Arizona governor and Uber kept self-driving program secret, emails reveal(theguardian.com)
theguardian.com
Arizona governor and Uber kept self-driving program secret, emails reveal
https://www.theguardian.com/technology/2018/mar/28/uber-arizona-secret-self-driving-program-governor-doug-ducey
78 コメント
One of the main points of the federation model was that it should be able to scale. Local regions rule themselves in nearly all things (the governor is the president of Alabama) and the regions people are only required to comply with a small number of issues that effect everyone (like national defense).
Of course we all know how that ended up. The federal powers increased steadily until we got where we are. This is one of the reasons you see so much anger against the coasts and cities.
Of course we all know how that ended up. The federal powers increased steadily until we got where we are. This is one of the reasons you see so much anger against the coasts and cities.
With some exceptions, Federal powers increased because we couldn't agree that human beings are not property, among other basic rights issues. The touchstone moments were not simple disagreements on policy, it was the pot boiling over into extreme violence. Even the Federal model requires some basic level setting for it to function. There's either a rebalancing or the whole thing falls apart.
Regardless, GP's points are notable just a local level. States have different weights assigned to executive and legislative. It appears that Arizona leans heavily towards a strong executive branch although I have to wonder if any laws were broken here. People expect corruption in US politics at this point but this incident has some staying power in the news.
Regardless, GP's points are notable just a local level. States have different weights assigned to executive and legislative. It appears that Arizona leans heavily towards a strong executive branch although I have to wonder if any laws were broken here. People expect corruption in US politics at this point but this incident has some staying power in the news.
More so federal funding has bent states into the shape that the federal government wants them to be. Funding with strings and the interstate commerce clause subdued the States’ power realistically.
I didn't mean that the federal agencies should necessarily have a role here. I would think that many US states also have agencies charged with administrating the law, something like a "Arizona Roadway Authority" or whatever.
My comment was trying to explain how even within a single hierarchy like a state's executive, which a Governor at the top and traffic engineers somewhere below, it's peculiar to see someone jump so many levels of that hierarchy to assume control of a rather low-level decision.
My comment was trying to explain how even within a single hierarchy like a state's executive, which a Governor at the top and traffic engineers somewhere below, it's peculiar to see someone jump so many levels of that hierarchy to assume control of a rather low-level decision.
I got trashed at a party with a leader of Uber's self driving program. He told me it was normal to lie about who was in control when incidents happen. He said everybody does it. He said nobody's numbers were as good as they said and deception is standard practice. After an extended conversation with him I have no doubt this crash is due to negligence
Sounds like rationalization. Even if it's TRUE it's still rationalization and a way to talk oneself into irresponsibility: lowering the bar.
Maybe I'm misunderstanding your comment, but this was months ago. No rationalizing needed.
Being trashed doesn't exactly provide confidence in what was spoken and the memory of it for later.
I was waiting for the moment when this would come up. A friend of mine had worked for Uber for a few years, and a couple of times when she came back to town, she'd mentioned over drinks that they have been "secretly" running autonomous vehicles around the ASU campus in Tempe for some years now.
Now THAT explains a lot of that "she came from the shadows" and "a human driver could not have prevented this" and "The car didn't do anything wrong" bullcrap we've been reading from the Tempe Police right after the accident, as well as the quick release of the highly misleading (in terms of lighting situation) dashcam video.
Don't bite the hand that feeds you...
Don't bite the hand that feeds you...
Yup, I was wondering why the police put up such a strong defence even before an investigation. I suppose this may explain at least some of that. I wonder if the post crash emails will ever come out.
The Uber-Arizona story is the American economic-political dystopia in microcosm.
For-profit corporation helps elected politician maintain power.
Elected politician uses the power of the state to promote the corporation and eliminate oversight that would have been in the public interest.
All of this is done in secret, of course. The rubes who elected you don't need to know what their "representatives" are doing in their names.
The politician wins, the corporation wins, and the public is the big loser.
For-profit corporation helps elected politician maintain power.
Elected politician uses the power of the state to promote the corporation and eliminate oversight that would have been in the public interest.
All of this is done in secret, of course. The rubes who elected you don't need to know what their "representatives" are doing in their names.
The politician wins, the corporation wins, and the public is the big loser.
I'm not sure what you mean by don't bite the hand, I understand that the police would have a vested interested in defending uber but I couldn't imagine the police destroying their credability over some stupid little thing in the big scheme of things.
I don't need to believe there was a conspiracy to protect Uber here. This spot apparently has a history of pedestrian crossing, and many police departments have a habit of assigning all blame to the pedestrian in a pedestrian/vehicle collision. This blame is reinforced by the driver's "evidence" (driver didn't see anything, but she wasn't looking) and the dashcam (can't see anything, but the quality is atrocious and not at all reflective of what most drivers would see).
One thing we should take away from this incident is that we need to seriously push back on immediate "pedestrian's fault" rulings when police make them after a pedestrian/vehicle collision. It was readily apparent within hours that the story didn't add up, but people only paid so much attention because it's the first such crash involving an autonomous vehicle.
The more interesting question is how that shockingly poor dashcam footage came to be. That quality is so poor that I find it hard to believe that Uber is actively using it as a sensor input. Hell, it's worse than most dashcams I've seen on Youtube...
