lidar because he believes that vision needs to be solved anyway for full autonomy.

 

[powertoold]

I think Elon‘s logic on LiDAR versus cameras goes like this:

What level of positional and size accuracy is needed for driving? Around 10cm

Within the next 3 to 5 years, given the pace of vision NNs, will cameras be able to achieve this accuracy? Yes

Therefore lidar is unnecessary.