3 ms·

Fantastic. The reality is that good mapping and some level of lidar (which is getting cheaper faster for low res/angle sensors) is somewhat game changing in ter

Fantastic. The reality is that good mapping and some level of lidar (which is getting cheaper faster for low res/angle sensors) is somewhat game changing in terms of self driving vehicle reliability. Tesla will miss out I think by not taking advantage of doing even the cheapest possible one sensor lidar to augment their existing system. $500 -> in the future $100 for the data lidar gives you is gold.

tobyjsullivan 6y ago
I imagine you're correct that lidar is a huge leg-up in the race to self-driving. However, I can see a scenario where, despite that, it is not the right choice for Tesla. It's important to remember that self-driving technology will be commodity at some point (like the lidar units themselves). Whoever offers the most affordable option meeting some baseline performance will dominate the market and licence to every self-driving fleet. It wouldn't surprise me if Waymo ended up meeting that criteria first and licensing the tech to everyone from Uber to Toyota. Once that tech is out, anyone else that succeeds in developing comparable tech effectively has an uphill battle ahead of them trying to sell their new, less proven version at a price and volume that can recoup their R&D investment. I think we'll see a lot of players leave the race once the first licensable product hits the market. In fact, I think we've seen a bit of that already. Which brings me back to Tesla. My guess is their strategy is not just to develop self-driving - they'd lose on equal ground (due to not being first) - but to develop "2nd gen" self-driving which has a specific competitive advantage of not requiring lidar and thus having lower total cost per unit. This allows them to enter the market late but still have a significant edge in that market. If a company lets their engineer's use lidar to develop self-driving tech, then the tech they build will be 100% dependent on lidar. That's Parkinson's law. If, on the other hand, the company adds a strict constraint that lidar is not an option, then the engineers might fail to build something that works - or else they will produce something with a unique advantage.
- donor20 6y ago
  This is not true. Google is collecting incredible levels of ground / lidar truthed imagery. In terms of then doing a generation function from imagery to 3D model (without lidar) they are going to be in a good spot if they want to be. This is the same approach they have taken in many many other areas - huge ground truthed datasets, then infer without the ground truth.
- echelon 6y ago
  > Which brings me back to Tesla. My guess is their strategy is not just to develop self-driving - they'd lose on equal ground (due to not being first) - but to develop "2nd gen" self-driving which has a specific competitive advantage of not requiring lidar and thus having lower total cost per unit. This allows them to enter the market late but still have a significant edge in that market. They'll be able to drive in the rain and snow.
  - cloudwalking 6y ago
    They won't, though, without HD maps. That's the irony. Nobody working on (real) SDCs is worried about rain or snow. Yes it's a problem right now, but a clearly solvable one. It hasn't been solved yet because it isn't a high priority.
- wstrange 6y ago
  Your analysis assumes Lidar is going to remain expensive. That is likely not the case...
- jimmydorry 6y ago
  I can't imagine waymo or any of the other groups that are basing their technology around lidar, managing to get to "commodity stage". LIDAR has a fixed range it is useful for, which is greatly shrunk when you also need to factor in vehicle response time. Cameras can see as far as humans can (further actually, as it can see things more reliably at greater distances than humans). Going beyond 25mph is already massively encroaching on the upper ranges of safety with LIDAR systems.
  - cloudwalking 6y ago
    This is simply not true, as evidenced by Waymo driving on roads faster than 25 mph
    - jimmydorry 6y ago
      Waymo doing select trial, or perhaps even limited runs above 25mph does not prove they can operate safely at such speeds. My math might be off since I don't use the imperial system, but LIDAR is only good up to 50-100m (closer to 50 than 100). Slamming the brakes on your vehicle to stop within the sensor distance is more dangerous (to the occupants and cars behidn you) and less effective (less likely to stop in time) the faster you're going. Can you go significantly faster than you can safely see and stop in time? Sure. Would it still be considered safe when doing so? Hell no.
- ironman1478 6y ago
  I do not think self driving technology will become a commodity be for a very very long time. It takes an enormous amount of processing power, both offline and online, and some incredibly specialized hardware to get right. I don't think the companies that actually succeed will actually sell their software. Waymo and probably are cruise are light years ahead of the competition and would want to take advantage of the monopoly they would have for as long as the can.
nbardy 6y ago
In that framing Tesla is already ahead in data collection. Pure numbers of cars dwarfs the advantage of lidar
- donor20 6y ago
  Tesla does not have the 3D data to go with their imagery - google does. The amount of data google can collect, then integrate into their ground truthed 3D maps (which street view collects) is incredible. If they cared to they could back generate from images only to 3D based on an absolutely amazing training data set.

