Driverless cars just got a whole lot harder. This technical paper by @amirrosenfeld raises some profound questions about the robustness of #DeepLearning as a perceptual mechanism. https://arxiv.org/abs/1808.03305
-
And imho, "Driverless cars just got a whole lot harder" is being dramatic. We've known DL has no robustness guarantees since... forever. I believe
@Waymo's tech is based on much more reliable concepts. Idea of end to end self driving is pretty much silly, always has been.
Much harder because our intuitions about what DL systems are doing and the range of circumstances in which they work are undermined rather dramatically. Would certainly be curious how Waymo gets around these issues...
-
-
Waymo's tech goes way back to before DL hype - Google got
@SebastianThrun to lead efforts after the darpa competition in 2005. I believe they use a lot of standard tools - SLAM, motion planning, control theory, etc. Deep Learning is really only needed for recognition, auxilliary. -
I presume that to be correct, but
@AmirRosenfeld’s paper leads to a new class of worries about robustness on the recognition side; not obvious what to do about it. - 2 more replies
New conversation -
-
-
...*if* it gets around these issues
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.