5 ms·

They are betting that hardware combination is the fastest to market, given the constraints of today software. When the ml stack is capable of leveraging purely

They are betting that hardware combination is the fastest to market, given the constraints of today software. 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.

afterburner 5y ago
Also you can't sell sexy Teslas if they have ugly lidars on top.
- mikejb 5y ago
  I think it's not just that. When Tesla started to with their FSD journey, they had to determine what sensors they can add to the car. Lidars back then were way more expensive than they are now, and it wouldn't have been feasible to add them at that time. 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.
waffle_maniac 5y ago
Tesla has really fallen behind. I think Karpathy will be fired this year if his team can't achieve at least L4.
- yupper32 5y ago
  > I think Karpathy will be fired this year if his team can't achieve at least L4. Ah so Karpathy will be fired this year. Because they're not reaching L4. FSD isn't even L3 yet.
  - waffle_maniac 5y ago
    “I would be shocked if we do not achieve Full Self-Driving safer than human this year. I would be shocked.” -- Elon Musk He set the milestone.
    - detaro 5y ago
      He told investors the same thing last year. Elon milestones mean nothing.
    - yupper32 5y ago
      He's said that every year since 2014: https://futurism.com/video-elon-musk-promising-self-driving-cars https://futurism.com/video-elon-musk-promising-self-driving-...
    - umeshunni 5y ago
      What comes first? Tesla FSD or the year of the Linux Desktop
  - ghaff 5y ago
    I do expect at least some companies will hit L4 within the decade(?) but it's going to be under limited conditions that won't include urban driving. Which could actually be a very useful capability but isn't the "don't own a car" future that some really are focused on. ------ 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.
    - yupper32 5y ago
      Oh sure, L4 within the decade for companies other than Tesla, totally doable. You could argue Waymo and Cruise are already there with geo limitations. But Tesla within a year with no lidar? Yeah, no. Not happening.
- ma2rten 5y ago
  Tesla has not fallen behind, it's rapidly catching up. It's just not that easy to catch up if you are 10 years behind the competition and handicap yourself with inferior hardware. Maybe you are right and Karpathy will get fired, but at that point it's time to sell your tesla stock.
- Fricken 5y ago
  I think you have to be ahead first before you can fall behind. Tesla never had that problem.
threeseed 5y ago
> when the ML threshold is crossed If, not when. Even the most cutting edge research today still pales in comparison to LiDAR.
- ivalm 5y ago
  Is it really an "if"? I think it would be pretty safe bet that in 100 years human-quality CV object detection will be solved (note, we both know that it is possible AND this doesn't require AGI). So then it's really a question of when (presumably you don't need the full 100 years).
  - ihattendorf 5y ago
    > note, we both know that it is possible AND this doesn't require AGI Not who you're replying to and not saying you're wrong, but how do we know this?
    - ivalm 5y ago
      We know in a sense that very simple animals do it and it doesn’t require decision making (in a sense that LiDAR only helps with perception).
      - mhh__ 5y ago
        So I can teach my dog to drive the car?
        audunw 5y ago
        Your dog can reliably detect objects, judge distance and avoid them. 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)

afterburner 5y ago

Also you can't sell sexy Teslas if they have ugly lidars on top.

mikejb 5y ago
I think it's not just that. When Tesla started to with their FSD journey, they had to determine what sensors they can add to the car. Lidars back then were way more expensive than they are now, and it wouldn't have been feasible to add them at that time. 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.

waffle_maniac 5y ago

Tesla has really fallen behind. I think Karpathy will be fired this year if his team can't achieve at least L4.

yupper32 5y ago
> I think Karpathy will be fired this year if his team can't achieve at least L4. Ah so Karpathy will be fired this year. Because they're not reaching L4. FSD isn't even L3 yet.
- waffle_maniac 5y ago
  “I would be shocked if we do not achieve Full Self-Driving safer than human this year. I would be shocked.” -- Elon Musk He set the milestone.
  - detaro 5y ago
    He told investors the same thing last year. Elon milestones mean nothing.
  - yupper32 5y ago
    He's said that every year since 2014: https://futurism.com/video-elon-musk-promising-self-driving-cars https://futurism.com/video-elon-musk-promising-self-driving-...
  - umeshunni 5y ago
    What comes first? Tesla FSD or the year of the Linux Desktop
- ghaff 5y ago
  I do expect at least some companies will hit L4 within the decade(?) but it's going to be under limited conditions that won't include urban driving. Which could actually be a very useful capability but isn't the "don't own a car" future that some really are focused on. ------ 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.
  - yupper32 5y ago
    Oh sure, L4 within the decade for companies other than Tesla, totally doable. You could argue Waymo and Cruise are already there with geo limitations. But Tesla within a year with no lidar? Yeah, no. Not happening.
ma2rten 5y ago
Tesla has not fallen behind, it's rapidly catching up. It's just not that easy to catch up if you are 10 years behind the competition and handicap yourself with inferior hardware. Maybe you are right and Karpathy will get fired, but at that point it's time to sell your tesla stock.
Fricken 5y ago
I think you have to be ahead first before you can fall behind. Tesla never had that problem.

threeseed 5y ago

> when the ML threshold is crossed If, not when. Even the most cutting edge research today still pales in comparison to LiDAR.

ivalm 5y ago
Is it really an "if"? I think it would be pretty safe bet that in 100 years human-quality CV object detection will be solved (note, we both know that it is possible AND this doesn't require AGI). So then it's really a question of when (presumably you don't need the full 100 years).
- ihattendorf 5y ago
  > note, we both know that it is possible AND this doesn't require AGI Not who you're replying to and not saying you're wrong, but how do we know this?
  - ivalm 5y ago
    We know in a sense that very simple animals do it and it doesn’t require decision making (in a sense that LiDAR only helps with perception).
    - mhh__ 5y ago
      So I can teach my dog to drive the car?
      - audunw 5y ago
        Your dog can reliably detect objects, judge distance and avoid them. 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)