In the piece — titled “Can You Fool a Self Driving Car?” — Rober found that a Tesla car on Autopilot was fooled by a Wile E. Coyote-style wall painted to look like the road ahead of it, with the electric vehicle plowing right through it instead of stopping.
The footage was damning enough, with slow-motion clips showing the car not only crashing through the styrofoam wall but also a mannequin of a child. The Tesla was also fooled by simulated rain and fog.
That’s what’s confusing me. Rober’s hypothesis is without lidar the Tesla couldn’t detect the wall. But to claim that autopilot shut itself off before impact means that the Tesla detected the wall and decided impact was imminent, which disproves his point.
If you watch the in car footage, autopilot is on for all of three seconds and by the time its on impact was already going to happen. That said, teslas should have lidar and probably do something other than disengage before hitting the wall but I suspect their cameras were good enough to detect the wall through lack of parallax or something like that.
Or it still may have short distance sensors for parking and that if it sees something solid on those it disables autopilot?
Completely disagree. You are assuming the same sensors that handle autopilot are the same sensors that disengage it when detecting close proximity. The fact that it happened the instant before he connected kind of shows that at a very close distance something is detecting an impact and cutting it off. If it knew ahead of time it would have stopped well ahead of time.
The original goal also wasn’t to uncover this, it was just to compare it to lidar per the article. I’m guessing we’re going to see a ton more things pop up testing this claim, and we’re likely to see tesla push an OTA update that changes the behavior so that people can’t easily reproduce it.