Make your robotaxi predictions for the 8/8 reveal

So Elon says that Tesla will reveal a dedicated robotaxi vehicle on 8/8. What do you think we will see? Will it look like this concept art or something else?

I will say that while this concept drawing looks super cool, I am a bit skeptical if it is practical as a robotaxi. It looks to only have 2 seats which would be fine for 1-2 people who need a ride but would not work for more than 2 people. I feel like that would limit the robotaxis value for a lot of people. Also, it would likely need a steering wheel and pedals for regulatory reasons even if Tesla did achieve eyes-off capability.

So I think this is concept art for a hypothetical 2 seater, cheap Tesla, not a robotaxi.

Could the robotaxi look more like this concept art but smaller? It could look a bit more like say the Zoox vehicle or the Cruise Origin, more futuristic box like shape IMO and seat 5-6 people.

Or maybe the robotaxi will look more like the "model 2" concept:

Other questions:
- Will the robotaxis be available to own by individuals as a personal car or will it strictly be owned by Tesla and only used in a ride-hailing network?
- What will cost be?
- Will it have upgraded hardware? Radar? Lidar? additional compute?
- Will Elon reveal any details on how the ride-hailing network will work?

Thoughts? Let the fun speculation begin!

 

[rentierparasit]

Hi, all --

But back to the topic at hand.

What could this plausibly be? Any radically different hardware poses a challenge to the "only Tesla has the data" story as well as the "every Tesla has robotaxi capabilities" story. A business model presentation seems like it would only give rise to derision. I've been following the thread but am too lazy to review it; was any kind of consensus ever reached?

Yours,
RP

 

[DrChaos]

Hi, all --

But back to the topic at hand.

What could this plausibly be?

It's going to be exactly like Optimus is to robotics. A shiny external platform with new hardware that is intended to distract from the key software and operational and business market gaps.

They can show off a small car prototype with robotaxi-compatible features. Canoo was there 5 years ago. Thats not the difficult or interesting part.

The giant problem which they used to understand was that the only way to make its unit cost low enough for robo economics to work is to built it off an already large volume platform, and the only way to get there is to sell a regular car which has known mass market demand. And Elon killed that.

Even with operational robo hardware and better FSD software, there's still a big gap to actual revenue producing robotaxi services. They will need geofencing and much more testing, and remote assistance and customer service, and fleets of people cleaning cars, repairing vandalism, charging them, getting them out of problems, dealing with taxi stands, hotels and airports and all of that is non-scalable local labor.

The only buyers of a robotaxi are robotaxi services and for all those reasons above their development will be slow. I predict 10 years to any significant robotaxi revenue. Waymo hardware and software and service is 5-10 years ahead of Tesla (other than unit hardware cost) and yet they aren't massively scaling.

I mean WTF?

Any radically different hardware poses a challenge to the "only Tesla has the data" story as well as the "every Tesla has robotaxi capabilities" story. A business model presentation seems like it would only give rise to derision. I've been following the thread but am too lazy to review it; was any kind of consensus ever reached?

No. It depends if you still think Tesla has magic or not.

I'm a skeptic. Elon is growing ever more shortermism (symptom of ego and amphetamine abuse and over leverage personally) and unwilling to listen to rational voices.

 

[rentierparasit]

It's going to be exactly like Optimus is to robotics. A shiny external platform with new hardware that is intended to distract from the key software and operational and business market gaps.

They can show off a small car prototype with robotaxi-compatible features. Canoo was there 5 years ago. Thats not the difficult or interesting part.

The giant problem which they used to understand was that the only way to make its unit cost low enough for robo economics to work is to built it off an already large volume platform, and the only way to get there is to sell a regular car which has known mass market demand. And Elon killed that.

Even with operational robo hardware and better FSD software, there's still a big gap to actual revenue producing robotaxi services. They will need geofencing and much more testing, and remote assistance and customer service, and fleets of people cleaning cars, repairing vandalism, charging them, getting them out of problems, dealing with taxi stands, hotels and airports and all of that is non-scalable local labor.

The only buyers of a robotaxi are robotaxi services and for all those reasons above their development will be slow. I predict 10 years to any significant robotaxi revenue. Waymo hardware and software and service is 5-10 years ahead of Tesla (other than unit hardware cost) and yet they aren't massively scaling.

I mean WTF?



No. It depends if you still think Tesla has magic or not.

I'm a skeptic. Elon is growing ever more shortermism (symptom of ego and amphetamine abuse and over leverage personally) and unwilling to listen to rational voices.

