Search results

Charge the car to 100% and use the "Apply Brakes When Regenerative Braking is Limited" mode. The pedal will move significantly when you let off the accelerator, turn it off and it won't move at all and you will coast.

It’s obviously not a sensor problem since you can clearly see curbs on the camera. With end-to-end there’s no way to know if the neural net “sees” the curb. All we know are the inputs and outputs.

It can if the reflectivity of the light and dark parts of the sign are different. But why wouldn’t you just use the cameras? Every AV has cameras. The problem is nobody has figured out how to get the performance necessary out of cameras. Tesla is still hitting curbs.

I honestly have no idea. I can’t understand how it could work in simulation but not in real life.

I’m not talking about Robotaxis where 99% would horrendous. Locals are trying to get way higher than 99% success. I bet they could easily do the turn 99 times in a row. It’s almost the simplest possible left turn from a side street on to a highway. I’ve never seen it have pedestrians. It...

It’s just a physics problem. With good perception you could probably do it 99% of the time with a simple planner that assumes that all vehicles continue traveling with their current speed and trajectory. There are no pedestrians, the cameras can see everything, and you don’t have a problem of...

https://disneyparks.disney.go.com/blog/2021/07/the-hall-of-presidents-sharing-the-story-of-libertys-leaders-for-50-years/

Sounds like 12.4 be more of the same but 12.5 will be when they get the first “9” for Chuck’s UPL.

it stops much farther than 2-3 feet from the limit line. Look at @AlanSubie4Life’s pictures! Not sure why you’re imagining a hypothetical situation where the limit line is placed where you can’t see the cross traffic, that’s not where limit lines are. Creeping is why the guy got the ticket...

The focus on cost makes no sense to me. Nobody has made a useful humanoid robot at any cost yet.

How would the case be different if there was a limit line? The complaint is that FSD stops too far back to properly see traffic!

It is illegal to stop too far back in California. Defendant convicted for stopping 10 feet back. https://caselaw.findlaw.com/court/ca-superior-court/1712024.html

My hypothesis is that it’s trained with simulated data. They stop where the front camera can see the line because otherwise the car “forgets” that it’s there and somehow that causes problems.

Pretty sure nearly every new car already has AEB. Automakers fulfill autobrake pledge for light-duty vehicles

Optimus will also be able to do basic service like tire rotation and rim rash repair.

Something else I didn’t consider is that supercharger utilization is very low. If they deploy an Optimus at each supercharger your car will be able to drive itself to the supercharger and get a charge while you sleep.

Also, robotaxis will not use public superchargers. They will go to depots where Optimus robots will clean and charge them. Another consideration is that because demand varies over the day robotaxis have plenty of time to charge so they don't need super fast charging.

Or they don’t want to burn their sources. If a company is trying to find a leaker they’ll send a slightly different version of an email to each recipient.

Maybe superchargers will not be needed for future models. Miniature fusion generators will be a $10 trillion dollar business. There’s a reason Musk said Tesla would be bigger than Apple and Aramco combined.

Not sure what you’re saying. Elon is definitely the one who has pushed the team to solve Chuck’s UPL. Seems very unlikely they’re devoting so much time to it otherwise. It’s a totally normal left turn. I have no idea where everyone here is driving. I think somehow it must look abnormal on video.

Most collisions at intersections happen when there is a traffic signal. https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/811366 I’m not sure it’s possible to exit Chuck’s neighborhood doing only right turns.

Sep 30th 2022. 😮

Close call at 1:53. Guessing it doesn't recognize moving objects? Not single stack yet?

https://static.nhtsa.gov/odi/inv/2022/INCR-EA22002-14496.pdf In the Warrenton Virginia crash the driver was going 70mph in a 45mph zone. I wonder how that's possible with FSD enabled...

I don't think it's possible. Sure you could probably train the system to recognize any particular attack but humans are inventive. AI is very bad at recognizing things that have never happened before. What advance will allow that? It doesn't seem like more training data will help. How will it...

Tesla: "we count all crashes in which the incident alert indicated an airbag or other active restraint deployed." NHTSA: "A review of NHTSA’s 2021 FARS and Crash Report Sampling System (CRSS) finds that only 18 percent of police-reported crashes include airbag deployments." Tesla: "In the 4th...

Well, if they are, I think it's a waste of time. There have been quite a few attacks on Waymo vehicles and none of them have been anywhere near as sophisticated as radar spoofing. It would not be very hard to make the radar system immune to spoofing. Being immune to jamming is probably much more...

@AlanSubie4Life your chariot awaits!

Bricks are cheap and effective against both human driven vehicles and Waymos. I'd much rather be in the backseat of a Waymo if someone throws a brick through the windshield.

Or they just won't do anything about it. All this concern about adversarial stuff is silly. These are not combat vehicles, there are many ways to stop them. When Waymo makes a military version then they can worry about this kind of thing.

Her reaction time is quite good. She gasps a fraction of a second after the car starts moving. Her response time is not so good. I saw potato video quality on the embedded video but when I went to Xitter it was much better.

Superhuman?

Guess we'll have to wait for GPT-5 level intelligence. I think the problem is you have a white car. P.S. That line is at least 8 feet from the front of the vehicle.

Determining distance to a stop line is not possible for even the most advanced AI.

Looks like here. https://static.nhtsa.gov/odi/inv/2022/INCR-EA22002-14496.pdf The argument is that only 18% of police reported crashes have airbag deployment and Tesla only reports crashes with an airbag or seatbelt pyro tensioner activation.

What does scientific mean in this context? If you used FSD 100% of the time would you have 5x fewer collisions than if you didn’t use it at all? I’m skeptical.

Tesla claims it’s already 5x safer. https://www.tesla.com/VehicleSafetyReport

If enabling FSD is truly 5x safer than driving without it shouldn’t Tesla insurance already be 5x cheaper for people who use it all the time?

It’s not clear how much of FSD is trained from simulation data. Simulated data could be generated for any camera position.

The only data we have is one bad turn and collision reports. Ultimately collisions are what matters once you have a large enough sample size. There are UPLs way more challenging than Chuck’s in Los Angeles. Though I suspect Waymo will avoid them.

How many miles without a ULT related collision is great?

I wonder if previous releases were “validated” and found to have an acceptable failure rate? When you have a 10-20% failure rate it doesn’t seem like “validating” would take very long. I’m curious what the actual process is here. Are there a bunch of inputs to the model that they can tweak? Or...

They told the CA DMV that they’ll start testing robotaxis when they get to 1-2 million miles per driver interaction. https://www.theverge.com/2021/5/7/22424592/tesla-elon-musk-autopilot-dmv-fsd-exaggeration

LOL. Watch the 2019 Autonomy Day presentation. That’s just the way Elon talks. They even had a robotaxi app back then!

Hopefully getting that first nine is the hardest. Maybe the local Tesla community can organize a UPL event to finally get enough training data?

In a separate recall Tesla was supposed to fix "the perceived duration of the vehicle’s static position at certain intersections with a stop sign, particularly when the intersection is clear of any other road users;" And in response Tesla claimed that they "Improved the longitudinal slowdown...

Here is the recall: https://static.nhtsa.gov/odi/rcl/2022/RCLRPT-22V037-4462.PDF As far as I can tell it just says they can't do "rolling stops". It does not require multiple stops. I didn't realize there is another recall addressing flaws that they have still failed to fix...

Tesla has a permit. https://www.dmv.ca.gov/portal/vehicle-industry-services/autonomous-vehicles/autonomous-vehicle-testing-permit-holders/