One thing we should take away from this incident is that we need to seriously push back on immediate "pedestrian's fault" rulings when police make them after a pedestrian/vehicle collision. It was readily apparent within hours that the story didn't add up, but people only paid so much attention because it's the first such crash involving an autonomous vehicle.
The more interesting question is how that shockingly poor dashcam footage came to be. That quality is so poor that I find it hard to believe that Uber is actively using it as a sensor input. Hell, it's worse than most dashcams I've seen on Youtube...
> That quality is so poor that I find it hard to believe that Uber is actively using it as a sensor input.
It probably isn't used as sensor input. It's probably an off-the-shelf dashcam (the cheapest they could find that could also record the driver) added as an afterthought. It also wouldn't surprise me if the windshield were tinted, which would make the dashcam recording even darker (since it's behind the windshield).
It probably isn't used as sensor input. It's probably an off-the-shelf dashcam (the cheapest they could find that could also record the driver) added as an afterthought. It also wouldn't surprise me if the windshield were tinted, which would make the dashcam recording even darker (since it's behind the windshield).
I don't know about AZ, but windshield tinting is usually illegal in TX, except (I believe) for certain newer methods of tinting that are designed to filter only IR frequencies. But I think those newer forms are also fairly rare. In any case, I definitely wouldn't assume the windshield was tinted.
You are severely mistaken. Window tinting is legal in AZ (resident) and a quick search tells me TX as well.
Maybe you’re thinking of the black out style tinting that you see in movies but most people tint to filter out like 5-20% light.
Maybe you’re thinking of the black out style tinting that you see in movies but most people tint to filter out like 5-20% light.
Brief googling indicates that windshield tinting, beyond a small strip at the top, in both Arizona and Texas is illegal without medical exemption.
"Transparent material that is installed, affixed or applied to the topmost portion of the windshield if:
(a) The bottom edge of the material is at least twenty-nine inches above the undepressed driver's seat when measured from a point five inches in front of the bottom of the backrest with the driver's seat in its rearmost and lowermost position with the vehicle on a level surface.
(b) The material is not red or amber in color."[0]
and
"Windshields Sunscreening devices must be applied above the AS-1 line. If there is no AS-1 line, sunscreening devices must end five inches below the top of the windshield. Sunscreening devices may not be red, amber, or blue in color. Sunscreening devices, when measured in combination with the original glass, must have a light transmittance value of 25% or more. Sunscreening devices, when measured in combination with the original glass, must have a luminous reflectance value of 25% or less. A clear (un-tinted) UV film is allowed anywhere on the front windshield without a medical exemption being required."[1]
[0] https://www.azleg.gov/ars/28/00959-01.htm
[1] https://www.dps.texas.gov/rsd/vi/consumerinfo/windowTint.htm
"Transparent material that is installed, affixed or applied to the topmost portion of the windshield if:
(a) The bottom edge of the material is at least twenty-nine inches above the undepressed driver's seat when measured from a point five inches in front of the bottom of the backrest with the driver's seat in its rearmost and lowermost position with the vehicle on a level surface.
(b) The material is not red or amber in color."[0]
and
"Windshields Sunscreening devices must be applied above the AS-1 line. If there is no AS-1 line, sunscreening devices must end five inches below the top of the windshield. Sunscreening devices may not be red, amber, or blue in color. Sunscreening devices, when measured in combination with the original glass, must have a light transmittance value of 25% or more. Sunscreening devices, when measured in combination with the original glass, must have a luminous reflectance value of 25% or less. A clear (un-tinted) UV film is allowed anywhere on the front windshield without a medical exemption being required."[1]
[0] https://www.azleg.gov/ars/28/00959-01.htm
[1] https://www.dps.texas.gov/rsd/vi/consumerinfo/windowTint.htm
You're not making the distinction between "window tinting" and "windSHIELD tinting".
Big difference. I'm not aware of anywhere that it's legal to tint your windshield below the top 4-5 inches. (Not saying it isn't but it certainly isn't in AZ or TX)
Big difference. I'm not aware of anywhere that it's legal to tint your windshield below the top 4-5 inches. (Not saying it isn't but it certainly isn't in AZ or TX)
> I'm not aware of anywhere that it's legal to tint your windshield below the top 4-5 inches.
In my country, you can put a tinted film on the whole windshield, as long as it allows at least 75% of the light to pass (the side and rear windows can be much darker: the film must allow at least 28% of the light). They're a common sight.
In my country, you can put a tinted film on the whole windshield, as long as it allows at least 75% of the light to pass (the side and rear windows can be much darker: the film must allow at least 28% of the light). They're a common sight.
I assume the poor dashcam footage came out because it fits the narrative they were all trying to push. I don't think there is any reason to believe it was the input to the driving system, but either way, it does not look good for Uber, and if it is apparently deliberately misleading, or, worse, has been tampered with, then that would take it to a new level.
The release of the interior shot might have been an attempt to throw the 'safety' driver under the bus, so to speak, and while her inattentiveness was a very serious matter, it also does not look good for Uber.
