Reply to post: I think your on the right track

'Autopilot' Tesla crashed into our parked patrol car, say SoCal cops

Corporate Scum

I think your on the right track

There are only a couple of sections of the outbound leg that have those side of the road markings for parking, and most of the road HAS a center divider, unlike the photo, so that narrow the location to just a couple spots. (You may have seen a more specific location, I'm working off the photo and being fairly local.) It looks like the are by the dog park but, another section does something similar by Broadway.

This town is great example of near total disregard for normal road markings. The road lines don't use the normal width, the dashed yellow lines for the center don't use the normal spacing, most of the lane lines are solid white instead of dashed and there are huge arrows everywhere. I'm not sure, but I think that the parking spaces where the police car was parked are limited to off peak hours, and as the Twitter thread points out that section is Wackadoodle from a road planning standpoint. It basically routes the right hand lane into parked cars.

So if we find out this is a blind spot for the Tesla's neural net, no one should be surprised. It does not play to the strengths of those systems to if they are dealing with a bunch of needlessly unique visual queues where there is only one place or set of training data.

Speaking of which, it would be a great idea if the NTSB was building up a shared pile of data from all of the crashes of cars equipped with autonomy, either full self drive or safety features. Every manufacturer should be using these incidents in their training data and testing. That data should include cases where the driver was in control as well, as the car should be learning to help us humans avoid the mistakes we are prone to.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon


Biting the hand that feeds IT © 1998–2021