Doesn't bode well for FSD...

[McRat]

Inertial directional data was available to cars at least by 1998.
It arrived with stability control.
Some of the early systems actually had to zero out before activating, which took about 100 feet. A warning was displayed, then vanished when the car 'learned' how the steering angle resolver and wheel dias affected where 'straight ahead' was actually pointed.

 

[Zacster]

The ironic part of all this is that the Tesla showroom is right there, actually directly OVER the tunnel on the Brooklyn side, right in the path.

 

[stopcrazypp]

I think you're overreacting.

This problem has been solved, years ago. A simple kalman filter to couple the INS and GPS.

If you're driving in a GPS denied environment for long enough (a tunnel), the car has no idea where you are. Having FSD doesn't mean it needs to follow the nav. It means it needs to not crash into the wall or the car in front it, until it gets out of the tunnel.

There are probably a million edge cases that need to be programmed into it, but showing one of them and saying "FSD is a long way off" doesn't prove anything*.


* I agree FSD is a long ways away, but not due to this scenario.

I had the same thought. Sometimes my GPS goes bonkers (same thing with snapping into a different road), but that doesn't means suddenly I forget how to drive. I don't see why FSD would be designed different.

 

[number12]

US Dept. of Transportation releases autonomous vehicle guidelines, expects full automation in 2025+

This article makes me feel like this debate will last for the good part of 10 years. That makes my head hurt.