The release of the interior shot might have been an attempt to throw the 'safety' driver under the bus, so to speak, and while her inattentiveness was a very serious matter, it also does not look good for Uber.
Yeah, they did not need to show the footage of her blatantly distracted away from the road. That was my biggest takeaway. I'm not sure who decided what footage got shown publicly but I immediately found that interesting to say the least.
I thought it looked a bit like they were throwing their human backup driver under the bus as well. Not cool.
Why is it not cool?
Certainly Uber the company is systemically responsible for this failure, but the driver also had a role in preventing harm, and they were negligent. There's more than enough blame to share.
Certainly Uber the company is systemically responsible for this failure, but the driver also had a role in preventing harm, and they were negligent. There's more than enough blame to share.
Imagine the police people having seen their governor (who effectively is their boss, as far as I know, so he is the "hand that feeds them") repeatedly be very, very protective and favorable towards Uber and the self-driving car testing program. Even if they never got an explicit order of the likes of "Courteously close your eyes when dealing with Uber!", they will be much more protective of Uber than they were if their boss was known as very critical regarding Uber. This is basic human instinct, and instinctive behavior often leads to decisions which might turn out to be obviously stupid in the long run.
This is extremely naive, have you ever worked in organizations?
I'm not being facetious, but a little top-down pressure can go a long way. This isn't about a grand conspiracy as much as it is about simple human nature, and the act to please those above you, keep your job, and move up yourself.
I'm not being facetious, but a little top-down pressure can go a long way. This isn't about a grand conspiracy as much as it is about simple human nature, and the act to please those above you, keep your job, and move up yourself.
Except the Tempe police don’t report to the governor, and the mayor of Tempe is a Democrat, and an opponent of the governor. There could have been as much pressure to make it look bad as to minimize it. But more likely there was mo pressure either way, and the police just blame pedestrians by default.
> but a little top-down pressure can go a long way
"The legal and philosophic aspects of obedience are of enormous importance, but they say very little about how most people behave in concrete situations. I set up a simple experiment at Yale University to test how much pain an ordinary citizen would inflict on another person simply because he was ordered to by an experimental scientist. Stark authority was pitted against the subjects' [participants'] strongest moral imperatives against hurting others, and, with the subjects' [participants'] ears ringing with the screams of the victims, authority won more often than not. The extreme willingness of adults to go to almost any lengths on the command of an authority constitutes the chief finding of the study and the fact most urgently demanding explanation."
https://en.wikipedia.org/wiki/Milgram_experiment
"The legal and philosophic aspects of obedience are of enormous importance, but they say very little about how most people behave in concrete situations. I set up a simple experiment at Yale University to test how much pain an ordinary citizen would inflict on another person simply because he was ordered to by an experimental scientist. Stark authority was pitted against the subjects' [participants'] strongest moral imperatives against hurting others, and, with the subjects' [participants'] ears ringing with the screams of the victims, authority won more often than not. The extreme willingness of adults to go to almost any lengths on the command of an authority constitutes the chief finding of the study and the fact most urgently demanding explanation."
https://en.wikipedia.org/wiki/Milgram_experiment
It's a bit misleading to post that quote without including the validity section as well. People are not as quick to abandon their ethics as the original results of that experiment suggest.
In 2012, Australian psychologist Gina Perry investigated Milgram's data and writings and concluded that Milgram had manipulated the results, and that there was "troubling mismatch between (published) descriptions of the experiment and evidence of what actually transpired." She wrote that "only half of the people who undertook the experiment fully believed it was real and of those, 66% disobeyed the experimenter". She described her findings as "an unexpected outcome" that "leaves social psychology in a difficult situation." In the journal Jewish Currents, Joseph Dimow, a participant in the 1961 experiment at Yale University, wrote about his early withdrawal as a "teacher", suspicious "that the whole experiment was designed to see if ordinary Americans would obey immoral orders, as many Germans had done during the Nazi period."
In 2012, Australian psychologist Gina Perry investigated Milgram's data and writings and concluded that Milgram had manipulated the results, and that there was "troubling mismatch between (published) descriptions of the experiment and evidence of what actually transpired." She wrote that "only half of the people who undertook the experiment fully believed it was real and of those, 66% disobeyed the experimenter". She described her findings as "an unexpected outcome" that "leaves social psychology in a difficult situation." In the journal Jewish Currents, Joseph Dimow, a participant in the 1961 experiment at Yale University, wrote about his early withdrawal as a "teacher", suspicious "that the whole experiment was designed to see if ordinary Americans would obey immoral orders, as many Germans had done during the Nazi period."
That's .. not how credibility works for the police.
I still see facts as:
1) Nothing was reflective
2) She was walking across the road between street lights in a random space.
3) If a person hit the woman it would not have been homicide or manslaughter even if we didn't know the driver was on their phone.
I would say the pedestrian was legally more than 50% at fault in this horrible accident.
http://www.alllaw.com/articles/nolo/auto-accident/driver-at-...
1) Nothing was reflective
2) She was walking across the road between street lights in a random space.
3) If a person hit the woman it would not have been homicide or manslaughter even if we didn't know the driver was on their phone.
I would say the pedestrian was legally more than 50% at fault in this horrible accident.
http://www.alllaw.com/articles/nolo/auto-accident/driver-at-...
