[stretchwithme]

I think it would be wise to physically test as many corner cases as possible under extreme conditions. At night, in the snow, going down a hill, birds flying across the road at the same moment a baby robot crawls on to it.

[nomel]

This, obviously, is one of the ongoing efforts and goals of Tesla AI, and the reason they collect so much data. There's some talk about it in the Lex Friedman podcast(s)[1].

[1] https://www.youtube.com/watch?v=cdiD-9MMpb0

[karlgkk]

Fine, that's great and all if you're into that.

But if that's what you need to build a FSD product, then you shouldn't be releasing the existing FSD product onto public streets.

[nomel]

Since, Waymo has crashes to, is there a crash-free self driving company? Is self driving something that should not be pursued? Is there a way to achieve it without real world use cases (requiring attentive drivers of course)?

[karlgkk]

Man, why are Tesla fanboys like this... I said /existing/. I do believe Tesla will pull it off, and it's a noble goal, just their track record for safety with this tech has been far from ideal, so far. Honestly, it looks like they're going to get away with it, which makes this all a moot point..

But, let's lay out some facts.

First of all, as far as I can tell, Waymo has not had any crashes at fault! It has had some safety issues, but its crash rate is substantially lower than FSD per mile. Even if it has had crashes where it was at fault, we're talking 1 or 2 at-fault crashes over 40 million + miles!

Meanwhile, Tesla's latest and greatest plows into lamp posts in non-adverse conditions. Fantastic.

Second of all, unlike Tesla, Waymo accepts responsibility and liability for crashes caused by its software. This is actually really crucial, since right now Tesla is pushing that onto the drivers, while selling it as a perfectly safe beta piece of software that is "Full Self Driving". Oxymorons all over the place.

> Is self driving something that should not be pursued?

I never said self driving shouldn't be pursued. What I'm saying is that Tesla's FSD implementation should not be used on public roads, as it currently stands.

> Is there a way to achieve it without real world use cases (requiring attentive drivers of course)?

Waymo created test tracks and then logged many hundreds of thousands of miles with cars with human drivers actively monitoring the system. This isn't perfect, of course, but they had a substantially better safety rate than Tesla

In summary: the issue isn't if Tesla can do it or not (I'm leaning towards "yes"), or if it's worth doing (it is). The issue is /how/ they're doing it and /who/ they're exposing that risk to.

[nomel]

I'm not interested in, nor do I own, a Tesla. I don't directly own stocks in Tesla. My employment is in no way related to Tesla, but it is in finding corner cases for real world problems.

It's difficult to address your response, since your interpretation of my question is so misaligned with where my question was coming from.

> What I'm saying is that Tesla's FSD implementation should not be used on public roads, as it currently stands.

This is the catch 22 that my questions were framed around.

Note that Waymo (1 per 2.7M) [1] and Tesla (1 per 6M) [2] have significantly lower accident rates than bare humans (1 per 0.36M), with FSD having 2x lower that Wayme. This suggests that running as is will prevent accidents.

[1] https://theavindustry.org/resources/blog/waymo-reduces-crash...

[2] https://www.tesla.com/VehicleSafetyReport (this comparison may not be exactly equal)

[dkenyser]

I'm assuming OP is suggesting that Tesla needs to test these conditions, not the end user on a public road where innocent lives are at risk...

I could be wrong though.

[nomel]

> Tesla needs to test these conditions

What are "these conditions" exactly? Tesla has 1.3 billion miles of data (impossible to collect without crowdsourcing far beyond the number of Tesla employees). There's probably thousands instances of something very similar to the conditions seen, but something made this a corner case/failure. Or, it's just not possible with the current tech.

You can't opt out of data collection when self driving is enabled. Tesla is aware of every disengagement, with the sensor data categorized and added to their tests, if it's found interesting. They also include synthetic data to manufacture scenarios. They are testing everything they can, and there's a chance this will be added to their tests.

> where innocent lives are at risk

I think a separate permit should be required for self driving, with that permit easily revoked for attention violations. Luckily, in the eyes of the law, the self driving doesn't exist. Driver's still goes to prison if they hit someone.

[TechDebtDevin]

Yeah, they also really need videos of the inside of your garage and videos of you with your family in the garage while the car isnt running apparently.