- A Tesla owner just repeated Mark Rober’s fake road wall test.
- In some tests, they achieved the same result and in others, not.
- The differences appears to be due to newer versions of FSD.
Earlier this week, Mark Rober sparked off a giant online battle by testing autonomous driving tech. In a long video, titled Can You Fool A Self Driving Car?, he set lidar against optical systems like the ones Tesla uses. The result? Instant backlash, praise, confusion—basically, the whole internet lost its mind. Now, someone else has re-run the same test, and, unsurprisingly, the results are both familiar and a little bit different.
In short, lidar tends to see more clearly, and more accurately, in certain situations compared to optical systems. That shouldn’t be much of a surprise, as after all, it’s a high-definition radar system that can sense objects in complete darkness.
Nevertheless, when Rober’s video highlighted Tesla’s failure to detect a wall that looked like a real road, fans of the brand came out with their pitchforks, so to speak. To their credit, Rober’s test didn’t use Full Self-Driving (Supervised) but rather Autopilot.
More: Tesla Autopilot Smashes Through Fake Road Wall While LiDar Lexus Stops Like A Pro
That’s where Kyle Paul, a Tesla owner himself, comes into play. He decided to rerun the same test with the same general parameters, but this time using FSD rather than Autopilot. He printed out his own wall that looked just like the real road it sat on and drove his Model Y up to it multiple times.
In every test, the Tesla failed to see the wall until he was literally inches away from it. As Rober suggested in an interview, it’s plausible that the ultrasonic parking sensors noticed the wall rather than the autonomous driving tech.
That said, Paul then switched things up by bringing in a Cybertruck to run the same test. Interestingly, it passed the test with flying colors by stopping on its own every time it began to approach the wall. What’s the difference? Other than the obvious fact that FSD is more advanced than Autopilot, the Cybertruck was on Tesla’s latest FSD hardware called HW4. The Model Y, a 2022 year model edition, wasn’t as it was running HW3.
The Missing Pieces
Notably, some in the comment section pointed out the tests that Paul didn’t do. For instance, he didn’t test FSD with a mannequin or in the rain—two factors that could offer a more realistic sense of how the system performs in everyday conditions.
Nevertheless, this should at least help calm the noise around Rober’s video. There’s clearly some truth to the criticisms, and those who continue to challenge Tesla’s approach to autonomous driving aren’t entirely off the mark.