-2
AI driver: review of Mark Rober's LIDAR showcase (youtu.be)
submitted 2 weeks ago by Eideen@lemmy.world to c/technology@lemmy.world
18 comments fedilink hide all child comments

I think AI driver give a good summary of some of issues with the Mark Rober showcase of LIDAR.

  • Some of the bias Mark Rober friendships with the LIDAR supplier.
  • Not using Tesla full self driving
  • Being so nervous that he accidentally disables autopilot, twice.
  • The one sided view in LIDAR, with not show examples of LIDAR have issues.
  • Give a way forward to do the tests with FSD
  • Mark Rober not giving the test his full attention, fun vs scientific process
you are viewing a single comment's thread
view the rest of the comments
[-] OptimusPrimeDownfall@discuss.tchncs.de 2 points 2 weeks ago

The friendships with LIDAR suppliers aside (always a problem with these kinds of things),

  1. the full self driving wouldn't have saved the wall.
  2. He's telling a story. Disabling autopilot because he's nervous is good storytelling because it's science communication, not science. Its also cause he knows it's super likely to fail.
  3. He did show LIDAR having troubles - the heavy rain. Its just good enough to peak through it to stop the car before it hits the kid. it's just that the self driving ALSO has trouble with this, and in fact worse troubles. LIDAR has trouble in places where the light is going to be blocked, a situation that a pure camera solution on a Tesla is not going to solve. Not unlikely to solve, NOT GOING to solve. In the real world, if light can't penetrate through a thing twice, it's unlikely to only do it once.
  4. Its not supposed to be scientific, it's science communication. Science is testing and retesting. He did one test and called it good.

Yup, LIDAR isn't a silver bullet for every situation to do sensing. But it's a damn sight better than pure cameras. And Musk would have known this if he was a good engineer. But hes not. Hes a spoiled, rich, apartheid-loving, racist asshole that thinks he's a good programmer and engineer.

[-] Eideen@lemmy.world 0 points 2 weeks ago
  1. Until someone re do the test, we will not know.
  2. yes if you do it knowingly, then redo it without talking control.
  3. Yes and no, he did put it down as a win.
  4. Mark is an engineer and has a good history of using scientific process. He has previously shown that it can be combined.

I know sadly that Tesla has lost a lot of respect due to Elon, especially over the last 4 years.

Engineer is in large part about balancing cost vs features. Yes the will be cases where LIDAR is better, but it comes at high cost. Think about it what is break over point the where the 0.1% edge case where LIDAR will do better, justify the cost.

If the test that Mark show to unexpected drivers I think many humans drivers will failed. Like spoting a dark stationary object in fog or heavy rain, is very difficult.

[-] OptimusPrimeDownfall@discuss.tchncs.de 1 points 2 weeks ago* (last edited 2 weeks ago)
  1. FSD won't solve the "the sensors don't detect the wall" issue
  2. That was not the point of the video.
  3. It did stop, yes. That is a win. Didn't kill the kid. The Tesla didn't. It killed the kid. That's a fail. What's your point?
  4. Why would you run a test multiple times with the high potential for damage to a very expensive vehicle when it's not supposed to be published in a journal? You talk about cost vs features for manufacturers but don't apply that same logic to Rober. Its also a one-off thing for Rober, so the math is better for him than the "cheaper, OK with kids dying" method of Musk.

I think you're underestimating how many situations a LIDAR system will be better than an all camera system. Its also a tradeoff in human lives. I'd rather it be slightly more expensive up front and not have kids die than cheap and kill kids.

And yes. Self driving should be better than humans. Cause humans suck as drivers. We have two cameras in one location in the car. Self driving must be better and make up for the limitations of humans. Cameras dont make up for them and thus are a terrible replacement for humans.

[-] mbirth@lemmy.ml 1 points 2 weeks ago
  1. Until someone re do the test, we will not know.

https://youtube.com/watch?v=9KyIWpAevNs

[-] Bell@lemmy.world 0 points 2 weeks ago* (last edited 2 weeks ago)

Why does no one mention why they said they dropped RADAR: Who do you trust when vision conflicts with RADAR? There were constant problems on this point. LIDAR is another redundant and possibly conflicting input.

[-] OptimusPrimeDownfall@discuss.tchncs.de 2 points 2 weeks ago* (last edited 2 weeks ago)

Yup, but that's going to be true in every environment. Conflicting or noisy signals are always going to be there when you have multiple sensors. Theres going to be conflicts between pure camera systems - what if a camera sensor goes buggy and starts putting out data that says there's always a thing to the left?

More systems giving data to establish ground truth is better. Dont Boeing yourself into thinking that one sensor is good enough - that's how you kill people.

Edit: you also know how they're doing the depth detection with cameras? With AI. You know, that thing that we keep having troubles hallucinating data with. So the data it's getting from the depth subsystem isnt ground truth, it's significantly worse and could be completely wrong.

[-] Bell@lemmy.world -1 points 2 weeks ago* (last edited 2 weeks ago)

No, the camera data is combined into one visual "truth" I would bet. There was a fatal accident where the radar didn't see a truck because it shot under it. At the time the radar was more trusted than the camera. The only way to solve that was to only have one sensor, and radar could never be the sole input.

[-] OptimusPrimeDownfall@discuss.tchncs.de 2 points 2 weeks ago

You clearly have no clue what you're talking about.

this post was submitted on 20 Mar 2025
-2 points (37.5% liked)

Technology

68442 readers
2415 users here now

This is a most excellent place for technology news and articles.

Our Rules

  1. Follow the lemmy.world rules.
  2. Only tech related news or articles.
  3. Be excellent to each other!
  4. Mod approved content bots can post up to 10 articles per day.
  5. Threads asking for personal tech support may be deleted.
  6. Politics threads may be removed.
  7. No memes allowed as posts, OK to post as comments.
  8. Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
  9. Check for duplicates before posting, duplicates may be removed
  10. Accounts 7 days and younger will have their posts automatically removed.

Approved Bots

founded 2 years ago
MODERATORS
L3s@lemmy.world
enu@lemmy.world
technopagan@lemmy.world
L4s@lemmy.world
L3s@hackingne.ws
L4s@hackingne.ws