The test for a self-driving car isn't "does it drive as well as a human," it's "does it drive as well as a self-driving car which normally beats a human should."
This might sound like it makes no difference, but the car had LIDAR. That the scene was badly lit, whether true or not, is irrelevant; the car didn't use ambient light to see. If it could hit her under these circumstances, then it could just as easily hit her at noon.
This might sound like it makes no difference, but the car had LIDAR. That the scene was badly lit, whether true or not, is irrelevant; the car didn't use ambient light to see. If it could hit her under these circumstances, then it could just as easily hit her at noon.
Which brings up the real question - do we want self-driving cars that ONLY drive as well as humans. I for one would be terrified at that criteria.
This is what gets me! I really wish they would talk about why the LIDAR didn't notice her. Was it off? Was it malfunctioning? Was there some other issue there? I've been puzzled about how it didn't see her!
The theory I've seen the most is that the LIDAR probably did detect her just fine, but Uber's software failed to act on the data.
According to Reuters, Uber had changed from a network of 7 LIDAR sensors on their vehicles to just one: https://www.reuters.com/article/us-uber-selfdriving-sensors-...
According to discussions I have read. The LIDAR was disconnected as part of a test of "visual only" navigation. Why didn't they keep the LIDAR on for safety?
If they did that I would say sue them out of existence. That killed a woman and harmed the industry.
Human drivers and other autonomous cars could have avoided this accident.
Uber's systems failed spectacularly, and should not have been on public roads.
Now we learn that Uber and the governor secretly collaborated to enable this, to neutralize any effective oversight, and to keep the public in the dark.
Blaming the victim doesn't illuminate any of these issues.
Uber's systems failed spectacularly, and should not have been on public roads.
Now we learn that Uber and the governor secretly collaborated to enable this, to neutralize any effective oversight, and to keep the public in the dark.
Blaming the victim doesn't illuminate any of these issues.
> I would say the pedestrian was legally more than 50% at fault in this horrible accident.
Which is fine for Uber-as-operator if they were operating in a state which uses a contributory negligence or modified comparative negligence rule, but a number of jurisdictions—Arizona, for instance—use a pure comparative negligence rule where, e.g., an injured party that is 99% responsible can recover 1% of damages from a party that is 1% responsible.
Moreover, it's no help for Uber-as-manufacturer if strict product liability principles are applied. Now, it's not clear to me that they would apply to Uber as manufacturer as long as the product isn't sold to a third party, but even if they don't apply under current law, Uber using operating a service directly with the product rather than selling it to other operators as an end-run around liability for an unreasonably dangerous product invites a change to the law which would extend such liability to manufacturer-operators.
Which is fine for Uber-as-operator if they were operating in a state which uses a contributory negligence or modified comparative negligence rule, but a number of jurisdictions—Arizona, for instance—use a pure comparative negligence rule where, e.g., an injured party that is 99% responsible can recover 1% of damages from a party that is 1% responsible.
Moreover, it's no help for Uber-as-manufacturer if strict product liability principles are applied. Now, it's not clear to me that they would apply to Uber as manufacturer as long as the product isn't sold to a third party, but even if they don't apply under current law, Uber using operating a service directly with the product rather than selling it to other operators as an end-run around liability for an unreasonably dangerous product invites a change to the law which would extend such liability to manufacturer-operators.
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.
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.
>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-... ).
I think it was both i) and ii).
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-... ).
I think it was both i) and ii).
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.
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.
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.
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.
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.
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.
These facts are only relevant if you disregard LIDAR and you believe that the camera showed what the human eye would see. Videos on youtube (https://www.youtube.com/watch?v=CRW0q8i3u6E and also https://www.youtube.com/watch?v=1XOVxSCG8u0&feature=youtu.be) and ABC news footage (http://abc7.com/video-shows-self-driving-ubers-fatal-collisi...) at the scene of the crash paint a much brighter picture. The street lights in the footage appear to be distinct while the background is dark, indicating the gain of the sensor would appear to be reduced. Try taking a picture of streetlights at night, if the street is visible, they will appear as stars. Additionally, the safety driver was not paying attention.
EDIT: In my opinion, the more you think about this crash the worse Uber looks. The LIDAR should have worked, the lighting was likely brighter than the video shows, the weather was good, no obvious distractor objects, the road was straight, no parties were impaired. The pedestrian had a big bulky bike and crossed at right angles to the car. This is nearly the most basic test of self-driving cars and it failed. The only way Uber would look worse is if it had happened in day time. On top of that, Uber disabled the Volvo auto-braking safety system. (https://www.sfgate.com/business/article/Uber-Disabled-Volvo-...)
EDIT: In my opinion, the more you think about this crash the worse Uber looks. The LIDAR should have worked, the lighting was likely brighter than the video shows, the weather was good, no obvious distractor objects, the road was straight, no parties were impaired. The pedestrian had a big bulky bike and crossed at right angles to the car. This is nearly the most basic test of self-driving cars and it failed. The only way Uber would look worse is if it had happened in day time. On top of that, Uber disabled the Volvo auto-braking safety system. (https://www.sfgate.com/business/article/Uber-Disabled-Volvo-...)
