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!

 

[diplomat33]

That's obviously not what I meant. Waymo is more advanced than Tesla in all aspects that I can think of. It's not like Waymo uses a rules based approach and no NN:s.

Let me put it like this: Are you aware of any technology that Tesla uses that Waymo doesn't? Are you aware of any novel approaches that Tesla invented or uses that Waymo doesn't?

It's my understanding that Waymo uses several end to end networks by the way, for smaller stuff like parking.

I mean, sure, they both use ML and NN. So I guess in a very superficial way, they use the same tech. But how Tesla and Waymo use the tech is very different. The approaches are very different. The NN are very different. The sensors are very different. The maps are very different. How the NN work together are very different. So I think it is a bit misleading to say that they use the same tech because it ignores all the differences in HOW they use the tech and the details in the tech are different (different sensors, different NN etc). It would be like saying that an amateur tennis player and the #1 professional tennis in the world use the same tech because they both use a tennis racket and tennis balls. Sure but they are literally leagues apart.

 

[spacecoin]

I mean, sure, they both use ML and NN. But how Tesla and Waymo use the tech is very different. The approaches are very different. The NN are very different. The sensors are very different. The maps are very different. How the NN work together are very different. So I think it is a bit misleading to say that they use the same tech because it ignores all the differences in HOW they use the tech. It would be like saying that an amateur tennis player and the #1 professional tennis in the world use the same tech because they both use a tennis racket and tennis balls. Sure but they are literally leagues apart.

Sure. Waymo has an existing mission-critical robotaxi business across a handful cities and Tesla is doing driver-assistance marketed as future robotaxi.
It's like comparing SpaceX and Virgin Galactic. not the same business model, not the same requirements.

However, I responded to this, and didn't think it merited more than ten seconds of my time..

Waymo does not have the "brain" that Tesla does, which is arguably significantly more important. Just saying.

Context matters..

 

[DanCar]

... Are you aware of any novel approaches that Tesla invented or uses that Waymo doesn't? ...

Perhaps not so novel:

1. Selling vaporware
2. Promising FSD since 2016 and not delivering
3. Giving us "quantum leap" tech, lol
4. Impressive salesman tactics that pump up the tech yet doesn't deliver
5. Selling very dumb summon as smart summon
6. Mind blowing release after release, yet we don't have much more than adjustable cruise control and lane keep assist.

 

[QtownTaxiGreg]

Mind blowing release after release, yet we don't have much more than adjustable cruise control and lane keep assist

Driving itself around an urban course and apparently able to respond appropriately to most curveballs from other road users is a way bit beyond lane keep assist if you ask me.

 

[Joey D]

That's obviously not what I meant. \

Waymo and Tesla do not use the same tech:

• Tesla uses off-the-shelf cameras. Waymo builds their own in-house customized HD cameras, lidar and radar.
• Waymo builds their own HD maps with their own vehicles. Tesla uses regular maps.
• Tesla trains a deep neural network, end-to-end, from camera input to control. Waymo trains individual neural networks to handle perception, prediction and planning separately.

The tech is quite different.

That's exactly what I mwant when I said Tesla and Waymo use different tech...and you replied "enlighten me."

Can Waymo's approach work on limited scale? (Limited number of expense to own and operate cars in a limited number of markets?) Of course. (It's operating in that capacity today).

Can Waymo's approach easily scale to just about anywhere with very large fleets of profitable vehicles? Can't see it.

Is Tesla's approach scalable? Absolutely. Can Tesla actually reach level 4/5 autonomy with their approach? Remains to be seen, but given the orders of magnitude of increase in AI hardware and software being advanced in the industry...I would say "very likely".

 

[Joey D]

Context matters..

It sure does. In context, again, Waymo does not have the same "brain" that Tesla does. They are using a different approaches: Waymo uses more and varied sensors, smaller, individual nets (many trying to utilize sensor fusion as input), lots of heuristic code to stich stuff together. Tesla's brain is much more akin to how humans work: Eyes (perceive the visual world), "think" (end to end neural net), plan and drive.

This is not controversial.