Hi, DrChaos --

> They can show off a small car prototype with robotaxi-compatible features. Canoo was there 5 years ago. Thats not the difficult or interesting part.

I don't think the latter part would be considered an obstacle for 8/8.

One big advantage of revealing some sort of dedicated robotaxi is that I don't think they need to show the ability to build it in volume, so no need to build another factory, no $Bs in cap-ex, etc. If you can sell some sort of ARKK-style business model then I think they could just say, "Initially we're going to make 150k/yr of these using the spare space in Austin.' If each RT is worth $100k because ROBOTAXI!, then that pencils out.

This is probably not worth thinking about, because who knows?

Yours,
RP

 

[stopcrazypp]

It's going to be exactly like Optimus is to robotics. A shiny external platform with new hardware that is intended to distract from the key software and operational and business market gaps.

They can show off a small car prototype with robotaxi-compatible features. Canoo was there 5 years ago. Thats not the difficult or interesting part.

The giant problem which they used to understand was that the only way to make its unit cost low enough for robo economics to work is to built it off an already large volume platform, and the only way to get there is to sell a regular car which has known mass market demand. And Elon killed that.

Even with operational robo hardware and better FSD software, there's still a big gap to actual revenue producing robotaxi services. They will need geofencing and much more testing, and remote assistance and customer service, and fleets of people cleaning cars, repairing vandalism, charging them, getting them out of problems, dealing with taxi stands, hotels and airports and all of that is non-scalable local labor.

The only buyers of a robotaxi are robotaxi services and for all those reasons above their development will be slow. I predict 10 years to any significant robotaxi revenue. Waymo hardware and software and service is 5-10 years ahead of Tesla (other than unit hardware cost) and yet they aren't massively scaling.

I mean WTF?



No. It depends if you still think Tesla has magic or not.

I'm a skeptic. Elon is growing ever more shortermism (symptom of ego and amphetamine abuse and over leverage personally) and unwilling to listen to rational voices.

I don't think a central fleet approach will make a profit anytime soon (for same reasons taxies struggle against Uber). Instead the model that might work is each car owner manages their own vehicle and Tesla takes a cut out of any revenue made and/or the owner uses it for personal use. Of course, it's a long way from there.

 

[spacecoin]

"No more rebuys Mr Musk":

No More Rebuys, Mr Musk

Even if you’ve always won, it’s still gambling until you get up from the table.

niedermeyer.io

"Instead, Tesla is going back to a now-familiar well and going “balls to the wall,” in Musk’s juvenile phrasing, on autonomous driving. But here too Tesla is attempting to surf a wave that has already expended much of its strength, over what will have been eight years of selling self-driving fairy dust this October. It’s been possible (though unpopular) to utterly debunk Tesla’s entire approach to driving automation for nearly a decade now, but I’ll save that task for another time. At this point no technical understanding is necessary to see that Musk has been promising the Kingdom of Self-Driving Heaven on Earth by the end of the year so many times, and with so little effort to explain why he has been so consistently wrong about it, that this genre of stock pump belongs in a category of religio-capitalism with the likes of such freaks as Zion Oil and Gas.

In this final choice, electing to hype a nonsensical robotaxi using a thoroughly debunked approach to driving automation instead of shoring up Tesla’s crumbling core car business, we see the logical conclusion of what this story has always been: a gambler who found himself in an industry that bankrupted most (if not all) of its gamblers a century ago. Musk ability to capture the public imagination and leverage popular narratives about technology gave Tesla a totally unpredictable opportunity to succeed at the car business, but rather than solidifying and protecting the nearly 2 million unit/year business he created against all the odds, he just kept doubling down on hype. And unlike William Durant whose wild stock market gambling run created General Motors, Tesla doesn’t have a board of directors or investor base capable of ditching him in favor of an Alfred P Sloan type to tackle the very different challenges of leading a maturing automaker with real scale."

 

[Daniel in SD]

Florida first? maybe.

https://twitter.com/x/status/1782435330680889755

 

[DrChaos]

I don't think a central fleet approach will make a profit anytime soon (for same reasons taxies struggle against Uber). Instead the model that might work is each car owner manages their own vehicle and Tesla takes a cut out of any revenue made and/or the owner uses it for personal use. Of course, it's a long way from there.

I don't think that will work either---regulations or liability. Is each car owner going to be able to take off work for an emergency to fix a tire, deal with a customer in person, find a tow truck, and get screamed at by drugged up karens and kens?

