NTSB: Autopilot steered Tesla car toward traffic barrier before deadly crash

Post by lsaferite » Fri, Jun 08, 2018, 3:27 AM UTC

crote wrote: post #148

panopticon wrote: post #127

Earlier quoted context omitted.

Out of curiosity, are you not in the U.S.? These types of lane dividers are incredibly common throughout the country. Note that I'm not defending the safety of them; I'm just surprised to see someone call them out as a hazard as they're such a common sight.

Dutch citizen here, I have never seen something like this. To compare, here's a similar situation in The Netherlands: https://www.google.nl/maps/@52.0928662,5.1734685,3a,75y,281.... Points of interest: - big solid white arrow, line-fade is nearly impossible - white-green sign indicating that the road is splitting - big shoulder in the continuation of the division - loads of grass instead of immediately using a metal…

Post by kartik-bh » Fri, Jun 08, 2018, 3:32 AM UTC

fixermark wrote: post #192

MBCook wrote: post #48

Earlier quoted context omitted.

That’s the thing, it clearly didn’t detect the obstacle. Given that there were no (known) obstacles, it behaved correctly. Fix the obstacle detection.

The problem is it's entirely possible the obstacle detection is un-fixable without swapping out the sensor suite the Tesla ships with. Tesla is quite clearly expecting the problem to be solvable with that sensor suite alone (this is more-or-less the line I was given by a Tesla seller when I was shopping for a car), so it's not going to be in the T-sheet for them to recall every Tesla on the road to fit a LIDAR to the…

Post by dominotw » Fri, Jun 08, 2018, 3:38 AM UTC

salawat wrote: post #358

Pfhreak wrote: post #299

Earlier quoted context omitted.

This seems testable. IANAAE (I Am Not An Automotive Engineer), but why can't you run both the new and old code side by side and if the actions they take are materially different investigate further? Like, if in one case the new code would want you to move left, and the old code goes straight, one of those behaviors is probably wrong. If the driver corrects, then the new code is probably correct, but if the driver doe…

It has already been touched on by another commenter, but testability of Machine Learned systems outside of a training dataset is pretty much a crapshoot. An ML solution stops "learning" after training and only reacts. To illustrate the difference, have you driven on roads under construction lately? As humans, when you've driven the same road hundreds of times, you start to do the same thing as a machine learned imple…

Post by stephengillie » Fri, Jun 08, 2018, 3:48 AM UTC

ingenieroariel wrote: post #24

devy wrote: post #7

Earlier quoted context omitted.

Add to that, the last bullet point from page 2 of the official preliminary report[1]: At 3 seconds prior to the crash and up to the time of impact with the crash attenuator, the Tesla’s speed increased from 62 to 70.8 mph, with no precrash braking or evasive steering movement detected. That to me is a strong indicator that Tesla's AP2 can not recognize a crashed attenuator, probably one of the strongest argument that…

This happens with regular adaptive cruise control on my Ford, when the cruise control set point is a lot higher than the current speed: the car accelerates after another car clears the lane and ram into things.

Post by justwalt » Fri, Jun 08, 2018, 4:05 AM UTC

menacingly wrote: post #47

Tesla has to realize these "shame the dead dude" posts are PR nightmares, right? They are reason alone for me to never consider one, that a private moment for my family might end up a pawn in some "convince the public we're safe using any weasel stretch of the facts we can" effort. If this is disruption, I'll wait for the old guard to catch up, lest I be disrupted into a concrete barrier and my grieving widow fed mis…