Again, Tesla still needs to prove that "it's way" can achieve Level 3/4/5 autonomy. But once / if they do...it is more or less immediately scalable. (Provided they train the nets locally enough to drive in a similar fashion as humans do in a given geographic area.)

 

[JulienW]

On 8/8, if he says anything, he'll probably just announce that he meant 8/8/2088.

So you are trying to say that Elon understands the meaning of time? He was just Tweeting and 8/8 sounded cool so he sent it. Now up to the teams to scramble and put together a plausible presentation.

 

[spacecoin]

That's exactly what I mwant when I said Tesla and Waymo use different tech...and you replied "enlighten me."

Can Waymo's approach work on limited scale? (Limited number of expense to own and operate cars in a limited number of markets?) Of course. (It's operating in that capacity today).

Can Waymo's approach easily scale to just about anywhere with very large fleets of profitable vehicles? Can't see it.

Is Tesla's approach scalable? Absolutely. Can Tesla actually reach level 4/5 autonomy with their approach? Remains to be seen, but given the orders of magnitude of increase in AI hardware and software being advanced in the industry...I would say "very likely".

I think you misunderstand why Waymo is doing what they're doing.

It's not like Waymo doesn't have state of the art 3d perception, or a less advanced "brain" than Tesla. Waymo drove as "well" as Tesla does now 7-8 years ago.

• They limit their ODD so they can validate the system to some resonable level.
• They have detailed maps - for safety.
• They have physical measurement of ranges - for safety.
• They have longer range sensing and all of it multimodal - for safety.

Tesla's approach isn't more scalable. They require a human as fall back because their reliability is no where near human levels of performance. They are over-reliant of correct map data. They just blow past stop signs if isn't on the map.

For every release they got to Chucks corner to test. Why do you think that is? Don't they have all the data needed from there by now? How many other hard UPL:s are they testing? They can't test them all, and as you will see, they will not be removing the driver for this very reason until they DEFINE AND LIMIT THEIR ODD.

Waymo don't yet do highways with customers. Even with their sensing capability. Why do you think that is? They can drive there. They've done so for 12 years.

Level 5 autonomy is not real. It's a fantasy. Anyone with some basic understanding of safety critical systems will tell you this. Perhaps in 30-40 years we'll have 90% coverage of the US. Sooner if there are some massive research level breakthrough, never if we don't. Existing approaches are mostly brute force.

 

[Joey D]

I think you misunderstand why Waymo is doing what they're doing.

It's not like Waymo doesn't have state of the art 3d perception, or a less advanced "brain" than Tesla. Waymo drive as "well" as Tesla does now 7-8 years ago.

• They limit their ODD so they can validate the system to some resonable level.
• They have detailed maps - for safety.
• They have physical measurement of ranges - for safety.
• They have longer range sensing and all of it multimodal - for safety.

You keep on saying that Waymo does these approaches "for safety." You ignore humans don't need ANY of these things "for safety." What humans need is to not be distracted while driving, and to have a greater spacial awareness (eyes on the back and side of head). Tesla's approach does in fact supply this.

Tesla's approach isn't more scalable.

Of course it is. Why are you parsing your argument this way? This is not debatable. You can take any tesla manufactured over the past what, 6 years, and drop it on a road anywhere in North America...and it can drive. Supervised yes. Again, the question is of course still open if Tesla will reach level 3,4,5 with their approach. But once / if they do, it is immediately and widely deployable.

You keep saying "it's not scalable because it's not yet officially level 3." Makes no sense.

Waymo don't yet do highways with customers. Even with their sensing capability. Why do you think that is?

2 reasons:
1) Because they haven't programmed the heuristics to be able to deal with what is in effect a completely different set of "rules" that humans use when driving on a limited access highway vs. the city. (Passing, construction zones, merging at speed, etc.)

2) Because if you don't get it right, accidents on the highway are more likely to result in more property damage and bodily injury (due to the speeds being traveled), vs. city driving where speeds are lower.