 

[brkaus]

I don't think a central fleet approach will make a profit anytime soon (for same reasons taxies struggle against Uber). Instead the model that might work is each car owner manages their own vehicle and Tesla takes a cut out of any revenue made and/or the owner uses it for personal use. Of course, it's a long way from there.

Uber drivers don't understand how much they are depreciating their car, that's why uber model works. (IMHO of course)

 

[QtownTaxiGreg]

I don't think that will work either---regulations or liability. Is each car owner going to be able to take off work for an emergency to fix a tire, deal with a customer in person, find a tow truck, and get screamed at by drugged up karens and kens?

In that model there would be another job created. Robo Rescue services would spring up for the owners to sub out the grunt stuff.

Uber drivers don't understand how much they are depreciating their car, that's why uber model works. (IMHO of course)

Absolutely. It's a revolving door of operators here.

Uber was recently brought kicking and screaming into the tax system here and they are leaving in droves. On TradeMe there is a dozen white Model Y mostly with taxi registration in Auckland. Mostly at dealers. They have abandoned their get rich at NZ taxpayer's expense schemes and gone home.

Some of the private sale ones have plates that are dead.

Dodgy doesn't begin to cover it.

Until last month they got paid their fares from Uber no questions asked less Uber's magnificent cut. No tax involved. Tax all up to the driver. They could legally turn over up to 60 grand without registering for sales tax and then simply forget to declare any of it and IRD was none the wiser.

Now Uber takes sales tax and gives more than half of it back to the driver as credit for the sales tax assumed paid on expenses. The rest they pay to IRD. Stupid generous but it gets them into the loop whether they like it or not. Apparently not.

 

[petercc9]

If mapping adds 10% safety and 30-40% ride quality it’s probably a no brainer to do it.

For robotaxis, hd-maps probably helps a lot with pickups and drop offs too.

I agree with you 100% regarding today’s AI reliability and capability. ML alone is not yet ready for safety critical applications.

For example, It’s a lot safer to measure a distance with a laser than guessing in from a 2d image using ML (but it’s gotten quite good at it at least in optimal conditions).

It's too expensive...even Huawei gave up...and they had widespread 5G map streaming. Sounds like this was a dead-end and Tesla was correct to focus on vision while others wasted time with it.

Huawei scraps expensive precision maps in self-driving cars

Chinese tech companies seek alternative navigation systems

asia.nikkei.com

"High-definition maps are very expensive," said Huawei senior executive Yu Chengdong at Auto Shanghai show on April 16. "We have been collecting data in Shanghai for one or two years, but we still have not been able to cover all 9,000 kilometers of the city's roads."

The survey cost for a map with 10-cm accuracy is 10 yuan ($1.44) per kilometer, but for accuracy to 1 cm costs up to 1,000 yuan, according to a 2021 white paper by industry group China Industry Innovation Alliance for Intelligent and Connected Vehicles.

 

[Daniel in SD]

It's too expensive...even Huawei gave up. I think this seals its future fate.

That's only $1.2 million to map all the roads in Shanghai to 1cm accuracy (which seems like overkill!)
How does that compare to taxi revenue for Shanghai?

 

[QtownTaxiGreg]

I dunno. Tech moves on.

All the new cars running around videographing their surroundings from ever so slightly different ride heights and road positions, reporting back to some giant brain computer at Tesla HQ over the free month of 'super connected subscription', whatever that is, would build a handy 3D map pdq.

 

[brkaus]

I dunno. Tech moves on.

All the new cars running around videographing their surroundings from ever so slightly different ride heights and road positions, reporting back to some giant brain computer at Tesla HQ over the free month of 'super connected subscription', whatever that is, would build a handy 3D map pdq.

They don't need cars on any free subscription to collect whatever data they want (assuming someone checked the I approve button at some point).

 

[EVNow]

That's only $1.2 million to map all the roads in Shanghai to 1cm accuracy (which seems like overkill!)
How does that compare to taxi revenue for Shanghai?

Time, rather than money ?

 

[MarkDave]

Hi, all --

But back to the topic at hand.

What could this plausibly be? Any radically different hardware poses a challenge to the "only Tesla has the data" story as well as the "every Tesla has robotaxi capabilities" story. A business model presentation seems like it would only give rise to derision. I've been following the thread but am too lazy to review it; was any kind of consensus ever reached?

Yours,
RP

Data being the answer belies a huge (probably purposeful) misunderstanding on behalf of Tesla.

AlphaGo was good, it was trained on humans playing Go.