tobyjsullivan 6y ago

I imagine you're correct that lidar is a huge leg-up in the race to self-driving. However, I can see a scenario where, despite that, it is not the right choice for Tesla. It's important to remember that self-driving technology will be commodity at some point (like the lidar units themselves). Whoever offers the most affordable option meeting some baseline performance will dominate the market and licence to every self-driving fleet. It wouldn't surprise me if Waymo ended up meeting that criteria first and licensing the tech to everyone from Uber to Toyota. Once that tech is out, anyone else that succeeds in developing comparable tech effectively has an uphill battle ahead of them trying to sell their new, less proven version at a price and volume that can recoup their R&D investment. I think we'll see a lot of players leave the race once the first licensable product hits the market. In fact, I think we've seen a bit of that already. Which brings me back to Tesla. My guess is their strategy is not just to develop self-driving - they'd lose on equal ground (due to not being first) - but to develop "2nd gen" self-driving which has a specific competitive advantage of not requiring lidar and thus having lower total cost per unit. This allows them to enter the market late but still have a significant edge in that market. If a company lets their engineer's use lidar to develop self-driving tech, then the tech they build will be 100% dependent on lidar. That's Parkinson's law. If, on the other hand, the company adds a strict constraint that lidar is not an option, then the engineers might fail to build something that works - or else they will produce something with a unique advantage.

donor20 6y ago
This is not true. Google is collecting incredible levels of ground / lidar truthed imagery. In terms of then doing a generation function from imagery to 3D model (without lidar) they are going to be in a good spot if they want to be. This is the same approach they have taken in many many other areas - huge ground truthed datasets, then infer without the ground truth.
echelon 6y ago
> Which brings me back to Tesla. My guess is their strategy is not just to develop self-driving - they'd lose on equal ground (due to not being first) - but to develop "2nd gen" self-driving which has a specific competitive advantage of not requiring lidar and thus having lower total cost per unit. This allows them to enter the market late but still have a significant edge in that market. They'll be able to drive in the rain and snow.
- cloudwalking 6y ago
  They won't, though, without HD maps. That's the irony. Nobody working on (real) SDCs is worried about rain or snow. Yes it's a problem right now, but a clearly solvable one. It hasn't been solved yet because it isn't a high priority.
wstrange 6y ago
Your analysis assumes Lidar is going to remain expensive. That is likely not the case...
jimmydorry 6y ago
I can't imagine waymo or any of the other groups that are basing their technology around lidar, managing to get to "commodity stage". LIDAR has a fixed range it is useful for, which is greatly shrunk when you also need to factor in vehicle response time. Cameras can see as far as humans can (further actually, as it can see things more reliably at greater distances than humans). Going beyond 25mph is already massively encroaching on the upper ranges of safety with LIDAR systems.
- cloudwalking 6y ago
  This is simply not true, as evidenced by Waymo driving on roads faster than 25 mph
  - jimmydorry 6y ago
    Waymo doing select trial, or perhaps even limited runs above 25mph does not prove they can operate safely at such speeds. My math might be off since I don't use the imperial system, but LIDAR is only good up to 50-100m (closer to 50 than 100). Slamming the brakes on your vehicle to stop within the sensor distance is more dangerous (to the occupants and cars behidn you) and less effective (less likely to stop in time) the faster you're going. Can you go significantly faster than you can safely see and stop in time? Sure. Would it still be considered safe when doing so? Hell no.
ironman1478 6y ago
I do not think self driving technology will become a commodity be for a very very long time. It takes an enormous amount of processing power, both offline and online, and some incredibly specialized hardware to get right. I don't think the companies that actually succeed will actually sell their software. Waymo and probably are cruise are light years ahead of the competition and would want to take advantage of the monopoly they would have for as long as the can.

nbardy 6y ago

In that framing Tesla is already ahead in data collection. Pure numbers of cars dwarfs the advantage of lidar

donor20 6y ago
Tesla does not have the 3D data to go with their imagery - google does. The amount of data google can collect, then integrate into their ground truthed 3D maps (which street view collects) is incredible. If they cared to they could back generate from images only to 3D based on an absolutely amazing training data set.