FSD/Tesla driving software

I apologize if this video is too much. I guess I should explain my reasoning here. I do think it's a problem that with no depth sensor or lidar that it can't see a physical wall if for whatever reason the camera gets fooled which is the point of the wall he made in the video. At the same time it also for me randomly brakes like it sees something when nothing is there. Previous adaptive cruise systems haven't done this for me that were more than camera only. This should be proof that the Tesla plan to go camera only didn't work.

My Model 3 is HW3. My Model Y is HW4. Apparently according to a post on Reddit that I saw earlier the HW4 cars have lidar behind the bumper that has not been activated. I have not tried to look for the module. If that is the case I fully believe enabling it assuming the software is properly set up will fix my issues with the system. It's a real damper on an otherwise great car.
 
This isn't a Tesla bashing thread, but this video Mark Rober uses his own Model X to show the severe shortfalls of a camera only system. In case anyone isn't familiar Mark Rober isn't just some influencer getting clicks he actually worked on the Mars rover at NASA, so he actually knows a thing or two, especially concerning equal and fair testing. This should be a wake-up call to the Tesla faithful but doubt it. Long story short Mark paints a Looney Toons ACME style tunnel on a huge styrofoam panel. Teslas camera only system can't process that the picture isn't real and hits the styrofoam block, the lidar does and the Lexus stops. Mythbusters was able to do a similar thing. One of the myths was "Can you fool a security system with different methods." A representative of a large security company on national tv said that "Their fingerprint scanner could not be fooled." Grant lifted fingerprints with scotch tape made a silocone mold and unlocked a fingerprint scanner. If Tesla's camera only system is fooled by a styrofoam block with an image painted on it they aren't close.

 
Last edited by a moderator:
This would fool about 97% of the drivers I drive along with most of the time.

Fortunately, 97% of those 97% are swerving around so much while stuck on their phones that they'll have a real chance at missing it altogether.

So yes, Humanity will prevail by pure statistics.
 
This would fool about 97% of the drivers I drive along with most of the time.

Fortunately, 97% of those 97% are swerving around so much while stuck on their phones that they'll have a real chance at missing it altogether.

So yes, Humanity will prevail by pure statistics.
Bingo. We shouldn't judge autonomous vehicles against perfection, we should judge them against a representative driver. A lot of people die or are severely injured in motor vehicle accidents each year. About 5 people died in my town just this year in avoidable crashes. And it's only March 17th.
 
Let me beat the same drum I always do wrt FSD.
The cheaper and more reliable way to do this is with smart roads and intercar communication on a standardized protocol.
Why Tesla abandoned the use of radar when every cheap car has it for both adaptive cruise control and autobrake I have no idea.
 
Camera only FSD will likely be good enough as a 16 year old driver, but will never be as good as lidar. However please remember that the way we design our roads over the year had changed and will continue to change, so maybe eventually anything difficult for camera based FSD will be changed for public safety. Tampering with road features that help FSD or lidar work will eventually be a felony like messing with stop signs are today.
 
Let me beat the same drum I always do wrt FSD.
The cheaper and more reliable way to do this is with smart roads and intercar communication on a standardized protocol.
Why Tesla abandoned the use of radar when every cheap car has it for both adaptive cruise control and autobrake I have no idea.
1) Other companies have a lot of patents already (Waymo, Uber, Chinese, etc) on Lidar based FSD.
2) Tesla has a lot of patents for camera based FSD and that would be the opposite of 1).

Of course Elon yells the loudest so everyone on Wall Street thinks he is the best..... Lidar will always be better than camera other than being more expensive and looks ugly on the car. That's as obvious as someone color blind will not see the road as well as someone not color blind.
 
I don't think anyone in their right mind thinks cameras are best. It's just that lidars are expesive in general, and the type of lidars required for anything remotely close to real self driving is insanely expensive all things considered (I believe the ginormous things on Waymos are a few grand at automotive builder rates if bought by the boatload), and that the hope was to make cameras as good eventually.

This is close to insignificant in the price of say a self driving quarry dump truck that costs millions already, but accounts for quite a chunk in the price of a passenger vehicle.

And this is where reality takes a turn while muskality flies straight ahead on impulse alone, as we are absolutely not there yet, despite real progress.

And then the real achievements of engineers and teams that are genuinely savvy and talented and are making genuine progress gets drowned in the "If a pregnant woman gives birth after nine months, nine pregnant women will give birth after one month" sales pitch. And there's not even nine of them. And it all rolls to Musklandia from there.
 
Last edited:
I have spent a lifetime implementing controls in factories. If you want a fail safe sensor - you MUST use unlike technologies redundantly. Every type sensor can be fooled at some point.

You would think engineers at these car companies would know this.
 
To be fair, the video used Autopilot rather than full self-driving for the test. Autopilot requires the driver to stay attentive even more.

Basically, he tested the cruise control. My car's cruise control would've plowed over the dummy too
 
Back
Top Bottom