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]

Makes sense...key being "testing phase." In short, what I take from this is that a company cannot market, sell, etc. a system as L3, L4 or L5...if human supervision is required when operating in those regimes. Of course, your system can be developed with the goal to achieve L3/4/5... and part of that testing / development process will almost undoubtedly utilize human supervision.

I assume "liability" to the manufacturer also comes with an officially marketed / sold L3/L4/L5 system.

Correct.

There is a distinction between testing phase and commercial deployment. Engineers can design/develop say a L4 system. Being able to deploy said L4 commercially in a viable way, where the company is liable for faults, is another matter.

 

[HitchHiker71]

Which do you all think will happen first?

L3 operation for Tesla owners (with a very restrictive ODD, like Mercedes)
or
L4 operation by Tesla owned vehicles (within a small geofenced area, similar in scope to Waymo), while FSD for owners remains L2

I assume eventually one of these has to happen (unless Tesla gives up).

IMHO what we finally now have with FSD(s) is an official non-beta L2 ADAS system. The next goal should, in turn, logically, be an L3 ADAS system. Jumping up to L4/L5 is likely going to take several years - regardless of what anyone thinks/says. Whether it's due to regulatory hurdles, lawsuits from people dying or being critically injured while using the L3 robotaxi system - because yes - it's going to happen - just a question of when, , unforeseen hurdles that require the system to be leveled up (you know - like what has happened for pretty much every prior version of FSD up until this point), hardware upgrades, camera upgrades, or a combination thereof. My issue with Musk pushing robotaxi is that he keeps trying to push L4/L5 as the goal - and while that is the ultimate goal - and sure go "balls to the wall" toward that goal - don't supplant other goals such as getting FSD(s) up to an approved L3 system in the near term. As you said - whether L4/L5 is limited only to the robotaxi Tesla owned vehicles - those are details TBD.

 

[HitchHiker71]

Sure, if we say vision only and HW4 as well.

Rumor mill says the robotaxi will be using HW5 and will also use updated Tesla radar units (like HW4 does). As much as Musk maintains that FSD must be VO (Vision Only), for a true L4/L5 system, I highly doubt this will ever work because humans have inherent limitations and difficulties navigating their vehicles during inclement weather. Any L4/L5 system will need to surpass humans even under the worst inclement weather conditions, and that's only going to realistically occur with radar/lidar added IMHO. FSD12 still has trouble during inclement weather events - even moderate rain - and limits the vehicle to speeds lower than surrounding flow of traffic for example. It's better than 11.4.9 was, but still has inherent limitations due to the cameras having "limited visibilities" during inclement weather. FSD12 is constantly flashing a red indicator on the screen stating that FSD is degraded during moderate inclement weather.

 

[Joey D]

...Any L4/L5 system will need to surpass humans even under the worst inclement weather conditions, and that's only going to realistically occur with radar/lidar added IMHO.

Agree and disagree.

At the limit, a vision only AI system can be "as good as the best human drivers in aggregate." This means that it is possible that vision only AI may be demonstrated to be safer in aggregte than human driving...because it is driving "like the best drivers, in all situations, at all times." (Does not get tired / distracted, drives with the same competance even in all geographic areas, etc.)

Sure, if you want the car to have "super human" abilities...such that it can operate in areas where even the best humans would fail...then yes, you need super-human sensing (or networking with other vehicles, etc.). I just don't think that is required for an L4/L5 system to be demonstrated to be safer than human drivers on aggregate.

 

[jebinc]

I will match any bet from any amount of people that by the end of 2026 Tesla won’t have a single robotaxi charging for rides anywhere on public roads.

But, it will be "Mind-blowing"....

 

[jebinc]

I already have a friendly bet with WholeMarsCatalog from 1.5 years back that Tesla wouldn't be any form of Level 3 anywhere by September 2024.
He would allow me to change his Twitter bio to whatever I wanted for a year if Tesla wasn't L3 by then. I doubt he'll honor it.

The official Teslascope account took the same bet one month later. The stupidity is simply fascinating.

He's the biggest shill of Elon/Tesla, so no, he will never allow that!

 

[HitchHiker71]

Agree and disagree.

At the limit, a vision only AI system can be "as good as the best human drivers in aggregate." This means that it is possible that vision only AI may be demonstrated to be safer in aggregte than human driving...because it is driving "like the best drivers, in all situations, at all times." (Does not get tired / distracted, drives with the same competance even in all geographic areas, etc.)

