Camera Based Autopilot Fails Wile E. Coyote Test

[Gareth]

Avoiding the clickbait title, aka Tesla Fails Wile E. Coyote Test.

Mark Rober uses lidar to map out Space Mountain, then tests a lidar based car safety system vs Tesla's camera based system (via Autopilot as the automatic emergency braking didn't do as well). Seems like a fair and balanced test to me, showing the benefits of lidar in some real world situations and some far-fetched. Although not sure if there are any benefits to point out for a camera based system, other than aesthetics of the vehicle itself (i.e. no lidar hump).

https://www.youtube.com/watch?v=IQJL3htsDyQ

[jonny]

The Tesla is smelling distinctly un-rose-like after these tests.

[Black]

Good video, thanks.

[Ken Sp.]

There is substantial skullduggery in the altered video.
Entertaining, but not accurate.
Discussed earlier in this post.
https://forums.macresource.com/read.php?1,3007531

[ztirffritz]

I usually like Mark’s videos but this is clickbait. You can’t really call it a failure if a human would have failed the test too. If someone tried to pull that stunt on me while driving without telling me I’d probably have done the same as the car. And driving into fog or rain so thick you can’t see is the same. A human driver would have done no better. LIDaR is not perfect but it does indeed offer some advantages. As does FLIR. To maximize benefit though Tesla has chosen to leave LIDaR out of the equation because historically it costs too much. It’s getting better and cheaper though. I wouldn’t be surprised if it gets integrated into Teslas at some point.

[pdq]

Ken Sp. wrote:
There is substantial skullduggery in the altered video.
Entertaining, but not accurate.
Discussed earlier in this post.
https://forums.macresource.com/read.php?1,3007531

…or not.

A group of Tesla fans and investors has inadvertently exposed Tesla’s shadiness regarding crashes involving Autopilot by attempting to claim that the advanced driver-assist system was not active in a crash test.

It was not only active, but it also disengaged itself less than a second before the crash—a known shady behavior of Tesla’s Autopilot.

… In NHTSA’s investigation of Tesla vehicles on Autopilot crashing into emergency vehicles on the highway, the safety agency found that Autopilot would disengage within less than one second prior to impact on average in the crashes that it was investigating…

This would suggest that the ADAS system detected the collision but too late and disengaged the system instead of applying the brakes.

Cast in the light friendliest to Tesla, you could argue that returning control to a human operator makes sense whenever any automated control system detects circumstances beyond its ability to handle. But FSD effectively saying "It's all yours, bud" 500 ms before certain impact feels more like an attempt to create legal confusion that favors Tesla in liability lawsuits.

miley-shocked003:

[Acer]

So, the accusation is that he turned off the Autopilot intentionally just to get the viral footage, all while saying he was testing Autopilot? Sorry, but that sounds ridiculous for a long-time, respected youtuber like Rober. Now if it were Mr. Beast.....hmmm.

[Lux Interior]

ztirffritz wrote:
A human driver would have done no better.

Isn't it supposed to be better than a human driver?

I keep hearing arguments that there would be less accidents overall if every car had Tesla tech.

[Ombligo]

Tesla is fast becoming marketing hype. There are other EV cars with better build quality and technology. Tesla's charge slower than KIA, Hyundai, Volvo, Porsche, and BYD. Their auto brake mechanism isn't as good as others. The body style is essentially unchanged since the introduction.

and that is't even getting into the issues being created by the CEO.

[Bill in NC]

just shows the limitation of cameras versus lidar.

as I've mentioned before.

yes, I know lidar was expensive when tesla first came in the scene.

but what's their excuse now?