3 ms·

How does Tesla fail so badly at basic collision avoidance?? Surely the LIDAR is able to detect an imminent collision, but at least in the recent fire truck and

by lph 8y ago

How does Tesla fail so badly at basic collision avoidance?? Surely the LIDAR is able to detect an imminent collision, but at least in the recent fire truck and lane-split barrier incidents, the cars didn't even slow down?! That seems really fundamental. Edit: so they don't use lidar... Is this a machine vision failure, then?

stevenwoo 8y ago
People have stated in older threads that the lane assist will not brake for stationary objects - if it did, it would brake all the time due to all the false positives. The lane following is pretty impeccable right until the collision for the Chinese accident. https://www.youtube.com/watch?v=fc0yYJ8-Dyo https://www.youtube.com/watch?v=fc0yYJ8-Dyo
Cogito 8y ago
My understanding is that they detect moving obstacles much more easily than stationary. This is because almost everything in the environment is stationary to the car, so it is hard to distinguish between stationary things in your way and stationary things just off to the side. So if the car in front of you slows down, it's obvious that you'll hit it if you don't slow down. A car parked partially in your lane might just look like regular noise to the sensors. That is, using the current technology, if they (correctly) classified the parked car as an obstacle then they would (incorrectly) classify so many non-dangerous situations as obstacles so as to make the system infeasible; the car would emergency brake constantly as it drives down the road. Some would say that until those obstacles can be classified correctly such technology should not be used at all, but I think it's probably doing more good than harm at the moment and hopefully a good solution can be found quickly. Even if they do fix this problem soon, I'm sure we'll hear plenty of stories where something goes wrong anyway, but hopefully they won't seem so obviously avoidable (who hits a parked car!)
- s_m_t 8y ago
  How dumb of an idea is it to attach the camera/lidar/what have you to a 2 axis rail and move it around at a high speed to introduce some parallax movement to the scene?
  - Cogito 8y ago
    I assume you could achieve the same thing with less complexity and more cheaply using a fixed system. Either having more than one sensor or using mirrors or something. In fact, I'm sure that the camera systems in place are already doing this to some degree. I think it's more that it's not yet good enough at picking out these stationary objects that are in the path of travel.

by lph 8y ago

stevenwoo 8y ago

People have stated in older threads that the lane assist will not brake for stationary objects - if it did, it would brake all the time due to all the false positives. The lane following is pretty impeccable right until the collision for the Chinese accident. https://www.youtube.com/watch?v=fc0yYJ8-Dyo https://www.youtube.com/watch?v=fc0yYJ8-Dyo

Cogito 8y ago

My understanding is that they detect moving obstacles much more easily than stationary. This is because almost everything in the environment is stationary to the car, so it is hard to distinguish between stationary things in your way and stationary things just off to the side. So if the car in front of you slows down, it's obvious that you'll hit it if you don't slow down. A car parked partially in your lane might just look like regular noise to the sensors. That is, using the current technology, if they (correctly) classified the parked car as an obstacle then they would (incorrectly) classify so many non-dangerous situations as obstacles so as to make the system infeasible; the car would emergency brake constantly as it drives down the road. Some would say that until those obstacles can be classified correctly such technology should not be used at all, but I think it's probably doing more good than harm at the moment and hopefully a good solution can be found quickly. Even if they do fix this problem soon, I'm sure we'll hear plenty of stories where something goes wrong anyway, but hopefully they won't seem so obviously avoidable (who hits a parked car!)

s_m_t 8y ago
How dumb of an idea is it to attach the camera/lidar/what have you to a 2 axis rail and move it around at a high speed to introduce some parallax movement to the scene?
- Cogito 8y ago
  I assume you could achieve the same thing with less complexity and more cheaply using a fixed system. Either having more than one sensor or using mirrors or something. In fact, I'm sure that the camera systems in place are already doing this to some degree. I think it's more that it's not yet good enough at picking out these stationary objects that are in the path of travel.