> If a person hit the woman it would not have been homicide or manslaughter even if we didn't know the driver was on their phone.
That's because the bar for vehicular manslaughter is incredibly high. There is a huge range of shitty road behaviour that is not illegal, but you should never do, for the sake of yourself and other people.
Driving faster then you can see is one of those behaviors. Hitting someone because you had legal right of way is another. Right of way is good and all, but your first job as a driver is to avoid crashes. Right of way is just a set of guidelines for avoiding crashes... And a way to assign legal blame when one happens.
When I sea kayak, I have right of way against large boats in some situations. I'd be a complete, and dead idiot if I were to exercise my legal right to stay on course.
That's because the bar for vehicular manslaughter is incredibly high. There is a huge range of shitty road behaviour that is not illegal, but you should never do, for the sake of yourself and other people.
Driving faster then you can see is one of those behaviors. Hitting someone because you had legal right of way is another. Right of way is good and all, but your first job as a driver is to avoid crashes. Right of way is just a set of guidelines for avoiding crashes... And a way to assign legal blame when one happens.
When I sea kayak, I have right of way against large boats in some situations. I'd be a complete, and dead idiot if I were to exercise my legal right to stay on course.
If you even believe the video is representative of what a human would see (which is totally false, as recent local videos of that stretch have shown), then the car was driving beyond its headlights. That's unsafe driving.
Deer are not reflective. Random objects that have fallen off a truck are not reflective. People are often not reflective, and there is no law requiring them to be.
And this woman was wheeling a bright pink bike, which shows up quite well. (It may also have had standard spoke reflectors, which might not show up in this video. hard to say.)
And finally, there weren't any crosswalks in that area for her to use. Yes, she misjudged the speed of the car, but her choice to cross the road at all is pretty solid.
Deer are not reflective. Random objects that have fallen off a truck are not reflective. People are often not reflective, and there is no law requiring them to be.
And this woman was wheeling a bright pink bike, which shows up quite well. (It may also have had standard spoke reflectors, which might not show up in this video. hard to say.)
And finally, there weren't any crosswalks in that area for her to use. Yes, she misjudged the speed of the car, but her choice to cross the road at all is pretty solid.
>If you even believe the video is representative of what a human would see (which is totally false, as recent local videos of that stretch have shown), then the car was driving beyond its headlights. That's unsafe driving.
This point needs to be understood by everyone discussing the issue.
This point needs to be understood by everyone discussing the issue.
Well there is a crosswalk at the light..intersection of Curry & Mill.
What's interesting is the X paved walkway in the median. You can see it here: https://www.google.com/maps/place/Tempe,+AZ/@33.4364759,-111...
What's interesting is the X paved walkway in the median. You can see it here: https://www.google.com/maps/place/Tempe,+AZ/@33.4364759,-111...
I asked this [1] in ask HN sometime back
https://news.ycombinator.com/item?id=16696681
No comments and it is removed from the ask page as far as I can see....
Why are people so indifferent and why are they just standing by and not asking relevant questions and not demanding for proper test/procedures in place before continuing the test of SDVs?
https://news.ycombinator.com/item?id=16696681
No comments and it is removed from the ask page as far as I can see....
Why are people so indifferent and why are they just standing by and not asking relevant questions and not demanding for proper test/procedures in place before continuing the test of SDVs?
Accidents like this will spur the AV community of companies to form a body (AVI?) to protect AVs from bad press by enacting a set of rules for all AVs to abide by.
Eventually, gov't policy will take AVI rules and recommendations and formalize them into their own books. But until then, the whole thing is deeply experimental. And regulation is premature. Otherwise, you may be crushing potentially great ideas unknowingly.
Eventually, gov't policy will take AVI rules and recommendations and formalize them into their own books. But until then, the whole thing is deeply experimental. And regulation is premature. Otherwise, you may be crushing potentially great ideas unknowingly.
>But until then, the whole thing is deeply experimental. And regulation is premature
If it is deeply experimental, then it should be conducted in experimental settings, and not on public roads. If regulation is premature, then public tests are also premature. I am surprised I have to explain this. To be on public road the vehicle should pass a minimum benchmark that shows that it can consistently handle at least all the common scenarios..
There is no urgent need for self driving vehicles. It is not like they are going to solve some of the pressing issues that humanity is facing right now. It is just a curiosity, and can may be boost the profits of some corporations. That is about it.
I don't think it is worth or just to risk lives of unsuspecting public towards that end..
If it is deeply experimental, then it should be conducted in experimental settings, and not on public roads. If regulation is premature, then public tests are also premature. I am surprised I have to explain this. To be on public road the vehicle should pass a minimum benchmark that shows that it can consistently handle at least all the common scenarios..
There is no urgent need for self driving vehicles. It is not like they are going to solve some of the pressing issues that humanity is facing right now. It is just a curiosity, and can may be boost the profits of some corporations. That is about it.
I don't think it is worth or just to risk lives of unsuspecting public towards that end..
The best angle I have seen for this is: if we knew a safer way than road testing to teach something to drive we would do it. Teens are way more dangerous than experienced drivers, even after classes and 100 hours of training, but we still let them on the road. It’s imperfect but the only known way to get the experience needed to get better is drive in the real world before you are 100% safe too (and nothing ever hits 100).
