Hacker News new | ask | show | jobs
by ra7 2003 days ago
> While the claim of full Level 5 autonomy without lidar is controversial, it’s possible that most Teslas on the road today will be capable of being upgraded to L5 capability via a software update.

How can you claim the current hardware on Teslas is capable of L5 when they haven’t even demonstrated L4 capability? Just saying it’s capable for full autonomy doesn’t actually make it so, they need to actually prove it.

Also, you (and Elon Musk) talk about Level 5 so casually when all the self driving companies out there strictly claim to aim for only Level 4.

1 comments

Because humans are perfectly capable of doing this task with vision only. Which means it's effectively a visual modelilng task, and is only a question of computation technology, modelling and available FLOPS inside the car. It seems perfectly possible that the hardware in a current Tesla car is capable of this.
Sorry, this is just wishful thinking. Computers don’t have human brains which have evolved over millions of years. So just saying “if humans do it with only vision, so can computers” is frankly weak logic. The technology isn’t there yet to match human brain performance, unless you believe Tesla will achieve several breakthroughs in computer vision and AI to do this.
Who said that the technology was there? Did you even read the previous comments or just jumped onto the hate wagon because of preconceptions?

The question was why it is possible. Not why "it should work today", nor "why it will most definitely 100% happen guaranteed in the next couple of years".

So the technology isn’t there and nobody has gotten anywhere close to achieving those significant breakthroughs in vision or AI, but it’s “perfectly possible” with current hardware “because humans do it”? This answer makes no sense.

The real answer is nobody knows whether current hardware is enough or not, including Tesla. Until they demonstrate it, it’s just a claim. But they can’t sell the cars on FSD promise if they say it.

Humans don't need lidar -> that means all the information is present in the visual channel -> it is at least theoretically possible to do with cameras -> tesla cars have cameras -> tesla cars might have possibility to do it.
> it is at least theoretically possible to do with cameras

Theoretically, yes. But that’s different than confidently claiming it’s “perfectly possible” and that it’s only a question of computation and modeling.

What about the other hardware, the computer doing the actual computations?
Stereo vision. Teslas to this day dont ship stereo front facing cameras. They arent even full color.