Real FSD preview?

[emmz0r]

We have all read and heard that Tesla says it's about regulatory approval for FSD.
But what about having the features available as Level 2?

The argument for liability is moot since you have Level 2 AP/NoA with nag today , and you can use it where it's not supposed to.

So what are the arguments against having L2 "FSD", that is it operating within restrictions with nags?
For example in Europe you can't go higher than x G in a turn. Fine, just stay within the bounds for the FSD preview as well.

Tesla can't really hide behind regulations in my opinion. They could demonstrate and show what they got.

 

[diplomat33]

We have all read and heard that Tesla says it's about regulatory approval for FSD.
But what about having the features available as Level 2?

The argument for liability is moot since you have Level 2 AP/NoA with nag today , and you can use it where it's not supposed to.

So what are the arguments against having L2 "FSD", that is it operating within restrictions with nags?
For example in Europe you can't go higher than x G in a turn. Fine, just stay within the bounds for the FSD preview as well.

Tesla can't really hide behind regulations in my opinion. They could demonstrate and show what they got.

The features we have now are L2. So we already have L2 "FSD". Sure, Tesla could release all their alpha software now with nags but the software is not ready yet. When Tesla's dev software does become reliable enough and ready for deployment, Tesla does release it to the fleet as L2 with nags, and with the "beta" label.

 

[HenryFarkas]

Here's an interesting article about how China is approaching autonomous vehicle tech.

China’s cities will soon be crawling with self-driving robotaxis

 

[Daniel in SD]

Haha. The argument for liability is definitely not moot here in the US. The US regulates things through liability more than explicit rules and regulations.
In Florida Tesla could deploy FSD as Level 5 today as far as I can tell.

 

[Tam]

...We have all read and heard that Tesla says it's about regulatory approval for FSD.
But what about having the features available as Level 2?...

It's a chicken and egg problem:

In 2019, Tesla reported only 12.2 miles of Autonomous driving on California public streets.

In the meantime, Waymo reported almost 1.5 million miles.

https://www.dmv.ca.gov/portal/uploads/2020/06/2019AutonomousMileageReports.csv

 

[PACEMD]

It's a chicken and egg problem.

Are you saying Tesla is chicken or they have egg on their face/s?

 

[animorph]

I fully expect we will see a full set of FSD features in the car that will function as L2. Regulatory approval will just allow them to remove the nanny and start their robotaxi service.

 

[Daniel in SD]

I fully expect we will see a full set of FSD features in the car that will function as L2. Regulatory approval will just allow them to remove the nanny and start their robotaxi service.

Why not just release it in a state that doesn't require regulatory approval?

 

[DanCar]

I fully expect we will see a full set of FSD features in the car that will function as L2. Regulatory approval will just allow them to remove the nanny and start their robotaxi service.

As others have tried to point out, in the U.S. at a Federal level , and Arizona at a state level, there is no law that prohibits FSD or that requires a nanny or something that would require regulatory approval. Some places like China require permissions, others like Arizona and the U.S. do not.

 

[Tam]

I fully expect we will see a full set of FSD features in the car that will function as L2. Regulatory approval will just allow them to remove the nanny and start their robotaxi service.

I agree with others that it's not a regulatory approval problem.

It's a technical, capability problem.

Tesla has the ability to produce video demo but it is doubtful that owners can obtain the machine's competency to reliably brake for stationary vehicles.

It's just like human will be able to go to mars so that is not a question.

The question is how? Which rocket? When and which spaceships and let's choose one when none of them have been built yet! Elon Musk chose Falcon Heavy Rocket then shelved it and moved to BFR instead!

The same with FSD announced in 2016 with MCU1 and HW2.0.

Now, after 4 years, cars are being equipped with MCU2 and HW3.0

In 2016, we thought it's just a software problem but now we know that it's also a hardware choice too (HW3 not HW2.0).

Then now Tesla is showing us that it has been labeling cars and road signs which might have required more powerful servers such as Dojo Supercomputer too.

Just like in 2016, some might have thought it was just a regulatory approval problem but now, I suspect, it's a competency problem: How can FSD drive more than 12.2 miles demo video into owners' hands without a collision or fatality?

Elon explained that would require HW3, new rewrites, Dojo Supercomputer but I suspect we will need more than that.

Just like going to mars, we currently have no capability to do that and we have to build it. Yes, we can send astronauts the International Space Station but even if we got regulatory approval today for mars, there's no technical way to get there.

And that's the same with Tesla FSD. Yes, the current beta is impressive but even if we got regulatory approval today for Robotaxi, there's no way the FSD can avoid collisions and fatality in some scenarios.