> if we knew a safer way than road testing to teach something to drive we would do it.
Exactly. The tests and regulations for self driving vehicles only need to ensure this. That it has been taught to the limits of what can be done in an experimental setting, before it is allowed to be on public roads.
Right now this is not the case. A company can put a SDV on road, that is not even capable of avoiding hitting a pedestrian right in front of it in plain sight at more than breaking distance. I mean, there is nothing to ensure this does not happen. That is my whole argument.
There should be a minimum benchmark that any experimental SDV should pass consistently to be tested on public roads. There should also be some process to ensure the continued validity of the license thus given. For example, it should only be valid until a software/firmware update.
I short, authorities along with engineers and company execs need to sit down and think this through and create a framework before this program can continue..
Exactly. The tests and regulations for self driving vehicles only need to ensure this. That it has been taught to the limits of what can be done in an experimental setting, before it is allowed to be on public roads.
Right now this is not the case. A company can put a SDV on road, that is not even capable of avoiding hitting a pedestrian right in front of it in plain sight at more than breaking distance. I mean, there is nothing to ensure this does not happen. That is my whole argument.
There should be a minimum benchmark that any experimental SDV should pass consistently to be tested on public roads. There should also be some process to ensure the continued validity of the license thus given. For example, it should only be valid until a software/firmware update.
I short, authorities along with engineers and company execs need to sit down and think this through and create a framework before this program can continue..
>There is no urgent need for self driving vehicles. It is not like they are going to solve some of the pressing issues that humanity is facing right now. It is just a curiosity, and can may be boost the profits of some corporations. That is about it.
If you don't consider the second leading unnatural cause of death in the United States (only recently surpassed by suicide) killing 40,000 people a year an "issue", then sure.
If you don't consider the second leading unnatural cause of death in the United States (only recently surpassed by suicide) killing 40,000 people a year an "issue", then sure.
> If you don't consider the second leading unnatural cause of death in the United States (only recently surpassed by suicide) killing 40,000 people a year an "issue", then sure.
It is an issue, But not one among the "pressing issues humanity is facing right now" and definitively not worth risking the lives of unsuspecting public.
Interesting that you intend to solve the issue of "40,000 people being killed an year" by putting even more dangerous things on the road, that may or may not become better drivers in a period of god knows how long...
The only sure thing about SDV is that it ll save these companies a boat load of money during the time they are allowed to operate them on the roads...
It is an issue, But not one among the "pressing issues humanity is facing right now" and definitively not worth risking the lives of unsuspecting public.
Interesting that you intend to solve the issue of "40,000 people being killed an year" by putting even more dangerous things on the road, that may or may not become better drivers in a period of god knows how long...
The only sure thing about SDV is that it ll save these companies a boat load of money during the time they are allowed to operate them on the roads...
If the "lives being risked" are less than 40k a year, then the math says that yes, it is worth it.
Status quo is the risk. It is a decision to make the active choice of killing 40k people a year, to prevent something else.
Status quo is the risk. It is a decision to make the active choice of killing 40k people a year, to prevent something else.
To be clear, the parent poster was arguing that we need to start testing these vehicles even before some sort of tests and legal framework can be devised to ensure them to be road worthy..
My argument was that we have no such urgency, and we can and should wait till we have such a framework in place.
My argument was that we have no such urgency, and we can and should wait till we have such a framework in place.
If that delays things by X years, then that mean X times 40k people will die.
That's the tradeoff.
Yes these rules and regulations are important, but are they worth X times 40k lives?
Perhaps one could argue that they are. Just as long as you are honest about the trade off of "I am willing to sacrifice 40K times X lives, because this issue is THAT important".
That's the tradeoff.
Yes these rules and regulations are important, but are they worth X times 40k lives?
Perhaps one could argue that they are. Just as long as you are honest about the trade off of "I am willing to sacrifice 40K times X lives, because this issue is THAT important".
>if that delays things by X years, then that mean X times 40k people will die.
>That's the tradeoff.
No. With the unknown value X (and numerous other, very relevant unknowns that you have conveniently omitted), that is not a trade off. That is speculation...No, that is worse than speculation. That is just wrong.
That is like some one saying
"Hey you, you are so unlucky that you might kill X number of people every year if you drive a car, so you should not drive a car. Are the lives of X people really worth risking just so that you can drive around in a car?". See? without the value of X, that is just meaningless.
You just can't plug in variables and construct arguments as if they are of certain values and with no sort of quantification. That is like..fundamentally wrong....
>That's the tradeoff.
No. With the unknown value X (and numerous other, very relevant unknowns that you have conveniently omitted), that is not a trade off. That is speculation...No, that is worse than speculation. That is just wrong.
That is like some one saying
"Hey you, you are so unlucky that you might kill X number of people every year if you drive a car, so you should not drive a car. Are the lives of X people really worth risking just so that you can drive around in a car?". See? without the value of X, that is just meaningless.
You just can't plug in variables and construct arguments as if they are of certain values and with no sort of quantification. That is like..fundamentally wrong....
"Otherwise, you may be crushing potentially great ideas unknowingly."