 

[spacecoin]

You keep on saying that Waymo does these approaches "for safety." You ignore humans don't need ANY of these things "for safety." What humans need is to not be distracted while driving, and to have a greater spacial awareness (eyes on the back and side of head). Tesla's approach does in fact supply this.

Supply what?

Of course it is. Why are you parsing your argument this way? This is not debatable. You can take any tesla manufactured over the past what, 6 years, and drop it on a road anywhere in North America...and it can drive. Supervised yes. Again, the question is of course still open if Tesla will reach level 3,4,5 with their approach. But once / if they do, it is immediately and widely deployable.

You keep saying "it's not scalable because it's not yet officially level 3." Makes no sense.

So you're saying that a Waymo vehicle couldn't be driven with a safety driver anywhere before they added hd-maps and geo-fencing?

This is one of the many ZERO INTERVENTION drives from 2009 - that's FIFTEEN YEARS AGO....

2 reasons:
1) Because they haven't programmed the heuristics to be able to deal with what is in effect a completely different set of "rules" that humans use when driving on a limited access highway vs. the city. (Passing, construction zones, merging at speed, etc.)

Waymo doesn't have a heuristic approach. I can guarantee you they have state of the art NN approaches not available to Tesla.

2) Because if you don't get it right, accidents on the highway are more likely to result in more property damage and bodily injury (due to the speeds being traveled), vs. city driving where speeds are lower.

Yes, and also, you can't just stop on a highway if your safety margin/confidence level goes down. It is forbidden.

 

[Joey D]

So you're saying that a Waymo vehicle couldn't be driven with a safety driver anywhere before they added hd-maps and geo-fencing?

No, I'm saying that a Waymo vehicle will NEVER drive autonomously without adding HD maps, geofencing, and heiuristic code for the specific use case. I thought this was clear?

Waymo doesn't have a heuristic approach. I can guarantee you they have state of the art NN approaches not available to Tesla.

I can guarantee you that there is some level of heuristics involved, and that Waymo's approach uses more than Tesla's current approach.
They have DIFFERENT NN approaches. Different NN Training methods, data curation, etc.

What happens when Waymo's "sensors" detect something that is different than what the HD Maps expect?

https://twitter.com/x/status/1778188663253574035

Yes, and also, you can't just stop on a highway if your safety margin/confidence level goes down. It is forbidden.

Yes and no. You can't just stop...but you can pull over. This is required AFAIK with level 3 and 4 systems. Level 3: if the system asks the human to take over (and the human does not), it must be able to safely pull over (not stop in place). Level 4...if the system loses confidence, it must be able to safely pull over.

 

[aataskin]

I mean, sure, they both use ML and NN. So I guess in a very superficial way, they use the same tech. But how Tesla and Waymo use the tech is very different. The approaches are very different. The NN are very different. The sensors are very different. The maps are very different. How the NN work together are very different. So I think it is a bit misleading to say that they use the same tech because it ignores all the differences in HOW they use the tech and the details in the tech are different (different sensors, different NN etc). It would be like saying that an amateur tennis player and the #1 professional tennis in the world use the same tech because they both use a tennis racket and tennis balls. Sure but they are literally leagues apart.

And.. Waymo has cameras at the front of the car, something that in my opinion is crucial but gets continuously voted down by the fanboys in the F12 forum. LIke, can't keep them clean or some other nonsense like this. And it's Tesla, not Waymo, that relies exclusively on cameras. Go figure..

 

[spacecoin]

No, I'm saying that a Waymo vehicle will NEVER drive autonomously without adding HD maps, geofencing, and heiuristic code for the specific use case. I thought this was clear?

Never is a long time. Not this decade, likely. Neither will anyone else.

I can guarantee you that there is some level of heuristics involved, and that Waymo's approach uses more than Tesla's current approach.
They have DIFFERENT NN approaches. Different NN Training methods, data curation, etc.

Yes. ML currently doesn’t provide any safety guarantees, so it’s required.

What happens when Waymo's "sensors" detect something that is different than what the HD Maps expect?

https://twitter.com/x/status/1778188663253574035

Waymo dont trust maps blindly. Definitely not more than Tesla.

