Can Tesla FSD Nail the Same Route 3 Times? Is Waymo Really 100X Better?

CNBC recently said that Waymo can go 17,000 miles between critical disengagements, while FSD can only go 170 miles. Can Tesla Full Self Driving at least complete the same route perfectly 3 times in a row?

Tesla FSD v12.5.4 on Hardware 3

[GoatDirt]

There’s a reason why years after testing, Waymo still has zero service here. New York is still solidly human level.

[eastmanwebb]

Your channel is among the most important to me since I am using HW3 and driving in NYC the majority of the time. Thanks for all your hard work. It’s greatly appreciated.

[warrenlee]

i have been waiting your video so bad. thank you so much for the test.

[MrAgility]

Thank you for your videos. Your commentary makes the video so much better.

[mulindejohn]

Waymo doesn't count remote interventions. It only counts if someone physically has to move the car.

[Dr.Octogon]

FSD works pretty well, it just makes occasional mistakes. It will get better, but the comparisons to Waymo, I think, are misleading. Waymo does not work everywhere. So, if you want to make a comparison, they need to be compared on the same roads supported by Waymo.

[Jhagen-txsp]

Wow, FSD was impressive. I can't believe it works that well!

[kottuning]

Keep in mind there's literally been zero FSD training in Manhattan so we're at a huge disadvantage. I have however used it many times in Manhattan off hours flawlessly, during rush hour it really needs some serious training

[sylviawylie]

That was a really feature-packed video on why to never visit New York.

[TheAaronBrady]

I really like what you’re doing here - very thoughtful! The last disengagement could have been a simple intervention by pressing the accelerator. I accept that this shouldn’t have to be done at all, of course. Generally, I find my car learns from these interventions where I press it to be more aggressive. This is most notable when dealing with speed limits - when I press my car to speed up in a certain area, it won’t go so slowly next time.

[abend]

FSD in NYC has to be the acid test. There were far more challenging situations in these three runs than would be normal in most cities - certainly more than would be found in Phoenix where Waymo drives. I live in the Phoenix burbs.
Overall I found the performance of Tesla really impressive. Nice job on your part with the drives and your fair analysis.

[fractalelf]

Comparing Waymo to FSD is simply incorrect to begin with. Waymo is expensive, geofenced, brittle and unscalable. It’s a dead-end.

[xitijdesai]

Regarding Last left turn that you tookover. Next time try to just put some gas and see if the car turns it self. It will be better than disengaging the fsd and simply overriding the gas input.

[Accuaro]

Really interesting to see the progress of FSD, wonder what it'll be like in another 3-5 years.

[PS]

It is unfair to compare Tesla FSD and Waymo. Though you have given so many criticisms of FSD, you have to note that FSD can run anywhere in the US, unlike Waymo that has geo fencing as the map data is fed to Waymo cars while Tesla FSD has no such geo fencing limitations… drop the Tesla FSD anywhere and it will drive there…

[IDRIVE-mklv]

Love your content and how you are pushing the FSD to the limit there. I dont think we can compare Waymo to Tesla. We can't buy a Waymo and use it.
I agree with all your comment on FSD capability, phantom breaking specially the ones without anything around are scary and really poor experience. It will be great to see improvement.

[yl]

Unfortunately, in a Robotaxi, there is no such thing as “noncritical” disengagement. When it disengages for any reason, the rider is stuck and will need remote intervention.

[bradstewart]

0 non-critical disengagements could be considered a perfect score. Not compared to a good human driver, but probably would have compared well to a Waymo. Would love to see a what a Waymo dropped into Manhattan in heavy traffic would do.

[charliedoyle]

A better test that you could do, to make it more like a real AV test, is to drive on random, mostly unique routes in NYC, or perhaps busy areas in the NY Metro, over 100 mile increments.
Keep track of all interventions and disengagements, and critical interventions. It would be lots of work, but you'd have the best FSD tests on YouTube.

In a busy city like that, the difficult scenarios come fast and often, so it's the ultimate street-driving test; and the robotaxi business will only be worth doing in busy big cities, so it's more relevant than testing in light traffic, which most other channels do.
Over 100 miles, or 50 miles if you like, you would get a glimpse of how robotaxi-ready FSD is, which seems to be your goal.
AV companies are looking for the failures, so they seek out new scenarios, not look to confirm how great the ADS is.
A few brief drives, mostly going straight, isn't much of a test.

[erick]

I don't Tesla cares more about non critical disengagements . They think if the disengagement was non critical, than the car would have figured it out maybe few seconds later of the driver didn't intervene.
So wjen there is no driver Tesla will let the car figure out in most situations sooner or later