Along with crushing some human bodies along the way.
Along with crushing some human bodies along the way.
"We had to kill the pedestrians in order to save them..."
Secrecy in policy has no place in a democracy. Keeping your spies hidden or war plans makes sense, but not things that directly affect the public.
This article is pretty sensational, it's not really a secret. Ducey has been a big proponent of autonomous vehicles, it's known to anyone paying attention. When Uber, Waymo, and GM (the 3 biggest autonomous development operations in America) started testing in the Phoenix metro area it was publicly announced.
Michigan, Florida, and Georgia also have similarly lassiez faire policies towards autonomous vehicle testing, but most of the action has been around Phoenix because of Ducey's active courting of the autonomous driving industry. In the wake of the Uber incident, a violation of Ducey's trust in Uber's obligation to test responsibly, he has ordered Uber to suspend their operations:
https://mobile.twitter.com/biancabuono/status/97842803263227...
Interestingly, cities don't have much power to decide whether, or how, autonomous vehicles can test in their cities. When NY state governor Andrew Cuomo and Cruise Automation jointly announced GM's plan to begin testing in NYC, Bill de Blasio and the mayor's office were surprised, they hadn't heard anything about it, but NYC testing appears to be going ahead anyhow.
Michigan, Florida, and Georgia also have similarly lassiez faire policies towards autonomous vehicle testing, but most of the action has been around Phoenix because of Ducey's active courting of the autonomous driving industry. In the wake of the Uber incident, a violation of Ducey's trust in Uber's obligation to test responsibly, he has ordered Uber to suspend their operations:
https://mobile.twitter.com/biancabuono/status/97842803263227...
Interestingly, cities don't have much power to decide whether, or how, autonomous vehicles can test in their cities. When NY state governor Andrew Cuomo and Cruise Automation jointly announced GM's plan to begin testing in NYC, Bill de Blasio and the mayor's office were surprised, they hadn't heard anything about it, but NYC testing appears to be going ahead anyhow.
As well as rules and regulations being privitized. No one should have to pay someone like ASME to access the law.
If this was a company from any other sector, such as a bank, people would start crying corruption; not that there probably was any. I think the governor was trying to court one of the largest Silicon Valley companies and that's trying to develop some "exciting" new technology and bent over backwards for them, which is cool and all, but he didn't perform his due diligence.
I find it interesting how the governor just "trusted" that the company was obeying all federal laws and guidelines and didn't check to see if: (a) they were indeed complying (b) there were any outside of the car itself (c) if any exist about self-driving cars, how those guidelines ensure public safety.
This is the exact kind of thing that ends up driving knee-jerk legislation to go the opposite direction. Does anyone know what federal laws and guidelines (from NTSB or other dept) exist that govern self-driving cars?
I find it interesting how the governor just "trusted" that the company was obeying all federal laws and guidelines and didn't check to see if: (a) they were indeed complying (b) there were any outside of the car itself (c) if any exist about self-driving cars, how those guidelines ensure public safety.
This is the exact kind of thing that ends up driving knee-jerk legislation to go the opposite direction. Does anyone know what federal laws and guidelines (from NTSB or other dept) exist that govern self-driving cars?
Just guidelines at this point.
https://www.nhtsa.gov/press-releases/us-dot-releases-new-aut...
https://www.nhtsa.gov/press-releases/us-dot-releases-new-aut...
> NHTSA strongly encourages States not to codify this Voluntary
Guidance (that is, incorporate it into State statutes) as a legal
requirement for any phases of development, testing, or
deployment of ADSs. Allowing NHTSA alone to regulate the
safety design and performance aspects of ADS technology will
help avoid conflicting Federal and State laws and regulations
that could impede deployment.
Can someone clarify that this please? To me it appears that this body is saying that, "States, Please don't make laws requiring minimal standards for SDVs for things, including deployment on public roads, because that would impede deployment. Allow us to handle all that"
But they didn't do it, right? SDV's were on roads even before the "safety design and performance aspects of ADS technology" was regulated, right?
Can someone clarify that this please? To me it appears that this body is saying that, "States, Please don't make laws requiring minimal standards for SDVs for things, including deployment on public roads, because that would impede deployment. Allow us to handle all that"
But they didn't do it, right? SDV's were on roads even before the "safety design and performance aspects of ADS technology" was regulated, right?
>not to codify this Voluntary Guidance ... as a legal requirement for any phases of development, testing, or deployment
I don't think it's meant to impede anything. It's just saying "don't use this as a go-ahead to base your state's SDV laws and guidances since this document will change".
I don't think it's meant to impede anything. It's just saying "don't use this as a go-ahead to base your state's SDV laws and guidances since this document will change".
If you watch the video, it's clear uber/the safety driver were at fault. Which made the rapid announcement that "nothing could have been done" by the police so surprising when the evidence was so clear. But now with this coming out, it all makes sense.
I like that uber called the governor a "thought leader" as doublespeak for someone in their pocket, or who at least gave them their ear.
But doing this in secret is the real icing on the cake. great leadership there. Should we expect anything different from republicans/russians at this point?
I like that uber called the governor a "thought leader" as doublespeak for someone in their pocket, or who at least gave them their ear.