Yes and no. You can't just stop...but you can pull over. This is required AFAIK with level 3 and 4 systems. Level 3: if the system asks the human to take over (and the human does not), it must be able to safely pull over (not stop in place). Level 4...if the system loses confidence, it must be able to safely pull over.

Sure. But Afaik no one is allowed to ”pull over” on a highway, unless there’s an emergency.

Hence you need higher reliability.

 

[HitchHiker71]

I'm not knowledgeable about the differences between Waymo and Tesla FSD. What I can say is that Waymo has had driverless vehicles certified and legally allowed on roads within certain geofenced areas for many years now - approved to do so by the local authorities within those zones. Tesla, at least AFAIK, has never had driverless vehicles on any roads in this country. Tesla is not certified by said local authorities to do so at this time. Will FSD demonstrate the legal requirements via FSD to permit the authorities in these cities to do so? That is the claim that is supposedly coming on or after 8/8/2024 - but the reality remains to be seen. Hopefully Tesla has a good story to tell about all this on 8/8. If Tesla doesn't explicitly talk about the roadmap to obtain these approvals on 8/8 - and simply talks in generalities and hype about what is to come - that in and of itself will be telling on 8/8. Look for details, specifics, on timelines and roadmaps for actual approvals by the appropriate authorities in heavy urban areas where this is almost assuredly going to start for the robotaxi efforts. IMHO, come 8/8, we aren't going to here much of anything about a driverless robotaxi that actually compares to a Waymo driverless vehicle on 8/8, we're going to here about the rideshare network and the system that Tesla will be introducing for commercial rideshare businesses to start using - and we may see some mockups and perhaps even an actual concept car for the eventual robotaxi modular design efforts under way. Based upon the job postings I posted into this thread previously, we'll here a lot about a manned robotaxi system for the next 1-2 years best estimate - with the goal of eventually introducing a driverless robotaxi once all of the bugs have been ironed out. I doubt we are going to hear any specific details about driverless FSD certifications on 8/8. Color me skeptical in this regard.

We can argue semantics about which system is superior. Those in the Waymo camp - that is already actually doing driverless taxis and has been for quite a few years now - they've got real proven evidence of the capabilities of that system. Those in the Tesla camp currently have zero evidence in comparison - since there has never been nor are there any driverless FSD-based Tesla's on the road today - since FSD is only a L2 ADAS system today. The bet is this is going to change soon - it's a big bet to say the least. Whether that gambit proves to be a good bet - I guess we'll get a better sense of things on 8/8. I am skeptical of the timelines based upon past performance with Tesla personally. Eventually it may get there - but IMHO it's still 1-2 years out at best.

For those who pointed out that Waymo may or may not be able to drive on highways in a driverless vehicle - keep in mind that only FSD city streets is using E2E NN today - the highway stack is still the same codebase that was in 11.4.9. In other words, FSD is still stuck at L2 ADAS for all highway driving today. Will this change by 8/8? We shall see. If it doesn't, FSD will be in the exact same boat as Waymo for all intents and purposes, at least until the highway stack is moved to the E2E NN and starts learning how to drive on highways.

 

[Joey D]

Never is a long time. Not this decade, likely. Neither will anyone else.

"Never" is simply logical. If Waymo depends on lidar mapped roads, and the highways are never lidar mapped, Waymo will NEVER reach level 3/4/5 autonomy on the highway.

You seem to be arguing that "if Waymo can't do it...Tesla won't be able to either." I don't get this line of argumentation at all. Waymo's and Tesla's approaches each have pros and cons. In short the single biggest difference: Tesla's approach is more scalable, Waymo's approach is easier to implement with limited scale.

They will both have different obstacles to overcome if the goal is to reach broad "level 3/4/5 autonomy on any road in the world." That is Tesla's goal.

To be clear...I'm not even sure that's what Waymo's goal actually is. Waymo may just be striving for "autonomy limited to specific metro areas, in automobiles limited to fleet managers"...and that's it.

 

[EbS-P]

we may see a prototype but unless there is movement to allow driverless cars in a majority of major markets we won’t see actual production. So…. Here my take.

