[The Verge] Is Elon wrong about LiDAR?

[Carnos Jax]

I suspect you may be right kj. It may in fact be Tesla's belief that LiDAR, ultrasonics, etc. are the lazy man's solution to the 'vision' problem, only utilized because they couldn't figure out software side of things to the degree required.

[kj]

Carnos Jax wrote:
I suspect you may be right kj. It may in fact be Tesla's belief that LiDAR, ultrasonics, etc. are the lazy man's solution to the 'vision' problem, only utilized because they couldn't figure out software side of things to the degree required.

If lidar actually solved the vision problem, who cares whether it's lazy or not? The problem is, it isn't really a solution to the "vision" problem. The data from both cameras and lidar can be used to construct a 3d model of what's out there. The real challenge is how to make the right decisions based on what the sensors provide.

I guess since our brains process this information effortlessly and automatically, it's hard for people to understand how much is going on after the physical information is gathered by our senses.

Here's a real life example. My car will be steering along fine, following the road, etc. but when there is a sort of lane for slow vehicles to pull off into, it freaks out sometimes. It starts to dodge into it, thinking that's where the road actually goes. It hits the brakes. Sometimes it even relinquishes control. Lidar doesn't address that problem. The software needs to recognize what the lane is, and ignore it. How do you program it to do that? The camera sees it, lidar would see it more accurately, but how does the software recognize what it is and respond appropriately? I think that's the more important part of the whole process.

[sekker]

kj - as I posted earlier, the issues we have with FSD on my wife's Tesla are 100% software.

[Tiangou]

sekker wrote:
kj - as I posted earlier, the issues we have with FSD on my wife's Tesla are 100% software.

Cameras don't measure depth/distance effectively, can't penetrate fog, and blow out in direct sunlight. There's no way to program them to get around those limitations. "Ballparking" it is how Teslas slam into emergency vehicles on a regular basis.

[kj]

Tiangou wrote:
[quote=sekker]
kj - as I posted earlier, the issues we have with FSD on my wife's Tesla are 100% software.

Cameras don't measure depth/distance effectively, can't penetrate fog, and blow out in direct sunlight. There's no way to program them to get around those limitations. "Ballparking" it is how Teslas slam into emergency vehicles on a regular basis.
Do we know that's why they slam into emergency vehicles? At any rate, our human visual system does a pretty good job avoiding emergency vehicles, so there's no particular reason lidar is absolutely necessary.

There's a load more to FSD than measuring depth/distance accurately. There's also a load more that has to be done with lidar to use it for FSD. You still have to have software that recognizes objects, their movement, and what to do to avoid them, etc. Lidar would add some accuracy and help with some decisions, but it doesn't just automatically make FSD awesome. That's why a cost/benefit analysis still makes sense. There's still a huge amount that can be done with cameras, radar, and gps to make FSD a lot better. It will probably make sense to add lidar at some point, but that point isn't necessarily now. And honestly, you could add lidar and still have a schist autonomous driving machine. It's a small part of what has to be done.

[sekker]

Tiangou wrote:
[quote=sekker]
kj - as I posted earlier, the issues we have with FSD on my wife's Tesla are 100% software.

Cameras don't measure depth/distance effectively, can't penetrate fog, and blow out in direct sunlight. There's no way to program them to get around those limitations. "Ballparking" it is how Teslas slam into emergency vehicles on a regular basis.
Our Tesla has no trouble with distance accuracy. The issues are like those listed - there is a side road, so the computer SEES THE ROAD and then makes a poor decision to break and/or swerve.

LiDAR would not fix that at all.

I look forward to the day where the software is vision-limited.

[Carnos Jax]

And that's the thing, lidar doesn't bring anything to the table that cameras don't, and that's because of Tesla's software proficiency. And the best testament to that is Tesla's camera based system is performing on par with the lidar/radar/sonic sensor based systems of the other guys.

[kj]

sekker wrote:
[quote=Tiangou]
[quote=sekker]
kj - as I posted earlier, the issues we have with FSD on my wife's Tesla are 100% software.

Cameras don't measure depth/distance effectively, can't penetrate fog, and blow out in direct sunlight. There's no way to program them to get around those limitations. "Ballparking" it is how Teslas slam into emergency vehicles on a regular basis.
Our Tesla has no trouble with distance accuracy. The issues are like those listed - there is a side road, so the computer SEES THE ROAD and then makes a poor decision to break and/or swerve.

LiDAR would not fix that at all.

I look forward to the day where the software is vision-limited.
Exactly. That's a good way of putting it (not vision-limited). My car beeps annoyingly well when I'm approaching cars that have stopped. I've driven for hours behind people on curvy highways and it holds distance really well. I don't like how close it drives to the center line, and I would anticipate slowdowns/stops sooner than it does, but these things are just preference.

Off topic a bit, but the absolute worse thing about the self-driving is the frequent nags. They don't make sure you are paying attention to the road, they make sure you are staring at the screen. And wiggling the steering wheel often disengages the self-driving. Almost too annoying to bear.

[Carnos Jax]

kj wrote:
Off topic a bit, but the absolute worse thing about the self-driving is the frequent nags. They don't make sure you are paying attention to the road, they make sure you are staring at the screen. And wiggling the steering wheel often disengages the self-driving. Almost too annoying to bear.

Yes it’s a shame. It’s a result of the few idiots abusing the system and the all the whining about the abusers self inflicted demise. There’s got to be a better solution. Maybe a more powerful/better AI that recognizes if someone’s paying attention (just like regular humans do).

[Tiangou]

Carnos Jax wrote: Maybe a more powerful/better AI that recognizes if someone’s paying attention (just like regular humans do).

AI is not a solution. It's a problem. AI is not "smart." It's not capable of making judgments.

And you don't need judgments here.

All that you need is a gadget to check a piezo sensor to ensure that your hands are on the wheel and a mm wave sensor to check where your head and eyes are pointed. Move out of the pre-set range and alarms go off.