 

[HenryFarkas]

My M3 can't even make a 90 degree turn automatically when on FSD on city streets. It's still unable to take a sharp curve on a country road without my intervention so all this talk about FSD being delayed by regulators is quite premature. Once the car can go from my garage to my destination without any other intervention than my hand exerting a bit of torque on the steering wheel, it will be time to start writing to our representatives about letting us send our cars out to work for us as driverless taxis.And when we do that, we'll have to have a method of preventing the steering wheel from connecting to the tires, something that's impossible now because there's a mechanical connection. Otherwise, car thieves could order a Tesla Taxi, get in the driver's seat when the car arrives, and drive it to a chop shop. That would be inconvenient.

 

[bdub85]

True FSD is still a long ways off IMO. I'm hopeful that in the near future we'll have something that has a greater than 0% chance of getting from point a to point b without any human intervention other than AP nags. How near future? Who knows. There are just so many weird edge cases to figure out.

 

[Tam]

...quite premature...

True!

...car thieves could order a Tesla Taxi, get in the driver's seat when the car arrives, and drive it to a chop shop. That would be inconvenient.

A few lines of codes in the computer program would take care of this.

1) Disable all manual driving controls while on Robotaxi mode.

2) Disable the operation whenever sensors (either camera or weight) detects an occupation at the driver's seat.

3) Car can only go where the routes are paid on the app. Disable manual driving that deviates from the paid routes.

But that does not prevent a conventional method of stealing such as flatbed towing.

That doesn't prevent terrorism as past terrorists were too cheap to buy their own cars (or planes).

 

[spokey]

My M3 can't even make a 90 degree turn automatically when on FSD on city streets. It's still unable to take a sharp curve on a country road without my intervention so all this talk about FSD being delayed by regulators is quite premature. Once the car can go from my garage to my destination without any other intervention than my hand exerting a bit of torque on the steering wheel, it will be time to start writing to our representatives about letting us send our cars out to work for us as driverless taxis.And when we do that, we'll have to have a method of preventing the steering wheel from connecting to the tires, something that's impossible now because there's a mechanical connection. Otherwise, car thieves could order a Tesla Taxi, get in the driver's seat when the car arrives, and drive it to a chop shop. That would be inconvenient.

Don't know about city streets but my MY does the country curves. There are a couple 90s near my house and some a bit more hairpin and it handles them OK. BUT, a little scary as it tend to take them too fast and has uneven braking going in to the curve (braking while in the curve being a bad idea anyway).

 

[Mardak]

It's still unable to take a sharp curve on a country road without my intervention so all this talk about FSD being delayed by regulators is quite premature.

Maybe to help compare different people's experiences without actually sharing exact roads/gps locations, how fast would someone reasonably take these curves?

From my extensive testing, curves with at least a center dividing line normally taken at 20+ mph can be mostly be handled by current Autopilot. Curves or sharp right turns that one would take at 10mph the car definitely fails. And curves in between those speeds, Autopilot will attempt and often take too wide.

The primarily reason for the failures is that Autopilot currently relies solely on the main camera to determine lanes and heavily on what it currently sees. Normally getting closer to something makes Autopilot more confident, but in these cases, it's too close, so by the time Autopilot needs to turn, it has no vision of where you should be going. Both of these aspects are addressed with the rewrite to additionally use the side pillar cameras and fisheye wide camera as well as remembering where the curve is headed.

If Tesla releases no new explicit features with the rewrite, e.g., handling intersections, I expect this curve-taking ability to be one of the most noticeable differences. So assuming no more delays, the majority of Tesla's fleet that has FSD computer (potentially even without FSD software) will be able to take sharp curves without problem within a few months.

Given those timelines, is it still premature to be talking to regulators? Maybe so if the consistency of the rewrite's capabilities is not good enough, but even then it could be rapidly improving.

 

[MY-Y]

True!



A few lines of codes in the computer program would take care of this.

1) Disable all manual driving controls while on Robotaxi mode.
..

I'm pretty sure that my MY has a mechanical linkage for the wheel and brakes. Can't disable that with software - which is a good thing.

 

[Tam]

I'm pretty sure that my MY has a mechanical linkage for the wheel and brakes. Can't disable that with software - which is a good thing.

Yes, they are mechanically linked but they are also controlled by software.

The easiest line to write is to disable the car when the driver's seat/space is occupied. That way, the purposefully stalled car happens even before the car starts rolling.

If the software hijacker learns that trick, they can keep the driver's seat empty and wait for the car starts to roll then jump onto the driver's seat.

There are plentiful of seatbelt/weight/camera detections to activate the automatic disablement of the car even before the hijacker could get onto the driver's seat.