3 ms·

I would like to see the more detailed telemetry that the car gathered at the time. For a human driving the car, the dash cam together with personal testimony w

I would like to see the more detailed telemetry that the car gathered at the time. For a human driving the car, the dash cam together with personal testimony would probably put the issue to rest; although the fact that the pedestrian was in the roadway before they entered the lane looks very suspicious because a human likely would have seen them because of headlights on an open road even if not illuminated by streetlights. Nonetheless, the dash cam shows what it shows, and human sense and memory is fragile. Not so for a self-driving car -- the memory is not fragile, and should be preserved. If it is not preserved and there is no remaining copy of the car's telemetry at the time, that in itself is a detail that can point towards gross negligence on the part of Uber. If it is preserved, the question becomes, would a "reasonable person/software" have acted based on what was shown there, and that goes towards whether there is negligence or recklessness. But we (or at least, investigators) need to see that telemetry to make that decision.

hcknwscommenter 9y ago
>the dash cam shows what it shows The dash cam is either proof that: i) that dash cams are far inferior to human eyes in terms of dynamic range (certainly true); and/or ii) the video was purposely darkened or compressed to mislead the public into believing a human driver would have hit the pedestrian (see, https://arstechnica.com/cars/2018/03/police-chief-said-uber-victim-came-from-the-shadows-dont-believe-it/ https://arstechnica.com/cars/2018/03/police-chief-said-uber-... ). I think it was both i) and ii).
- cr0sh 9y ago
  I'd like to see the actual data it gathered as well. Plus, I'd like to see the software system, ML models, etc - to run the data against. The dash cam video is very suspicious. What I want to see is the actual computer vision camera footage; it's never been made clear whether the footage that has been publically made available is from a dashcam or CV camera; I suspect the former. On the notion of the LIDAR - the device spits out 64 beams; it appeared in the video the woman who was hit was wearing a fuzzy and dark color jacket (was she? I don't know for certain). If this is so - it is possible that any reflections for the LIDAR returns only came from the bottom half and the bicycle at best. The top half would have been either absorbed, or possibly filtered as "noise" prior to the classification algorithm. Which is why the ML algo's matter - how did the classifier classify the data? Did it think what was moving across the road was something other than a human? Without the data and the algorithm, there's no good way to know. But now I hear it's possible the LIDAR was turned off to run a "vision only" driving system. Was it? Who knows; we may never get the truth of what happened here. Form a personal standpoint, it saddens and perplexes me - and makes me upset to a degree - that this happened here in my city. I recently completed the Udacity Self-Driving Car Engineer nanodegree, and I tend to wonder if this incident will somehow potentially color any future employment opportunities in that field, should I pursue them... The whole political angle of this thing - sigh.
  - andrewla 9y ago
    The software system and ML models would only be necessary if an analysis of the underlying data showed that there should be a reasonable chance of avoiding the accident. If that were the case then the models and training process could well be distinction between simple recklessness and gross negligence. I doubt the LIDAR data would be that useful compared to the enhanced high fidelity video just because of how interpretable it would be to a jury.
- andrewla 9y ago
  The first point is, as you say, certainly true. The second point is an allegation of criminal obstruction of justice, which would be taken into account both as a crime in an of itself and also as an overt act in furtherance of a conspiracy. These are very serious charges, that would result in prison time for people throughout the program. Thus it's my tendency to assume good faith on the part of Uber since the consequences are so dire. Most likely the higher fidelity video and sensor logs are being held back pending a warrant or subpoena for them so that Uber can conduct an internal investigation to figure out how avoidable the collision was, and to avoid giving ammunition to a wrongful death civil suit. Until the footage is released to investigators or potentially to the public, I think both sides (unfortunate accident vs. gross negligence or recklessness) are possibilities. I personally think the latter is likely -- purely a gut feeling based on the video released. If the underlying footage was not automatically archived or otherwise somehow lost then I think that has to be significant evidence regarding "intent" in a criminal case for manslaughter.

hcknwscommenter 9y ago

>the dash cam shows what it shows The dash cam is either proof that: i) that dash cams are far inferior to human eyes in terms of dynamic range (certainly true); and/or ii) the video was purposely darkened or compressed to mislead the public into believing a human driver would have hit the pedestrian (see, https://arstechnica.com/cars/2018/03/police-chief-said-uber-victim-came-from-the-shadows-dont-believe-it/ https://arstechnica.com/cars/2018/03/police-chief-said-uber-... ). I think it was both i) and ii).

cr0sh 9y ago
I'd like to see the actual data it gathered as well. Plus, I'd like to see the software system, ML models, etc - to run the data against. The dash cam video is very suspicious. What I want to see is the actual computer vision camera footage; it's never been made clear whether the footage that has been publically made available is from a dashcam or CV camera; I suspect the former. On the notion of the LIDAR - the device spits out 64 beams; it appeared in the video the woman who was hit was wearing a fuzzy and dark color jacket (was she? I don't know for certain). If this is so - it is possible that any reflections for the LIDAR returns only came from the bottom half and the bicycle at best. The top half would have been either absorbed, or possibly filtered as "noise" prior to the classification algorithm. Which is why the ML algo's matter - how did the classifier classify the data? Did it think what was moving across the road was something other than a human? Without the data and the algorithm, there's no good way to know. But now I hear it's possible the LIDAR was turned off to run a "vision only" driving system. Was it? Who knows; we may never get the truth of what happened here. Form a personal standpoint, it saddens and perplexes me - and makes me upset to a degree - that this happened here in my city. I recently completed the Udacity Self-Driving Car Engineer nanodegree, and I tend to wonder if this incident will somehow potentially color any future employment opportunities in that field, should I pursue them... The whole political angle of this thing - sigh.
- andrewla 9y ago
  The software system and ML models would only be necessary if an analysis of the underlying data showed that there should be a reasonable chance of avoiding the accident. If that were the case then the models and training process could well be distinction between simple recklessness and gross negligence. I doubt the LIDAR data would be that useful compared to the enhanced high fidelity video just because of how interpretable it would be to a jury.
andrewla 9y ago
The first point is, as you say, certainly true. The second point is an allegation of criminal obstruction of justice, which would be taken into account both as a crime in an of itself and also as an overt act in furtherance of a conspiracy. These are very serious charges, that would result in prison time for people throughout the program. Thus it's my tendency to assume good faith on the part of Uber since the consequences are so dire. Most likely the higher fidelity video and sensor logs are being held back pending a warrant or subpoena for them so that Uber can conduct an internal investigation to figure out how avoidable the collision was, and to avoid giving ammunition to a wrongful death civil suit. Until the footage is released to investigators or potentially to the public, I think both sides (unfortunate accident vs. gross negligence or recklessness) are possibilities. I personally think the latter is likely -- purely a gut feeling based on the video released. If the underlying footage was not automatically archived or otherwise somehow lost then I think that has to be significant evidence regarding "intent" in a criminal case for manslaughter.