AlphaZero is transcendent. The Zero refers to 0 human data used in training. AlphaZero didn't get better than AlphaGo by being fed more data.

Go is a game that has more possible states than the number of atoms in the universe.

Driving is somewhat simpler.

For Tesla, data is used as a "straw man" of sorts to justify their public beta testing so they can really claim the revenue moreso than actually advance the state of their system. Of course, some data helps, more is better, but only to a limited extent, and the usefulness falls off as data accumulates. The "we need more data" belief also helps to serve as a motivating factor for people to buy FSD, paying Tesla for the privilege of beta testing the software. Or now, supervising it.

The "need" for actual human miles is actually pretty concerning. They really should have robust simulation by now, especially with generative ML tools.

 

[spacecoin]

Data being the answer belies a huge (probably purposeful) misunderstanding on behalf of Tesla.

AlphaGo was good, it was trained on humans playing Go.

AlphaZero is transcendent. The Zero refers to 0 human data used in training. AlphaZero didn't get better than AlphaGo by being fed more data.

Go is a game that has more possible states than the number of atoms in the universe.

Driving is somewhat simpler.

For Tesla, data is used as a "straw man" of sorts to justify their public beta testing so they can really claim the revenue moreso than actually advance the state of their system. Of course, some data helps, more is better, but only to a limited extent, and the usefulness falls off as data accumulates. The "we need more data" belief also helps to serve as a motivating factor for people to buy FSD, paying Tesla for the privilege of beta testing the software. Or now, supervising it.

The "need" for actual human miles is actually pretty concerning. They really should have robust simulation by now, especially with generative ML tools.

No. Sorry to say - this analogy with AlphaZero is silly.

Driving is not simpler for a computer than Go. Driving has an many many orders of magnitude more "board states" than Go and is safety critical.

The whole premise of AlphaZero and AlphaGo was to play millions of games with itself and learn by doing that. You cannot compare a real-world application where the computer cannot learn itself and "play" millions of "games" against itself. There is a real cost to doing that in robotics and in safety-critical, people can be injured and die. You simply can't build a million robots and have them fail in the world.

Autonomous robots isn't fundamentally gen ai "fill in the blanks" problem in the same way NeRF:s aren't really meant for "solving" parking or the BEV view. You need to be sure what's in the blanks, not just dream them, if you care about safety.

 

[spacecoin]

It's too expensive...even Huawei gave up...and they had widespread 5G map streaming. Sounds like this was a dead-end and Tesla was correct to focus on vision while others wasted time with it.

Huawei scraps expensive precision maps in self-driving cars

Chinese tech companies seek alternative navigation systems

asia.nikkei.com

If you run a robotaxi business with the types of sensors that they use, all cars are capable of map upkeep. So it's essentially free once you build the data and ML pipeline for it.

You can't compare that to the Huawei consumer car car where you need to do map generation and upkeep as a separate task. In general people need to understand that consumer cars are different than robotaxis, and the the business model in completely different. In mass produced consumer cars where you need to make millions (if you're lucky), the BOM is all you optimise for. Not safety, not hd-mapping.

 

[DrChaos]

Data being the answer belies a huge (probably purposeful) misunderstanding on behalf of Tesla.

AlphaGo was good, it was trained on humans playing Go.

AlphaZero is transcendent. The Zero refers to 0 human data used in training. AlphaZero didn't get better than AlphaGo by being fed more data.

yes it did---it simulated tons of outcomes and used those to train.

Go is a game that has more possible states than the number of atoms in the universe.

Driving is somewhat simpler.

The span of unknown and unpredictable situations is much larger and simulation doesn't cover it.

For Tesla, data is used as a "straw man" of sorts to justify their public beta testing so they can really claim the revenue moreso than actually advance the state of their system. Of course, some data helps, more is better, but only to a limited extent, and the usefulness falls off as data accumulates. The "we need more data" belief also helps to serve as a motivating factor for people to buy FSD, paying Tesla for the privilege of beta testing the software. Or now, supervising it.

The "need" for actual human miles is actually pretty concerning. They really should have robust simulation by now, especially with generative ML tools.

They do have simulation now. The simulations with generative ML tools also need lots of high quality data to train with.

 

[QtownTaxiGreg]

ML

BOM

Acronyms doing me under. I think I have the rest sussed but these two escape me...

 

[spacecoin]

Acronyms doing me under. I think I have the rest sussed but these two escape me...

Sorry. ML = machine learning. BOM = Bill of Materials (the components and their associated cost).