1
$\begingroup$

If a street is extremely crowded to an extent that the terrain is not visible from the point of view of the LIDAR (e.g. in google's self driving car), can it still manage to localize itself and continue to operate? I recall Sebastian Thrun saying that Google's car cannot navigate through snow filled roads since the onboard LIDAR cannot map the terrain beneath the snow (e.g. here).

[Edit : Based on the comments] Clarifying the context, here "not visible" means there is an obstacle between the LIDAR and the terrain

Improve this question
$\endgroup$
4
  • 1
    $\begingroup$ So it couldn't drive in the night, too? $\endgroup$
    – ott--
    Commented Jul 1, 2013 at 20:13
  • $\begingroup$ LIDAR should operate at night w/o difficulty. $\endgroup$
    – erichui
    Commented Jul 2, 2013 at 1:36
  • 1
    $\begingroup$ What does it mean to "continue to operate"? I would hope that the proper behavior of a self driving car is not to continue moving forward if there are people in front of it. $\endgroup$
    – Ian
    Commented Jul 2, 2013 at 14:04
  • $\begingroup$ @Ian By that i mean, can it still know where it is ? or will it freeze. My question is focused on localization and of course it will move only when it is safe to do so. $\endgroup$
    – atree
    Commented Jul 2, 2013 at 14:37

1 Answer 1

Reset to default
0
$\begingroup$

The snow problem you're referring to seems to that the car can no longer tell where the edge of the road is.

Even in heavy traffic, a good driver still maintains enough space from the car in front that plenty of road is visible, and the side of the road identifiable from LiDAR. This is not common practice, but I imagine autonomous cars would be programmed to do so. In that case, plenty of road is always visible for the car.

Google's car seems to have no problems in heavy traffic.

Improve this answer
$\endgroup$
3
  • $\begingroup$ What if the car is closely surrounded by huge trucks. There can be numerous such extreme scenarios. My question is, how can it handle the situation when LIDAR is compromised? $\endgroup$
    – atree
    Commented Jul 3, 2013 at 17:15
  • $\begingroup$ Well, options I can think of: 1. Inertial dead reckoning 2. (differential) GPS. And presumably all position estimates are combined continuously in a Kalman filter or similar. $\endgroup$
    – foobarbecue
    Commented Jul 3, 2013 at 18:02
  • $\begingroup$ Anyway... if it's surrounded that closely, the goal ceases to be "determine location relative to road" and becomes "don't hit any of these trucks." I mean, you could ask the same question about a human. "How can a human drive if there are so many trucks she can't see the road?" $\endgroup$
    – foobarbecue
    Commented Jul 4, 2013 at 16:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.