4 ms·

Seems like these components were authorized at the height of the self driving hype cycle as a way of future proofing the cars--that is, the owner could download

Seems like these components were authorized at the height of the self driving hype cycle as a way of future proofing the cars--that is, the owner could download a self-driving software update and the hardware would already be ready for it. In retrospect, this looks foolish: doing self driving without a LIDAR is borderline unethical. I wouldn't be surprised if newer models do away with these ASICs until the software is further along.

ec109685 7y ago
Huh? Even without full level 5 self-driving, these chips allow Tesla to control steering and breaking in many situations. Humans don’t have lidar and we are able to navigate the world. If the windows of the car was replaced by a low latency screen, showing exactly what we’d see if we looked out the window, we could drive just fine, so cameras do provide enough data to self drive. You just need a smart processor.
- raxxorrax 7y ago
  No, you would lack depth perception. It is possible to drive, but it would probably be significantly more difficult. I have a friend with a dysfunctional eye. He has a drivers license, but really doesn't like driving or catching things out of the air. But while humans can successfully correct for this issue through extended interpretation of perceived images and training, it becomes almost a necessity for a stupid computer. I believe a fast and efficient stereoscopy would make identifying objects much easier. But it isn't neither trivial to implement, nor very performant to my knowledge. A Lidar would be an alternative of course.
- Invictus0 7y ago
  The bar is not "being able to navigate the world". The bar is doing so 10x more safely than a human, and currently the human deaths-in-car-crash rate is something like 1.2 per 100 million miles driven [0]. For scale, that is the same distance from the Earth to the sun, and equivalent to 4000 trips around the equator. I understand it is theoretically possible to beat this without a LIDAR but you are seriously understating the difficulty of this task when you say you just need a smart processor. [0]: https://www.iihs.org/topics/fatality-statistics/detail/state-by-state https://www.iihs.org/topics/fatality-statistics/detail/state...
  - ec109685 7y ago
    Injury rate / accident rate is a good proxy for deaths, so you don’t need as much data as 1.2 in 100M would imply. Also, given humans drive a lot, even with those astronomical odds, if you drove 15k a year for 60 years, you’d have an almost 1% chance of dying in an auto accident.

ec109685 7y ago

Huh? Even without full level 5 self-driving, these chips allow Tesla to control steering and breaking in many situations. Humans don’t have lidar and we are able to navigate the world. If the windows of the car was replaced by a low latency screen, showing exactly what we’d see if we looked out the window, we could drive just fine, so cameras do provide enough data to self drive. You just need a smart processor.

raxxorrax 7y ago
No, you would lack depth perception. It is possible to drive, but it would probably be significantly more difficult. I have a friend with a dysfunctional eye. He has a drivers license, but really doesn't like driving or catching things out of the air. But while humans can successfully correct for this issue through extended interpretation of perceived images and training, it becomes almost a necessity for a stupid computer. I believe a fast and efficient stereoscopy would make identifying objects much easier. But it isn't neither trivial to implement, nor very performant to my knowledge. A Lidar would be an alternative of course.
Invictus0 7y ago
The bar is not "being able to navigate the world". The bar is doing so 10x more safely than a human, and currently the human deaths-in-car-crash rate is something like 1.2 per 100 million miles driven [0]. For scale, that is the same distance from the Earth to the sun, and equivalent to 4000 trips around the equator. I understand it is theoretically possible to beat this without a LIDAR but you are seriously understating the difficulty of this task when you say you just need a smart processor. [0]: https://www.iihs.org/topics/fatality-statistics/detail/state-by-state https://www.iihs.org/topics/fatality-statistics/detail/state...
- ec109685 7y ago
  Injury rate / accident rate is a good proxy for deaths, so you don’t need as much data as 1.2 in 100M would imply. Also, given humans drive a lot, even with those astronomical odds, if you drove 15k a year for 60 years, you’d have an almost 1% chance of dying in an auto accident.