But doing this in secret is the real icing on the cake. great leadership there. Should we expect anything different from republicans/russians at this point?
I had a passenger (circa 2015) who was gung-ho about driving for Uber. He'd driven for the taxi company for a few shifts, but couldn't make money. He'd ordered a taxi to take his family to the airport because they had luggage. He couldn't risk getting a miniature uber that wouldn't be big enough. He'd bought the cheapest black BMW he could find so that he could get Uber Black fares too, and was going through the state's procedures to become a vehicle-for-hire. He whined about how much commercial insurance was going to cost him...
Governor Ducey threw us taxi drivers under the bus as soon as he took office in January 2015. (edit2: the NFL's Superbowl was in Glendale, AZ that month.) Arizona has very minimal regulations for taxis and vehicles for hire -- basically a vehicle inspections, background checks for drivers and commercial insurance. Anyone who wants to become a taxi company is welcome, no medallion required. Taxi companies are trusted to take care of the regulations for themselves. Essentially Ducey said he was going to stop enforcement of the state's existing minimal laws pertaining to private transportation until the legislators came up with replacements.
Ducey's predecessor, Jan Brewer, vetoed a bill that gave personal vehicles transporting passengers for "ride sharing" an exemption from having commercial insurance - at the time Uber's insurance was "trust us, you're covered". The company I drove for spent a ton of money on their lobbyists to get that bill vetoed, and for a bill that was somewhat-fair for everyone.
> Arizona was not always a friendly state for Uber. In April 2014, then-governor Jan Brewer vetoed legislation that would have exempted taxi-hailing companies from insurance regulations imposed on traditional taxis.
(I believe this is a typo - the veto'd bill exempted personal vehicles used for "ride share" from the insurance regulations for taxis).
The taxi company I drove for is doing fine. They've adapted, and now lease unbranded vehicles that drivers can use with any of the app-based services. Their contracts pay better than the apps' non-surge fares. One of my driver-friends said the scene on the ground is okay - fare volume is down, but quality is up (calls are from people people who really want a ride).
(minor edits)
Governor Ducey threw us taxi drivers under the bus as soon as he took office in January 2015. (edit2: the NFL's Superbowl was in Glendale, AZ that month.) Arizona has very minimal regulations for taxis and vehicles for hire -- basically a vehicle inspections, background checks for drivers and commercial insurance. Anyone who wants to become a taxi company is welcome, no medallion required. Taxi companies are trusted to take care of the regulations for themselves. Essentially Ducey said he was going to stop enforcement of the state's existing minimal laws pertaining to private transportation until the legislators came up with replacements.
Ducey's predecessor, Jan Brewer, vetoed a bill that gave personal vehicles transporting passengers for "ride sharing" an exemption from having commercial insurance - at the time Uber's insurance was "trust us, you're covered". The company I drove for spent a ton of money on their lobbyists to get that bill vetoed, and for a bill that was somewhat-fair for everyone.
> Arizona was not always a friendly state for Uber. In April 2014, then-governor Jan Brewer vetoed legislation that would have exempted taxi-hailing companies from insurance regulations imposed on traditional taxis.
(I believe this is a typo - the veto'd bill exempted personal vehicles used for "ride share" from the insurance regulations for taxis).
The taxi company I drove for is doing fine. They've adapted, and now lease unbranded vehicles that drivers can use with any of the app-based services. Their contracts pay better than the apps' non-surge fares. One of my driver-friends said the scene on the ground is okay - fare volume is down, but quality is up (calls are from people people who really want a ride).
(minor edits)
I just saw the dashcam of the bicyclist getting hit. It seems like had waymo's lidar been installed, the cyclist would have been detected.
Move fast and break things may not be suitable for all development projects after all.
If after all the deregulation companies got at the federal level, plus even more deregulation in Arizona, they still had to do this, then something smells very bad here.
What's really surprising, however, is the sort of decisions a US governor (or maybe just Arizona) is apparently involved in. In my small European country, a license to operate some new and potentially dangerous technology is a matter for some executive agency, which would routinely involve at least one expert on the technology and a lawyer.
Now, obviously, that agency is part of the executive branch and theoretically below the Governor in a hierarchy. But in practice it's very similar to the federal US Government: the US President actually can't directly order some low-level bureaucrats at the FDA/SEC/etc to do anything. Such an order has to go through all the levels of that hierarchy. And if anyone refuses, their firing also has to follow that path.
Individual decisions are almost completely independent from top leadership. A series of questionable decisions will probably get you reassigned to the mail room, but even then past decisions would usually remain in force.
That system does not only strengthen the role of subject matter experts in the process and reduce the appearance of arbitrary favouritism. It's also in the best interest of politicians: to put your signature under an order allowing an individual safety-critical project without following some sort of established process is almost certain to end your career the first time one of those projects goes south.
Less relevant nitpicks:
- What exactly did Arizona stand to gain from a few hundred autonomous cars with minimum-wage drivers behind the wheel? It's not like Uber was going to move their software engineering to Tuscon, was it?
- Getting free use of Uber's offices in San Francisco is obviously bribery if it was for personal or campaign staff. Although I guess it's technically legal in the US if properly reported.