Sure, if you want the car to have "super human" abilities...such that it can operate in areas where even the best humans would fail...then yes, you need super-human sensing (or networking with other vehicles, etc.). I just don't think that is required for an L4/L5 system to be demonstrated to be safer than human drivers on aggregate.

Musk has persistently indicated that unless FSD is materially better than human drivers - typically at least twice as safe and up to a 10x factor ideally - then it's unlikely FSD will take hold and it's unlikely that regulatory approvals will be granted for autonomous driving. It's not enough for the L4/L5 systems to be "as good" as the aggregate human driver profile - it must be significantly safer - even and perhaps especially under less than ideal circumstances during inclement weather for example.

 

[Joey D]

Musk has persistently indicated that unless FSD is materially better than human drivers - typically at least twice as safe and up to a 10x factor ideally - then it's unlikely FSD will take hold and it's unlikely that regulatory approvals will be granted for autonomous driving. It's not enough for the L4/L5 systems to be "as good" as the aggregate human driver profile - it must be significantly safer - even and perhaps especially under less than ideal circumstances during inclement weather for example.

We will have to agree to disagree on this. To be clear, "as good as the best human drivers all the time" to me means "significantly safer than human drivers in aggregate." If a level 4 autonomy has an ODD that excludes "severe inclement weather" then fine. (That is also an indicator that humans should not be driving in such weather in the first place...certainly not at normal speeds.)

Musk has never said (to my recollection) along the following: "Take the theoretical best driver, and assume that driver drives his best in all conditions at all times...FSD must be 10x better than that driver all the time." Musk has said (again, my recollection) "FSD must be on average 10x+ safer than human drivers."

 

[jebinc]

We will have to agree to disagree on this. To be clear, "as good as the best human drivers all the time" to me means "significantly safer than human drivers in aggregate." If a level 4 autonomy has an ODD that excludes "severe inclement weather" then fine. (That is also an indicator that humans should not be driving in such weather in the first place...certainly not at normal speeds.)

Musk has never said (to my recollection) along the following: "Take the theoretical best driver, and assume that driver drives his best in all conditions at all times...FSD must be 10x better than that driver all the time." Musk has said (again, my recollection) "FSD must be on average 10x+ safer than human drivers."

But let's be honest with ourselves; you can't believe anything Musk says, right?

 

[Joey D]

But let's be honest with ourselves; you can't believe anything Musk says, right?

I believe some things, but not everything. I also believe there are many things he says that he truly believes in (in other words, he may be sincere) but in my estimation are overly optimistic.

 

[aataskin]

SAE levels have nothing to do with confidence levels. They are about what the system is designed to do:

• L1 can do either steering or acceleration/braking but not both at the same time.
• L2 can do both steering and acceleration/braking at the same time but is not designed to do all driving tasks.
• L3 is designed to perform all driving tasks on its own in an artificially limited ODD but cannot perform the fallback.
• L4 is designed to perform all driving tasks and fallback on its own but only in an artificially limited ODD.
• L5 is designed to perform all driving tasks and the fallback on its own in a wide ODD of a typical human driver (ie everywhere, day, night, all drivable road and weather conditions).

Confidence determines when the system is ready for deployment based on what ODD or what feature the company is trying to deploy. But the level is not determined by the confidence. The level is based on the engineering design of the feature/system.

A typical L2 system would be lane keeping + cruise control at the same time. but that's it. So it can stay in the lane and maintain a safe distance from a moving lead car at the same time but nothing else. So it cannot move out of the lane to avoid an obstacle, it cannot respond to traffic lights or stop signs etc... Hence, a L2 system always requires a human driver since it needs a human to perform the driving tasks that it is not designed to do. So confidence has nothing to do with the level itself. You could have L2 that is very reliable or you could have L2 that is not very reliable. And it would also depend on the ODD. On interstate highways, with clear lane markings, no construction zones, on a clear day, with light traffic, L2 might go a long time with no interventions because lane keeping and cruise control are the only driving tasks that are needed for awhile. But put L2 in the city where it is not designed to stop for red lights or make protected turns, and it will likely require frequent disengagements. We saw that in the early days of FSD before FSD beta was released.

L3 requires a human to perform the fallback but otherwise can perform the driving on its own. Again, confidence has nothing to do with being L3. You could have a L3 system that almost never asks the human to take over or a L3 system that is designed to ask the human to take over quite frequently because of a very limited ODD.

