4 ms·

Elon said a camera is all you need. Waymo decided to solve the problem using whatever they deemed necessary.

by melling 2y ago

mewpmewp2 2y ago
Still it's good to have multiple approaches going on at once.
- nobunaga 2y ago
  Except one is bound to fail considering the solutions required for the problem at hand. Its obvious which one that is.
dmix 2y ago
> Waymo decided to solve the problem using whatever they deemed necessary. Solving the technical side is only half the problem, otherwise we'd all be flying in Concorde jets. I'm curious to see if Waymo can scale up to mass production of their sensor suite and maintain them economically, where it makes sense as a business. There may be a big business in LIDAR factories and repairmen soon.
- melling 2y ago
  The Concorde is not permitted to fly supersonic over land because of the sonic boom. That means very limited routes. That is a technical problem. Hopefully, NASA solves the technical problem and commercial supersonic flight returns https://www.nasa.gov/centers-and-facilities/armstrong/taming-the-boom/ https://www.nasa.gov/centers-and-facilities/armstrong/taming...
  - dmix 2y ago
    That was mostly a social/policy problem which destroys the economics of it because you're limited to a very small amount of airports. Waymo is currently limited to a small amount of towns for a mix of economic, policy, and social reasons. Probably in that order. Just being sufficiently better than human drivers was never the entire barrier to entry for this to be a real business for Google, at least one that becomes nation-wide or global.
    - melling 2y ago
      How about making a prediction and we will follow up in 5 years. To me it sounds like you’re just being sort of vague. Yes, absolutely everyone knew there were social issues too. Where will self driving be in 5 or 10 years?
      - dmix 2y ago
        Predicting it on a macro level is a boring question IMO, the influence of policy interference and social outrage every time there's an accident - which is never fully connected to the realities on the ground of true capabilities/potential - is basically a black box. Like betting on elections. My interest atm is mostly the economics question of whether LIDAR will scale widely in near term combined with the social risks (ie, millions of cars with $30-50k in sensors, which atm cost as much as the car itself, so basically only e-taxis, no personal ownership + high cost of maintenance) vs cameras being good-enough... that's a far more interesting question. It's pretty obvious LIDAR/ultrasonic or a combination with cameras is superior. So maybe some ambitious person will find some mass-producible middle ground within <5yrs shrug
    - minwcnt5 2y ago
      Calling San Francisco a "town" diminishes what they've achieved. It is the second densest urban center and ride hailing market in the US, and a difficult place for humans to drive. I think being able to effectively replace Uber drivers there means that once they crack freeway driving and snow, the regulatory, economic, policy, and social hurdles will start to fall like dominos. I don't think this is how most people expected this to go. My personal belief was that self-driving cars would be limited to medium-density suburban areas initially (i.e. places where the driving is relatively simple, but there's still enough demand for it to be economically viable). It's astonishing that they were able to move from suburban trials to high density environments so quickly.
      - dmix 2y ago
        Everyone obsesses on the technical side which was necessary but scaling it up country wide as a functioning profitable business entity, even just urban/suburban, requires far more than just heavily subsidized R&D work and experiments as a minor taxi player in a few places. I have doubts the very slow pace of expansion was just a matter of regulations and nailing down the tech. It has to make sense as a business otherwise they aren’t going to dump billions into sensor factories (where each LIDAR suite currently costs around $75k to make), cars, and service infrastructure. Just like how Tesla wasn’t just a of matter of making the first prototypes of electric sports cars. The real magic and barrier to entry was the factories and infrastructure to make the economics work.
ryandvm 2y ago
> Elon said a camera is all you need. This was when I really started to doubt that guy's genius. That was a phenomenally stupid assertion. I can drive with 2 eyes because I have a bunch of other sensors as well as an entire world model in my head with 30 years of driving experience. Is it Halloween tonight? Better drive extra slowly. Did that white semi with the bright background sky just disappear? Probably not. Is that deer going to bolt in front of the car coming the other way? Maybe. Is the road slick because it recently rained here? etc.
- minwcnt5 2y ago
  "640k ought to be enough for anybody" vibes.

