Fifteen New Tests Reveal Tesla FSD Still Runs Down Child-Sized Mannequins | Hacker News Reader

Fifteen New Tests Reveal Tesla FSD Still Runs Down Child-Sized Mannequins

dawnproject.com

29 pointsVeserv30 comments

The people that can fix it are busy auditing twitter

I posit if those people could fix this problem, they would have fixed the problem in 2016

Here we are 8+ years after FSD introduction, and the thing still doesn't do emergency braking correctly yet.

Yes because they took LIDAR and other sensors away.

Its a data (features) problem, not an algorithm problem at this point. More of the same approach wont yield a breakthrough as is

Tesla never had LIDAR.

But Tesla had RADAR until the chip shortage took it away. An now Tesla is removing ultrasonic sensors (USS).

--------

If the problem couldn't be solved with Vision + RADAR + USS, why do they think they can solve the problem with Vision only?

Its clear that they're just cost-cutting, IMO. If the problem isn't solved either way, might as well deliver a cheaper car to make, with the FSD at $12,000 instead of $5000. Just a simple money-making scheme at this point.

Thanks for the correction

Yes, FSD is just a money making scheme at the behest of the gullible by now

Humans been at it for decades and still can't do it so well either

Humans don't cost $12,000 and claim that the problem is fixed.

The issue is the false advertising and false expectations being set.

EDIT: If this were just some company's experiment, I wouldn't care much. But this is a product that has been sold to the public, with "Full Self Driving" capabilities for the last 8 years. I have different sets of expectations here compared to whatever "beta-software" or "test software" Waymo (or whoever else) is doing.

Linked Youtube video of tests: https://www.youtube.com/watch?v=FgUPOM7oIxo

The video is interesting, seems like the vision system is detecting it but too late and then not actually disengaging.

I'd be interested to see if this can be replicated, as this research group seems more like an anti-tesla marketing company: https://dawnproject.com/contest/

But that's OK - not using additional sensors is cheaper, easier to calibrate and causes less noise in the data so the engineering cost is lower as well. Every complex technology project needs to have priorities, after all.

Sheesh, they even put shoes on the mannequin. Seeing one on the windshield really gets you thinking...

This is just a very poor taste advert for their lidar product.

Can we please not encourage this kind of thing.

What's wrong with the test? Does it not prove that Tesla's self driving will happily drive over a child, making zero effort to avoid it, and continue on it's merry way?

The problem with the test is that it's not a test. It's an advert for a product.

As long as they aren’t faking the test or results then it’s still relevant and incredibly concerning.

They did another similar test that looked like this, and other people weren't able to replicate it. They also have a conflict of interest, and they're very motivated to make it look as bad as possible. They're not disinterested "researchers".

https://www.youtube.com/watch?v=qTEP-DURtkg

What lidar product are you referring to? The Dawn Project and founder are unaffiliated with any lidar company.

Are you perhaps referencing the unrelated test [1] by a unrelated lidar or maybe AEB company that was making the rounds a few months ago? That test was demonstrating the Tesla AEB system and whether it would override the human pressing on the accelerator like the other car. This is unrelated to the Dawn Project tests which evaluate the ability to avoid child pedestrians without intervention as seen in the videos.

[1] https://youtube.com/watch?v=3mnG_Gbxf_w

Not just Tesla - all self driving cars are murder machines. All.

Neither Waymo nor Cruise has ever killed anyone. And they routinely drive in busy cities.

Haven’t killed anybody YET.

By this argument, all human driven machines are murder machines. All.

Does it still need to be said that the guy behind the Dawn Project, Dan O'Dowd, has a bit of a conflict of interest because he is trying to sell a competing product / service?

Is this relevant? Do you think that Teslas don’t actually mow down child mannequins? Are there flaws in the test?

It's relevant yes, because it does give us a good reason to think there are (possibly deliberate) flaws in the tests.

This is a person who is very motivated to attract a lot of media attention (he notoriously did big ad campaigns against tesla just months ago) and make Tesla look as bad as possible. They have no motivation or incentive to perform accurate, boring tests. It's not a disinterested researcher.

The video is pretty clear though.

The hands are off the steering wheel, the accelerator / brake pedals are clear in the Youtube video, and the youtube videos themselves are uncut footage.

What possibly could be "inaccurate" about these tests?

https://www.youtube.com/watch?v=FgUPOM7oIxo

I cannot say. It's possible the vehicle was tampered with, it's even plausible that they tried this a million times with 100 different mannequins and did a montage of edge cases. It's possible it only does this because the mannequin doesn't move like a human. It's possible it only does this at that exact location. I don't know.

What I do know is that other people tried to replicate their former "test" [1] and weren't able to. So there's likely something special going on, but I don't know what.

[1] This test: https://www.youtube.com/watch?v=qTEP-DURtkg

But the tests match the conditions that were replicated at CES 2022 by Luminar Technologies, with a live demo to boot.

This isn't the first report to show Tesla running down Mannequins. In fact, there's many, many, many independent tests that show this happening.

--------

https://twitter.com/TaylorOgan/status/1478802681141645322

This here is a test from a live Demo, with audience members of CES driving the Tesla.

I know the Tesla fanbois had a lot to say about the test (ie: they must have held the accelerator pedal, or whatever). This new test from Dawn Project proves that the pedal wasn't being used.

---------

This isn't a particularly new problem either. https://carbuzz.com/news/watch-a-tesla-model-3-run-down-a-pe...

As far as I'm concerned, this new test is just further replication of what I already knew. Unless you think there's some kind of worldwide conspiracy, spanning from China, to multiple different companies in the USA and around the world, all conspiring to make Tesla's FSD look bad... I don't see how you get out of this one...

Honestly I would trust a test performed by Waymo way more than a test performed by Dan O’Dowd.

It also gives him the most motivation to find flaws with the competition.

The only important question here is whether the test in question is reproducible/accurate.

This must be like Musk walking down the halls at Twitter, taking out fools.

Reply on news.ycombinator.com