You could have a L4 system that is not very reliable and as a result, needs a safety driver to intervene a lot. You could have a L4 system that is more reliable and the company decides to deploy it as a driverless robotaxi in a geofence. We see this with some of the L4 systems being tested. Some companies are testing L4 with safety drivers because their disengagement rate is still quite poor. Others like Waymo are deploying L4 without safety drivers in limited geofences because they have confidence the system is safe enough in that limited ODD.

Likewise you could have a L5 system that is not very reliable and requires human supervision or a L5 system that is safe and reliable enough that the company trusts it unsupervised.

On this last point, it should be noted why L5 is so much harder to deploy commercially than L4. With L4, you only need to achieve high safety in a limited ODD, like a small geofence, and you can deploy say a commercial robotaxi service. But to deploy L5 commercially, you need to achieve high safety basically everywhere, all the time.

cc: @jebinc

This is precisely what I thought, and this is why I think a Tesla robotaxi needs to be an L5 while Waymo only needs to achieve an L4 in a geofenced area. Waymo is basically there already while Tesla is, well, an L2...

 

[Joey D]

This is precisely what I thought, and this is why I think a Tesla robotaxi needs to be an L5 while Waymo only needs to achieve an L4 in a geofenced area. Waymo is basically there already while Tesla is, well, an L2...

Tesla does not need to be L5. That may be Tesla's goal, but for Tesla to be successful, they need deployable L4, but with a much , much wider ODD than what Waymo currently operates in.

Put another way, there is a potentially HUGE difference in applicability in different L4 systems. Waymo is currently very narrow ODD. How broadly Waymo is planning to expand that ODD is not clear. Tesla (or anyone else) can achieve a much, much more applicable system, and still "only" be L4.

 

[voyager]

Troy thinks Tesla will simply unveil the compact car as the robotaxi.

https://twitter.com/x/status/1776402318277165467

The smaller/sleeker the AV, the more nimble it is and the more road space there is to take evasive action.
He got that right. Remember that the ONE thing that's keeping regulators worried, is the consequences
when an AV does become involved in a traffic accident. The bigger the AV, the bigger the impact, etc.
Especially with the higher front-ends that MPVs and SUVs nowadays feature.

 

[spacecoin]

Tesla does not need to be L5. That may be Tesla's goal, but for Tesla to be successful, they need deployable L4, but with a much , much wider ODD than what Waymo currently operates in.

Put another way, there is a potentially HUGE difference in applicability in different L4 systems. Waymo is currently very narrow ODD. How broadly Waymo is planning to expand that ODD is not clear. Tesla (or anyone else) can achieve a much, much more applicable system, and still "only" be L4.

90% of the world-wide ride haling market revenue today is in major metro areas (and their airports). This can be/is solved using current technology/approaches; eg Waymo.

From a commercial perspective TaaS and personal car ownership need to be completely redefined to change that equation. My guess is that's 30 years out.

 

[spacecoin]

The smaller/sleeker the AV, the more nimble it is and the more road space there is to take evasive action.
He got that right. Remember that the ONE thing that's keeping regulators worried, is the consequences
when an AV does become involved in a traffic accident. The bigger the AV, the bigger the impact, etc.
Especially with the higher front-ends that MPVs and SUVs nowadays feature.

Like the Cybertruck you mean? It doesn't even pass European pedestrian safety laws/tests.

 

[Ciaopec]

Hey great to find this thread. I have one of the original "robo taxis" a 2016 MS90D. I have trouble with technology and have been unable to connect the car to wifi and because of this I know that I'm missing some software. If anyone could help me along I'm still looking forward to my hands free cross country experience!
TIA

 

[QtownTaxiGreg]

Hey great to find this thread. I have one of the original "robo taxis" a 2016 MS90D. I have trouble with technology and have been unable to connect the car to wifi and because of this I know that I'm missing some software. If anyone could help me along I'm still looking forward to my hands free cross country experience!
TIA

What have you tried so far?

 

[Joey D]

90% of the world-wide ride haling market revenue today is in major metro areas (and their airports). This can be/is solved using current technology/approaches; eg Waymo.

That's great if you can operate and scale your technology beyond 3 cities profitably and quickly using automobiles that can be cost effective. Waymo's approach has not proven to solve this.

 

[spacecoin]

That's great if you can operate and scale your technology beyond 3 cities profitably and quickly using automobiles that can be cost effective. Waymo's approach has not proven to solve this.

No tech has. Waymo is closest by a wide margin.

 

[Joey D]

No tech has. Waymo is closest by a wide margin.

You are correct that no tech has. Waymo being "closest" is nothing but wild speculation.