Tesla Autopilot drove into Wile E. Coyote-style fake road wall in the middle of the road in a camera versus lidar test.

all 37 comments

sorted by: hot top controversial new old
[–] 80 points 2 years ago (15 children)

While most companies developing self-driving technologies have been using a mix of sensors (cameras, radar, lidar, and ultrasonic), Tesla insists on only using cameras.

The automaker removed radars from its vehicle lineup and even deactivated radars already installed in existing vehicles.

I did not know about this. I thought they were also using a mix of sensors, since they originally were. But they stopped putting them in new ones and even disabled them on vehicles that already had them?!

This is some next level stupidity.

  • source
  • hideshow 15 child comments
  • [–] 35 points 2 years ago (7 children)

    A while ago people called Musk out on only using cameras when better options existed(for reasons exactly like this one), and he had a little tanty about it like the normal adult man he is. I have no doubt turning off existing radar features is a spiteful move because he got his feelings hurt.

  • source
  • parent
  • hideshow 7 child comments
  • [+] -21 points 2 years ago (6 children)

    Doubt thats the reason. More like no need to maintain two versions of their software.

  • source
  • parent
  • hideshow 6 child comments
  • [–] 14 points 2 years ago* (5 children)

    No sane person would ever have 2 separate versions of their software for different sensors. It would be 1 software with integration of the different sensors.

    It's not "Cameras OR Lidar", it's Cameras AND Lidar.

  • source
  • parent
  • hideshow 5 child comments
  • [+] -14 points 2 years ago (4 children)

    Thats my point. If the new model only have cameras, then it only makes most financial sense to develop only for cameras.

  • source
  • parent
  • hideshow 4 child comments
  • [–] -1 points 2 years ago* (2 children)

    That depends on how much you think good will toward your customers affects the financial sense.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 34 points 2 years ago (2 children)

    Yeah this is the reason Tesla is massively overvalued. Self drive tech won't be coming from Tesla because Musk interfered and told his engineers to remove lidar.

    Musk is nothing but an investor but he believes his own nonsense about being a genius. He made some good bets but he also made some bad ones (Twitter being one) and Tesla is not going to justify irs crazy shareprice. Even having fallen 50% from peak it is way overvalued.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 14 points 2 years ago (9 children)

    The comments say that the tests weren’t exactly right. Something about not using FSD after stating their were going to.

    I think it shouldn’t matter, the other vehicle using FSD and didn’t murder a bunch of fake children.

  • source
  • hideshow 9 child comments
  • [–] 14 points 2 years ago* (8 children)

    Supposedly there's a difference between 'Full Self Driving' and 'Autopilot' (the later is what Mark used in the video). I'm not convinced there's any meaningful difference for a test like this though.

    Changing the software behind them isn't going to magically give cameras the power to see through visual hindrances like rain/fog, or detect range to a nearby object that appears far away (like a painted wall).

  • source
  • parent
  • hideshow 8 child comments
  • [–] 3 points 2 years ago (3 children)

    Mark states he used it because it was more conservative but commenters are saying it wasn’t on for the wall test.

  • source
  • parent
  • hideshow 3 child comments
  • [–] 4 points 2 years ago (2 children)

    Hmm....

    Taking a close look just now; there's two conflicting shots.

    At 15:34 the blue lines and rainbow effect 'turn on' on the center screen. Then the shot from inside the car shows no rainbow/blue lines during impact at 15:42.

    I then started looking around a bit and found:

    He posted the 'raw footage', which shows Autopilot disengage before impact, but not brake. Unfortunately when you compare the two, it's clear these are two seprate takes: the youtube version had Autopilot engage at 39mph, while the 'raw footage' shows it engage at 42mph.

    Combined with the obvious advertising/conflict of interest; this one's gonna have Mark in some hot water... Tesla may even have a solid defamation case here :/

    https://teslanorth.com/2025/03/16/busted-mark-robers-misleading-tesla-test-sparks-outrage/

  • source
  • parent
  • hideshow 2 child comments
  • [–] 7 points 2 years ago (1 child)

    So the question then remains, can the autopilot disengage if it senses an imminent collision so that the manufacturers aren't held liable?

  • source
  • parent
  • hideshow 1 child comment
  • [–] 0 points 2 years ago

    It certainly could, and wouldn't really surprise me tbh. Whether it does or not would need to be tested further as unfortunately I think Mark has pretty well destroyed his credibility with this.

    It seems pretty clear there was at least two takes of hitting the wall, both times Autopilot was disengaged shortly before impact, and Mark wasn't honest about it.

  • source
  • parent
  • [–] 3 points 2 years ago (3 children)

    There is a huge difference between Autopilot (basically hasn't been updated in like four years) and FSD. Autopilot is basically just glorified lane keeping while FSD attempts to drive end to end (stop signs, traffic lights, etc.).

    Not sure how it would change this test though. FSD is much more vision based than Autopilot.

  • source
  • parent
  • hideshow 3 child comments
  • [–] 4 points 2 years ago (2 children)

    There is a huge difference

    Yes, FSD is built on a completely different software stack.

    FSD is much more vision based than Autopilot.

    No. They're both 100% vision-based.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 12 points 2 years ago

    This is hilarious. It wasn't even a front on photo of the road but a perspective corrected on, for the side on, wile e coyote perspective shot. Head on, the photo would look like it took a sudden, immediate corner, not even a curve. Tesla's software just said 'this is perfectly normal'.

  • source
  • [–] 3 points 2 years ago

    its vision plus neural net system is designed to work like a human driver

    Looks at street death toll.

  • source
  • [+] -29 points 2 years ago (8 children)

    The problem is that at scale you can’t have every car on the road using lidar at the same time.

    They’ll mess with each other’s systems.

    Same with radar.

    So while this seems like a solved answer, it’s not necessarily as simple as it sounds

  • source
  • hideshow 8 child comments
  • [–] 31 points 2 years ago (4 children)

    This is objectively false. Early lidar systems did have crosstalk interference and mitigations have been designed and working for years.

    It's like saying you can't have WiFi cause if all your neighbors do it won't work. As these systems hit the real world mitigations get designed to handle these types of issues.

  • source
  • parent
  • hideshow 4 child comments
  • [–] -1 points 2 years ago (3 children)

    The mitigations work in small numbers, none of them seem like they would work in large amounts of traffic

  • source
  • parent
  • hideshow 3 child comments
  • [–] 3 points 2 years ago (2 children)

    Open your phone and see how many WiFi networks are in your immediate vicinity.

  • source
  • parent
  • hideshow 2 child comments