If, not when.
Even the most cutting edge research today still pales in comparison to LiDAR.
Not who you're replying to and not saying you're wrong, but how do we know this?
That's all the person is saying. Simple animals can use only vision to do what we're using lidar and radar to do. But neither camera, lidar nor radar or any combination of them guarantees that you'll be able to make a computer drive a car in all situations.
For me, intuitively, the problem of reconstructing a distance field from cameras must be way harder than say, trying to predict what a person on a bike will do next, or detect road lines on the road in heavy rain or snow. So it seems very likely that an "AI" capable of driving a car in all situations would be powerful to not need lidar or radar (though I don't see the point of dropping radar, as it gives you some ability to "see" around objects which can make cars better than humans)
Because they already have the data advantage for ML.
They can't add them now to new vehicles because they promised the vehicles back then are only a software update away from full autonomy [1]. Building on Lidar now would mean developing on 2 heavily differentiating stacks. Going back on the promise of old Tesla's being "FSD capable" would introduce a huge liability.
Long story short, Tesla's stance on Lidar determined 8 years ago, without the option to revise the decision with future developments.
[1] Note this has turned into "we'll only have to replace the computer in the car", which is still doable, contrary to adding sensors to the existing vehicle.
Ah so Karpathy will be fired this year. Because they're not reaching L4. FSD isn't even L3 yet.
------
Level 4 _ High Automation
System capability: The car can operate without human input or oversight but only under select conditions defined by factors such as road type or geographic area. • Driver involvement: In a shared car restricted to a defined area, there may not be any. But in a privately owned Level 4 car, the driver might manage all driving duties on surface streets then become a passenger as the car enters a highway.
Example: Google’s now-defunct Firefly pod-car prototype, which had neither pedals nor a steering wheel and was restricted to a top speed of 25 mph.
This is the kind of setup I can't wrap my head around. The car might "require" you to take over when you exit the highway, but it can't exactly "make" you. If you fall asleep on the freeway and the car isn't willing/able to drive at the end of your journey, or in edge cases if you were to pass out, etc. what does it do when it gets to your designated exit, or to the final exit of a designated highway? Are there going to be a bunch of cars all stacked up with flashers on the shoulder by every off-ramp waiting for their people to wake up / quit playing on their phones and engage manual mode?
As you say, city driving is hard, but there are a lot of trucks that cross the US on freeways that are easy to automate.
You're right, though, that the challenge is that it also makes it easier for everyone to opt for cars over mass transit.
But Tesla within a year with no lidar? Yeah, no. Not happening.
-- Elon Musk
He set the milestone.
When the ml stack is capable of leveraging purely camera sensors, Cruise and others like them, own the fleet and can swap out the hardware. Tesla does not "own" the fleet per se. So perhaps its different bets on which cars will still be on the road when the ML threshold is crossed.