You only need a new vehicle to reduce ROI time frame. I argue the M3 of MY will make fine robo taxis all the need to do is make them drive by wire. That’s right delete the steering wheel and steering column. No need for a new assembly line. How many special built taxis have ever sold well in numbers outside of the UK?

Two seats makes no sense. 5 seat hatch. That’s what I think.

 

[ItsNotAboutTheMoney]

"Never" is simply logical. If Waymo depends on lidar mapped roads, and the highways are never lidar mapped, Waymo will NEVER reach level 3/4/5 autonomy on the highway.

You seem to be arguing that "if Waymo can't do it...Tesla won't be able to either." I don't get this line of argumentation at all. Waymo's and Tesla's approaches each have pros and cons. In short the single biggest difference: Tesla's approach is more scalable, Waymo's approach is easier to implement with limited scale.

They will both have different obstacles to overcome if the goal is to reach broad "level 3/4/5 autonomy on any road in the world." That is Tesla's goal.

To be clear...I'm not even sure that's what Waymo's goal actually is. Waymo may just be striving for "autonomy limited to specific metro areas, in automobiles limited to fleet managers"...and that's it.

Waymo uses HD mapping but it's not an inherent dependency. It gives confidence. The better the sensor-based model becomes the less benefit there is to the HD mapping.

HD mapping is a cost burden, so naturally you'd want to get away from that and navigate likes humans with simple maps, sensors and local knowledge.

 

[Joey D]

Waymo uses HD mapping but it's not an inherent dependency. It gives confidence. The better the sensor-based model becomes the less benefit there is to the HD mapping.

HD mapping is a cost burden, so naturally you'd want to get away from that and navigate likes humans with simple maps, sensors and local knowledge.

I would argue that Tesla has done something similar. Tesla has vehicles (mules) equipped with Lidar / Radar for the purposes of "training / tuning / validating" it's vision based object model.

HD Mapping is more than a cost / time burden...it's a source of sensor / decision conflict that when it occurs in real-time, it must be dealt with.

What happens today when Waymo's "real-time" perception of the world does not match it's pre-defined map? No easy answer....the best solution is to have no pre-mapping at all...to have enough confidence in the real-time perception to be able to rely on it.

 

[Joey D]

Two seats makes no sense. 5 seat hatch. That’s what I think.

I disagree here. The majority of urban "taxi" rides are one or two people, point to point.

Surely, there is a need for different taxi form factors for different scenarios, but the overwhelming majority of taxi / uber trips today (intra-city) is one or two people.

You only need a new vehicle to reduce ROI time frame.

Mostly agree. For entities (including Tesla) that may want to operate a FLEET of robotaxis as a business, you want the cheapest cost to purchase and maintain. This is where the next-gen platform comes in.

 

[spacecoin]

"Never" is simply logical. If Waymo depends on lidar mapped roads, and the highways are never lidar mapped, Waymo will NEVER reach level 3/4/5 autonomy on the highway.

You seem to be arguing that "if Waymo can't do it...Tesla won't be able to either." I don't get this line of argumentation at all. Waymo's and Tesla's approaches each have pros and cons. In short the single biggest difference: Tesla's approach is more scalable, Waymo's approach is easier to implement with limited scale.

That's not what I am trying to say. I am trying to say that MACHINE LEARNING (current techniques) alone isn't likely able to unless there are major research progress.

You should listen to this for five minutes:

They will both have different obstacles to overcome if the goal is to reach broad "level 3/4/5 autonomy on any road in the world." That is Tesla's goal.

At least it was in marketing? 2019: Robotaxis next year (except to DMV). Now: Robotaxis never? FSD (Supervised) ie a Level 2 with a large ODD.

To be clear...I'm not even sure that's what Waymo's goal actually is. Waymo may just be striving for "autonomy limited to specific metro areas, in automobiles limited to fleet managers"...and that's it.

Waymo's goal is to develop an autonomy system and deploy it as robotaxis, make money. When the system cost is more modest they plan to sell the system to anyone that wants it. It's probably already dropped to 1/10 of when they started already.