5 ms·

Because LIDAR is expensive. If camera-based neural networks eventually get good enough that LIDAR provides minimal additional value, they will drop it. This is

Because LIDAR is expensive. If camera-based neural networks eventually get good enough that LIDAR provides minimal additional value, they will drop it. This is what Musk is betting on. We're not there yet for sure, and it's not clear to me yet whether that's a realistic goal for the 5-10 year time frame.

hef19898 7y ago
Musk is betting on a lot of things lately. I would rather have him focus on one or two hard problems and solve these first.
- vinceguidry 7y ago
  I don't think his original two bets have been fully solved yet, electric cars and private space exploration. He seems as dedicated to those two goals as he ever was.
- whatshisface 7y ago
  Musk's job isn't to solve problems, it's to find people who can solve problems, put them together, and pay them.
  - coldtea 7y ago
    Musk's job is to secure investment (and government deals) with promises and signs of hope.
navigatesol 7y ago
>Because LIDAR is expensive So is the battery technology, and electric vehicles in general. Is there anything unique about LIDAR to assume the costs won't decrease significantly over time?
- CrazyStat 7y ago
  Doesn't matter. Even if the cost of LIDAR decreases drastically to, say, $100/car, companies will still optimize it out if they can.
snowwrestler 7y ago
There is some precedent for this. Precise digital surface models are useful in geomorphological studies, and the state of the art for producing a new one in the field is laser scanning. But earlier this decade, researchers started using differential processing of sequential images taken from a platform moving arbitrarily (plane, helicopter, drone, etc.) and were able to produce DSMs with elevation resolutions below 10cm, which is sufficient for a lot of studies. See this paper on "structure from motion": https://www.sciencedirect.com/science/article/pii/S0169555X12004217 https://www.sciencedirect.com/science/article/pii/S0169555X1... The big difference between this and driving is that in SfM, the platform moves arbitrarily around the feature of interest until it has enough data to build the model. Driving is far more linear; your car can't circle an unknown object 5 times until it understands what it is. So, it might be more important for driving applications to maximize the number of different channels of sensor data available to integrate for creating a model.

hef19898 7y ago

Musk is betting on a lot of things lately. I would rather have him focus on one or two hard problems and solve these first.

vinceguidry 7y ago
I don't think his original two bets have been fully solved yet, electric cars and private space exploration. He seems as dedicated to those two goals as he ever was.
whatshisface 7y ago
Musk's job isn't to solve problems, it's to find people who can solve problems, put them together, and pay them.
- coldtea 7y ago
  Musk's job is to secure investment (and government deals) with promises and signs of hope.

navigatesol 7y ago

>Because LIDAR is expensive So is the battery technology, and electric vehicles in general. Is there anything unique about LIDAR to assume the costs won't decrease significantly over time?

CrazyStat 7y ago
Doesn't matter. Even if the cost of LIDAR decreases drastically to, say, $100/car, companies will still optimize it out if they can.

snowwrestler 7y ago

There is some precedent for this. Precise digital surface models are useful in geomorphological studies, and the state of the art for producing a new one in the field is laser scanning. But earlier this decade, researchers started using differential processing of sequential images taken from a platform moving arbitrarily (plane, helicopter, drone, etc.) and were able to produce DSMs with elevation resolutions below 10cm, which is sufficient for a lot of studies. See this paper on "structure from motion": https://www.sciencedirect.com/science/article/pii/S0169555X12004217 https://www.sciencedirect.com/science/article/pii/S0169555X1... The big difference between this and driving is that in SfM, the platform moves arbitrarily around the feature of interest until it has enough data to build the model. Driving is far more linear; your car can't circle an unknown object 5 times until it understands what it is. So, it might be more important for driving applications to maximize the number of different channels of sensor data available to integrate for creating a model.