Hacker News new | ask | show | jobs
by AlotOfReading 364 days ago
One example from the first day, where the vehicle spends a bit of time driving into oncoming traffic:

https://youtu.be/_s-h0YXtF0c?t=0h7m15s

6 comments

Anyone who owns a Tesla with “full self driving” knows how this is going to go. Teslas using only camera vision just don’t have the sensor package and programming to actually perform self driving. I just don’t see this going well for Tesla as it’s more likely to reveal the weaknesses in their technology than be a showcase.
Given that and given that they know the data better than anyone, what's your hypothesis as to why they would deploy if the data isn't in their favor?
Musk needs to keep hyping the next big thing to justify the insane valuation. So it was going to launch regardless of how ready or safe it was.

Still seems nuts to me, they're pretending Waymo doesn't already regularly drive without supervision.

https://en.wikipedia.org/wiki/List_of_predictions_for_autono...

But if you're correct, then won't it be a massive harm to the valuation once they are at fault in many accidents over the coming weeks? They have the Q2 earnings call coming up next month.

It seems like it would be a really bad idea to put this out there if they know from the data that it will cause lots of accidents.

It depends on how much autonomous driving they actually do. My guess is that they will only drive a miniscule amount in favorable conditions for the foreseeable future. That would maximize the hype:risk ratio.

It's also possible they quietly rely on remote operators to a large extent. The fake robot PR stunt wasn't even that long ago, the company might try the same thing here.

Yeah, well, companies like Enron show that covering up mistakes with other mistakes ends badly. Fake it til you make it works if you make it, blows up otherwise.
So you're saying they developed it up until this level where it can mostly drive itself, but now they've pivoted toward collapsing the company?

I just don't see how this train of thought makes sense.

Same phenomenon as this:

https://www.youtube.com/watch?v=1-5s4JlBesc

Nobody's gonna be the one to tell Elon that the stuff isn't ready.

So you think Elon is not interested in numbers? Or you think Tesla employees are presenting fabricated data to him?

The phenomenon in that video is only possible when management is heavily disconnected from ground truths & data, but that seems the opposite to how Musk's companies are run.

One can only delay the much-hyped thing so many times before the markets get irritated, and Musk isn't great at acknowledging that he screwed up.
So you think they chose to launch it yesterday despite the data telling them that they will be at fault in lots of accidents over the coming weeks, and will presumably be forced to roll it back? You think they decided that's somehow better than delaying?
I mean, if that were to happen, they wouldn't exactly be the first. It's not particularly rational, but "launch very late thing in an unacceptable state and then pull it a few weeks after" is not an uncommon industry pattern. Admittedly, it's unusual for _hardware_ (though not totally unknown; see the Humane AI thing, say).
Pulling something a few weeks later isn't uncommon when it's unknown how a service will go, but in this case they actually have data that pretty much tells them how this is likely to go.
Honestly, I have been hearing this since years now. I was suspicious about it for a long time. But they have been allowing vision based FSD for a long time now and the data shows it's getting better, fast. I am ready to park my skepticism (sometimes cynicism) for some time and give them a benefit of doubt out of optimism. Tesla knows that if things go wrong with Robotaxi launch, they are screwed. They wont' take that risk unnecessarily.
> They wont' take that risk unnecessarily.

Tesla: a company that doesn't take risks unnecessarily.

exactly!
Waymo has had many such incidents like this, but they're still (correctly) considered much safer than human drivers.

Whilst a wheel wobble & veering somewhere it's not supposed to go looks bad, it's very difficult to do worse than the average human driver in terms of safety.

It's actually the opposite. Humans, for all their faults, are amazingly safe drivers. Depending on exactly how you choose to measure things, we achieve anywhere from 5-8 9s of reliability or more.

It's statistically unlikely that we'd see an issue like this on the first day of a limited deployment if FSD was hitting those numbers.

That's true. As part of this deployment they haven't driven enough miles to prove safety, but from their FSD data overall they should have a pretty good understanding of where they stand.
> from their FSD data overall they should have a pretty good understanding of where they stand

They know where they stand when there is a safety driver behind the wheel. I'd expect if that data were really good, they'd be less secretive about it. But still, it says very little about where they stand without that driver.

How many accidents have you been in?
Including being not the driver? Four.

How is it relevant though?

If I was in a car with a human driving that badly, I'd leave ASAP! No-one beyond a learner would traverse in such an uncertain, jerky way. And this is on a day with perfect visibility and no adverse conditions that are visible in the video.
But there was no traffic. It changed its mind, but it handled it pretty perfectly otherwise.

People have some sort of beef with Elon and thus FSD, but Waymo has made WAY WAY worse mistakes

Here's a human doing the same thing in the same video for the same reason: https://youtu.be/_s-h0YXtF0c?t=906
They shouldn't be on the road either. That's not a positive argument for the Tesla experience, it's a negative one for whoever was driving that car.
I don't like what that person did, but I don't think their license should be taken away for it.
I believe that driving is a privilege, not a right.

However, I'm not saying that their licence should be taken away (we've all made mistakes - I should have made that clear), but if that is representative, it is dangerous. I rode motorbikes for many years, and you can spot people who are dangerous very quickly from their 'car body language'. And an error like that in another circumstance could kill someone.

I knew people whose driving had deteriorated like this (my late mother and her friends spring to mind). They refused to accept they were not capable any more, and she used to do things that terrified me. I had discussions (kindly!) about her driving standards and errors and she refused to accept that she was making the errors she was. And they were not as bad as what I saw in the video.

And so it’s ok to break traffic rules? In my country, as a human this will get you easily a one month license suspension.
No, just a bit of perspective. I'm sure the video record of your car would be squeaky clean with no violations whatsoever.
If I said I've never had a driver's licence, can we get back to the topic at hand?

What's the thesis here, humans are imperfect drivers and therefore we should accept self driving cars driving into incoming traffic, and if anyone objects, focus on their faults instead?

We allow imperfect humans, we should allow similarly imperfect robots. The robots will get better. The humans, in aggregate, will not.
Humans on the aggregate can definitely get better, that's why Norway has 2.14 / 100k traffic-related deaths per year and the United States has 12.84 / 100k.
It's not a binary. That particular case is pretty egregious since the car signals to turn, gets into the turn lane, actually starts turning, and then swerves back (across the oncoming traffic lanes in the middle of an intersection!) and tries to get back into the correct lane to drive straight, but only manages to achieve that after driving for a bit in the lane for oncoming traffic. That is a major fuckup. And this all happens on a road with fresh clear marking, on a sunny day with perfect visibility.
License should be just revoked instantly for the sake of Tesla. At least with not fully autonomous I can override it. They need to get a feedback that they are not ready.
Texas just passed a law to license autonomous driving, but does not take effect until this fall.