• Quick note - the problem with Youtube videos not embedding on the forum appears to have been fixed, thanks to ZiprHead. If you do still see problems let me know.

Ed Self-Driving Cars: Pros, Cons, and Predictions

Evaluate Self-Driving Cars on a scale of 1-5 (1 = Terrible, 3 = Meh, 5 = Great)

  • 1

    Votes: 10 6.6%

  • 2

    Votes: 11 7.2%

  • 3

    Votes: 24 15.8%

  • 4

    Votes: 28 18.4%

  • 5

    Votes: 79 52.0%
  • Total voters
    152
  • Poll closed .
Rolfe said:
Click to expand...
There are plenty of circumstances where lidar will fail, or will generate false positives (it's true that the lidar was probably seeing the water rather than the dummy through the water) but lidar vs. cameras is something of a false dichotomy. It's more like vision-only vs. vision-plus-other-sensors.

As for the claim that the industry is moving away from lidar, that's not remotely true. Vision-only is increasingly popular for applications where people don't die if you make an error, because it's cheap and often good enough. I'm hardly an expert in this area, but I've used computer vision for applications where I would have used less sophisticated sensors in the past. It gives you more flexibility, allows you to do things like like detect and count human bodies, etc. And it's also just cheap.

But there's no way in hell I'd do this for a car. It's a decision to blind yourself in certain conditions and to eliminate important redundancies, and that just seems unconscionable to me.

I've seen it suggested that Tesla's decision to go vision-only was motivated by supply chain issues rather than any real confidence in the approach. Maybe--the timing is about right. If that's the case, I'd expect that they'll re-introduce other sensors at some point, with some face-saving marketing cant.
 
Last edited:
And it seems Musk is betting “full self drive” will happen in the same manner the LLM AI came out of huge datasets, but the processing of those is done either in the cloud or using a powerful graphic card running a cut down version. Do the Teslas sold up to now have anything like the required hardware? Or will they need a hardware upgrade?
 
As I understand it, Teslas sold nowadays as having FSD capability won't need any hardware updates. I have heard of old Teslas having hardware updates to make them compatible, or I think I have.

I still don't know what to think about it. It may be Tesla has made a mis-step putting all their eggs in the camera basket, but maybe they haven't. I have heard that there is already data showing that Teslas running FSD have a significantly lower number of accidents than Teslas not running the software, and an even lover number of accidents than the set of all cars not running FSD.
 
Darat said:
And it seems Musk is betting “full self drive” will happen in the same manner the LLM AI came out of huge datasets, but the processing of those is done either in the cloud or using a powerful graphic card running a cut down version. Do the Teslas sold up to now have anything like the required hardware? Or will they need a hardware upgrade?
Click to expand...
The training is done at their supercomputing clusters, but the model needs to be stored and run locally. The most recent on-board hardware is a custom SoC, I think the same one they're using in their newest clusters. Apparently v12 was trained on HW3, but running in emulation on HW4 on new cars.

I would not expect that the hardware currently being sold with Teslas will be able to handle L5 autonomy, if they ever get there. You can't run the v13 software on any Tesla purchased before 2023, although they'll probably roll out a hobbled branch for older hardware at some point. Tesla makes some promises about retrofitting older cars, but they come with some caveats.
 
Last edited:
Youre at a T-junction entering a bigger road which is congested. A kindly driver waves or flashes their lights to say "go ahead, join us", so you wave a "thank you" and edge out.

It happens all the time round here. I wonder how self-driving will handle that (and scores of similar situations)?
 
novaphile said:
Traffic will be worse.

All the self-driving cars will be taking up space on the road, because the owners have not paid for them to be parked somewhere.

This means the cars will be tootling around, looking for passengers, or at best, parked on the side of roads, making all roads narrower.

Even if owners make their cars drive home again after delivering the owners to work, that doubles the amount of traffic, because you're now dealing with four trips per day, instead of two.

Every person who hires a self driving car to take themselves somewhere, is making traffic worse, if they previously would have used a bus, train or bicycle.

No matter how you slice it, self-driving cars make traffic worse, unless they are not used as taxis.

I've already explained how every venue becomes grid locked, when everyone summons their self-driving car at the same time after an event.

I look forward to cycling past the resulting mess.
Click to expand...
Thanks for the correction.
 
arthwollipot said:
Just quickly, what was the LIE!!!!???!???
Click to expand...
Aha!

The lie is that the Tesla wasn't on autopilot when it crashed.

However, that lie is from the Tesla fans.

(See above)

Teslas drop out of autopilot immediately before a crash so that Elmo can say "they weren't on autopilot".

The article above suggests that the car should be stopping before disengaging.

NB. Very dangerous 'panic stops' while driving on autopilot on the freeway, seems to be still a thing.

There was a sped-up official Tesla video showing a flawless drive, that included a dangerous 'panic stop' while going around a bend.

Critics at the time felt that the stop was caused by pedestrians walking on a foot path on the opposite side of the road. The car didn't seem to understand that wasn't in its path while going through the corner.

(My guess, since the system is based on vision, is that current problems are caused by flying debris or insects in flight. I've seen similar behaviour with my car, where the emergency braking was triggered by a dust-devil (AKA 'whirly') crossing the road in front of me, full of dust and leaves.)
 
Rolfe said:
Well, here's someone else who has recreated the stunt. Just another Tesla fan though, so it can be discounted.

https://twitter.com/x/status/1902816452773810409
Click to expand...
ETA: Got it wrong. Skipped around too mcuh with no sound. I missed then that author of video had to break himself because Tesla did not see the wall. He validated original video.

ETA2: Forgot. Later tests had problem of wrong brightness. That might have helped Cybertruck to spot it. That part of video would need a redo.
See this subthread:
https://twitter.com/x/status/1902854254228070777
 
Last edited:
Brake.

The model Y had to be stopped by the driver every time because it wasn't seeing the "wall". However the guy then borrowed a Cybertruck which has newer hardware and/or software, and it did it right. Saw the wall and came to a controlled halt every time. The Cybertruck doesn't have LIDAR either. The conclusion was that Tesla is developing the system and that it can be done without LIDAR.
 
Last edited:
At this point, Tesla is toast...
Their stock is plummeting, their sales have fallen off a cliff, their reputation is horrible and their biggest shareholder is pushing for Musk's involvement with the company to be terminated...
:ROFLMAO:

All this will will severely impact on their ability to finance any further improvements and its doubtful that they will even be able to survive...

As for LIDAR, their prices have been dropping (they were a small manufacturing number run, so of course that pushes up the price significantly)

Be interesting to see if other manufacturers go the LIDAR or camera route- or a combination of both, or neither...

(personally I can't see many wanting a 'self driving car' if it is significantly more expensive than a non self driving version- and reliable emergency braking systems have been in use in trucks for years, so why reinvent the wheel???)

2 mins long, from Volvo- seven years ago...
 
Rolfe said:
Brake.

The model Y had to be stopped by the driver every time because it wasn't seeing the "wall". However the guy then borrowed a Cybertruck which has newer hardware and/or software, and it did it right. Saw the wall and came to a controlled halt every time. The Cybertruck doesn't have LIDAR either. The conclusion was that Tesla is developing the system and that it can be done without LIDAR.
Click to expand...
Did you see my ETA2? At the time of test of Cybertruck brightness of image wall was wrong. Might have made it more visible to system.
(As for break braking mistake, I shouldn't have been posting at night. I know the difference yet a I chose wrong word. Sigh.)
 
You must log in or register to reply here.

Back
Top Bottom