"Can You Fool A Self Driving Car?"

PhotoEye

Active Member
Joined
Dec 29, 2021
Messages
841
Reaction score
410
DE Number
173
I found this piece to be well produced, very funny, and very informative. Over 11 million views so far!

 
It’s also stirred up a hornets nest.
 
A bunch of Teslastans claim it can’t be accurate, despite the fact that it is.

Main complaints:
1) There’s a clip in which AP disengaged and then was re-engaged. Why it disengaged is anybody’s guess, and actually an indicator of Tesla having a problem, since Mark touched neither the brake nor accelerator, but therefore everything must be a lie. 🤷‍♂️

2) It’s not using FSD. Woe is me. Not that FSD suddenly develops magical powers to see through fog or anything with only vision, but something something “FSD IS GREAT”? 🤷‍♂️

3) “if you love teaching kids, why tear down their hero?” 🙄

4) “careful, Elon is vengeful”

That’s basically the summary of the responses from the Tesla corner as far as I can gather.
 
A bunch of Teslastans claim it can’t be accurate, despite the fact that it is.

Main complaints:
1) There’s a clip in which AP disengaged and then was re-engaged. Why it disengaged is anybody’s guess, and actually an indicator of Tesla having a problem, since Mark touched neither the brake nor accelerator, but therefore everything must be a lie. 🤷‍♂️

2) It’s not using FSD. Woe is me. Not that FSD suddenly develops magical powers to see through fog or anything with only vision, but something something “FSD IS GREAT”? 🤷‍♂️

3) “if you love teaching kids, why tear down their hero?” 🙄

4) “careful, Elon is vengeful”

That’s basically the summary of the responses from the Tesla corner as far as I can gather.
Got it! Must be a big nest with over 11 million views.
 
Got it! Must be a big nest with over 11 million views.
Go to X and search “elonmusk markrober” and you’ll get the gist. Basically what I said, plus two others:

1) why make my kids watch other kids get run over by a Tesla, that’s evil (it’s a mannequin ffs, and illustrates the point) 🤷‍♂️
2) obviously mark is a sellout, not an engineer, who only cares about putting Tesla in a bad light, despite the fact that he owns one 🤷‍♂️

In short: lots of people inhaling tons of copium trying to square how Elon could say “vision is all you need” and then discover… it isn’t. Evidently that cognitive dissonance can be painful.
 
Go to X and search “elonmusk markrober” and you’ll get the gist. Basically what I said, plus two others:

1) why make my kids watch other kids get run over by a Tesla, that’s evil (it’s a mannequin ffs, and illustrates the point) 🤷‍♂️
2) obviously mark is a sellout, not an engineer, who only cares about putting Tesla in a bad light, despite the fact that he owns one 🤷‍♂️

In short: lots of people inhaling tons of copium trying to square how Elon could say “vision is all you need” and then discover… it isn’t. Evidently that cognitive dissonance can be painful.
I don't do X, so thank you very much for the summation!
 
Oh I almost forgot my favorite: “These tests don’t matter because people don’t crash in hurricane level rains, heavy fog, or cartoon walls! In fact, it proves Elon’s brilliance for not focusing on these exceptions, and instead only focusing on the real world cases that 95% of us will run into! Cheaper is better because it scales faster, saving more lives than lidar which is more expensive and won’t be in as many cars, so Elon is brilliant!”

🙄

(Lest you think I’m picking on Elon or being political, I’m not; these are the top things that come up, and I am simply reframing them in a way that is, well, skeptical and sarcastic rather than drinking the koolaid)
 
The video was intriguing until the final part. While we all recognize the advantages of using lidar or radar over vision alone, the last example is not a compelling one. Even a human could be deceived into believing it was the road ahead and plow right through it. Would the Tesla have stopped if it had encountered a printed brick wall? Perhaps, but it appears that they deliberately manipulated the system in a scenario that would never happen in real life.

FSD undoubtedly has its flaws and limitations, but even I found the final example to be absurd. They effectively demonstrated the point in the other examples comparing lidar and vision alone.

Best part of the video was using lidar to map out the Disney rides 😂
 
The video was intriguing until the final part. While we all recognize the advantages of using lidar or radar over vision alone, the last example is not a compelling one. Even a human could be deceived into believing it was the road ahead and plow right through it. Would the Tesla have stopped if it had encountered a printed brick wall? Perhaps, but it appears that they deliberately manipulated the system in a scenario that would never happen in real life.

FSD undoubtedly has its flaws and limitations, but even I found the final example to be absurd. They effectively demonstrated the point in the other examples comparing lidar and vision alone.

Best part of the video was using lidar to map out the Disney rides 😂
"Even a human could be deceived into believing it was the road ahead and plow right through it. Would the Tesla have stopped if it had encountered a printed brick wall? Perhaps, but it appears that they deliberately manipulated the system in a scenario that would never happen in real life."

But was it not the case that the LIDAR equipped car DID stop in time? If so, doesn't seem like "manipulation" to me.
 
I think the main argument that LIDAR is too expensive isn't even valid anymore, the Chinese market cars all have modules that are way sleeker and integrated compared to Waymo I-Pace's...

I did watch it, and I thought Mark admitted he might have touched the brakes to disengage Autopilot out of 'self-preservation' reaction? But it's unfortunate, his work is typically much less sloppy and better executed when he does tests.
 
But was it not the case that the LIDAR equipped car DID stop in time? If so, doesn't seem like "manipulation" to me.
Go drive around town and send me a picture of a road where this EXACT scenario is. We don’t live in a cartoon world.

All the other examples proved the differences between LiDAR and vision only with possible real world scenarios. The last example was purely for clicks and it worked.
 
Go drive around town and send me a picture of a road where this EXACT scenario is. We don’t live in a cartoon world.

All the other examples proved the differences between LiDAR and vision only with possible real world scenarios. The last example was purely for clicks and it worked.
I think both conditions co-exist; namely, a valid functional test that does not occur in reality.
 
Go drive around town and send me a picture of a road where this EXACT scenario is. We don’t live in a cartoon world.

All the other examples proved the differences between LiDAR and vision only with possible real world scenarios. The last example was purely for clicks and it worked.
It's happened before! Underestimate humanity at your own peril.


 
Go drive around town and send me a picture of a road where this EXACT scenario is. We don’t live in a cartoon world.

All the other examples proved the differences between LiDAR and vision only with possible real world scenarios. The last example was purely for clicks and it worked.
An adjacent example is the back of a semi truck. Can sometimes match the color of the sky. I don't have a dog in the fight, but the point of the last example like the rest is that edge cases do exist.
 
Go drive around town and send me a picture of a road where this EXACT scenario is. We don’t live in a cartoon world.

All the other examples proved the differences between LiDAR and vision only with possible real world scenarios. The last example was purely for clicks and it worked.
But you see, the point is that LiDAR is not subject to the same limitations as a vision-only approach, or even as a human. You're illustrating the point brilliantly, without meaning to.

Even if this is unrealistic, and even if it would fool a human, it did not fool the LiDAR-equipped vehicle. In order to claim that it is irrelevant, you'd have to come up with a scenario in which a cameras-only approach would similarly succeed where a multi-sensor and LiDAR approach fails. I cannot think of such an example. Can you?

So, the point is: a multi-sensor approach, including but not limited to LiDAR, is safer and catches more scenarios than a vision-only approach, both contrived and realistic.

It seems, to me, that we agree more than we disagree on this one. :P
 
Back
Top