Elon said a camera is all you need. Waymo decided to solve the problem using whatever they deemed necessary.

mewpmewp2 2y ago

Still it's good to have multiple approaches going on at once.

nobunaga 2y ago
Except one is bound to fail considering the solutions required for the problem at hand. Its obvious which one that is.

dmix 2y ago

> Waymo decided to solve the problem using whatever they deemed necessary. Solving the technical side is only half the problem, otherwise we'd all be flying in Concorde jets. I'm curious to see if Waymo can scale up to mass production of their sensor suite and maintain them economically, where it makes sense as a business. There may be a big business in LIDAR factories and repairmen soon.

melling 2y ago
The Concorde is not permitted to fly supersonic over land because of the sonic boom. That means very limited routes. That is a technical problem. Hopefully, NASA solves the technical problem and commercial supersonic flight returns https://www.nasa.gov/centers-and-facilities/armstrong/taming-the-boom/ https://www.nasa.gov/centers-and-facilities/armstrong/taming...
- dmix 2y ago
  That was mostly a social/policy problem which destroys the economics of it because you're limited to a very small amount of airports. Waymo is currently limited to a small amount of towns for a mix of economic, policy, and social reasons. Probably in that order. Just being sufficiently better than human drivers was never the entire barrier to entry for this to be a real business for Google, at least one that becomes nation-wide or global.
  - melling 2y ago
    How about making a prediction and we will follow up in 5 years. To me it sounds like you’re just being sort of vague. Yes, absolutely everyone knew there were social issues too. Where will self driving be in 5 or 10 years?
    - dmix 2y ago
      Predicting it on a macro level is a boring question IMO, the influence of policy interference and social outrage every time there's an accident - which is never fully connected to the realities on the ground of true capabilities/potential - is basically a black box. Like betting on elections. My interest atm is mostly the economics question of whether LIDAR will scale widely in near term combined with the social risks (ie, millions of cars with $30-50k in sensors, which atm cost as much as the car itself, so basically only e-taxis, no personal ownership + high cost of maintenance) vs cameras being good-enough... that's a far more interesting question. It's pretty obvious LIDAR/ultrasonic or a combination with cameras is superior. So maybe some ambitious person will find some mass-producible middle ground within <5yrs shrug
  - minwcnt5 2y ago
    Calling San Francisco a "town" diminishes what they've achieved. It is the second densest urban center and ride hailing market in the US, and a difficult place for humans to drive. I think being able to effectively replace Uber drivers there means that once they crack freeway driving and snow, the regulatory, economic, policy, and social hurdles will start to fall like dominos. I don't think this is how most people expected this to go. My personal belief was that self-driving cars would be limited to medium-density suburban areas initially (i.e. places where the driving is relatively simple, but there's still enough demand for it to be economically viable). It's astonishing that they were able to move from suburban trials to high density environments so quickly.
    - dmix 2y ago
      Everyone obsesses on the technical side which was necessary but scaling it up country wide as a functioning profitable business entity, even just urban/suburban, requires far more than just heavily subsidized R&D work and experiments as a minor taxi player in a few places. I have doubts the very slow pace of expansion was just a matter of regulations and nailing down the tech. It has to make sense as a business otherwise they aren’t going to dump billions into sensor factories (where each LIDAR suite currently costs around $75k to make), cars, and service infrastructure. Just like how Tesla wasn’t just a of matter of making the first prototypes of electric sports cars. The real magic and barrier to entry was the factories and infrastructure to make the economics work.

ryandvm 2y ago

> Elon said a camera is all you need. This was when I really started to doubt that guy's genius. That was a phenomenally stupid assertion. I can drive with 2 eyes because I have a bunch of other sensors as well as an entire world model in my head with 30 years of driving experience. Is it Halloween tonight? Better drive extra slowly. Did that white semi with the bright background sky just disappear? Probably not. Is that deer going to bolt in front of the car coming the other way? Maybe. Is the road slick because it recently rained here? etc.

minwcnt5 2y ago
"640k ought to be enough for anybody" vibes.