AP2.0 Cameras: Capabilities and Limitations?

[buttershrimp]

Newer Version

 

[SlicedBr3ad]

Dashcam video of deadly self-driving Uber crash released

They needed more Lidars to see in the dark. Goes to show to everyone saying Lidars will solve vision Because they can see in the dark. Also, if the vision was available, This literally was a basic two objects on direct path of collision problem.

Experts: Uber self-driving system should have spotted woman

Exactly what I was saying but this guy is allegedly an expert.

 

[buttershrimp]

Newer Version

@Kristoffer Helle is back!

 

[VT_EE]

Experts: Uber self-driving system should have spotted woman

Exactly what I was saying but this guy is allegedly an expert.

Based on that video, a human driver would have hit her too. There is no way I would have been able to see the pedestrian in time to stop or swerve. Autonomous vehicles will not stop all accidents from happening, but they will help prevent a majority of them. People still have to be smart and not cross dark roads directly in front of traffic. It's sad that Uber is getting bashed for this when it was clearly not their fault (as stated by the police).

 

[buttershrimp]

Based on that video, a human driver would have hit her too. There is no way I would have been able to see the pedestrian in time to stop or swerve. Autonomous vehicles will not stop all accidents from happening, but they will help prevent a majority of them. People still have to be smart and not cross dark roads directly in front of traffic. It's sad that Uber is getting bashed for this when it was clearly not their fault (as stated by the police).

It’s interesting, I bring a lot of bias to the situation with uber, I am upset for the engineers and the fact that someone is being aggressive in early deployment over time, before this video surfaced, I hoped that the incident would shown a person emerging from bushes on their bike next to a lane... instead it shows a very avoidable scenario, certainly a scenario that regulators and citizens would deem unacceptable to test on public roads.

Whether or not it could happen or should be deemed acceptable given the stakes of autonomous driving, is my point. They’ve decided to be on the edge of the technology and to be aggressive, to the point where they have had mutiple incidents in testing.

This accident is going to make things harder for everyone including Tesla to develop technology. Regulatory environment will tighten because of a company that seems to be cavalier.

 

[SlicedBr3ad]

It’s interesting, I bring a lot of bias to the situation with uber, I am upset for the engineers and the fact that someone is being aggressive in early deployment over time, before this video surfaced, I hoped that the incident would shown a person emerging from bushes on their bike next to a lane... instead it shows a very avoidable scenario, certainly a scenario that regulators and citizens would deem unacceptable to test on public roads.

Whether or not it could happen or should be deemed acceptable given the stakes of autonomous driving, is my point. They’ve decided to be on the edge of the technology and to be aggressive, to the point where they have had mutiple incidents in testing.

This accident is going to make things harder for everyone including Tesla to develop technology. Regulatory environment will tighten because of a company that seems to be cavalier.

looking at the video and saying "i couldn't avoid it either" is correct because humans need headlights to see and biker literally stepped into the path of the vehicle. now imagine if we could see at night and same scenario. or alternative, imagine daytime and person A walking the bike slowly but across inappropriate place on the road and directly into path in front of person B who is driving at 40 mph. Person B (driver) is responsible to see and avoid an accident. Saying UBER automation (being person B) and seeing like its daytime couldn't avoid this is total BS. now i'm unhappy that police department went out and published "uber unlikely" responsible. shows lack of understand how this technology works.

 

[malcolm]

This accident is going to make things harder for everyone including Tesla to develop technology. Regulatory environment will tighten because of a company that seems to be cavalier

TBH some sort of incident like this was inevitable giving the variety, extent and pressure of the competition between manufacturers.

Prospect of greater fragmentation of state legislation which will make a coast-to-coast demo interesting to set up.

Can it be any coast-to-coast route? Canada town becomes 'snowtonomous' research hub as automakers plan on making self-driving cars that can operate in snow and sleet- Technology News, Firstpost

 

[Uncle Paul]

Wondering why a woman would push her bicycle directly into the path of an oncoming car.

You could see, also, that the human monitoring the vehicle was not engaged in the monitoring he was supposed to be doing. Certaintly not looking where they were going as well.

Classic example of why human error causes so many accidents.

 

[CK_Stuggi]

The car was driving straight for three seconds prior to the collision. How come the pedestrian did not see and react to the certainly visible approaching lights of the car?
In this situation, even a skilled attentive driver in full control would have had a very difficult time to react by a combination of hard breaking and a left swerve to avoid a collision.
The 2 questions I'd have in this case are;
1. LIDAR is usually reasoned to detect objects that cameras and radar can't - why did lidar fail here?
Too short range/long processing?
Not usable at this speed?

2. Why, given the driving speed and lighting situation, why were no high beams used?
Given the location (33.436824,-111.942800) this would have been an option for a human driver. An option probably not many human drivers would have taken, but certainly one that existed and that could have increased the ability for the controller of the car (human or other) to recognize the situation and react accordingly. And maybe help the pedestrian become alerted to the oncoming vehicle.
Not sure how legal using high beams in that area would be (not usually driving in the US)?
This is a tragic incident and I hope the backlash to autonomous driving will not be excessive. It would be very interesting to see more info regarding the sensor info around this.
Considering that in 2016 around 6,000 pedestrians were killed in traffic (by humans), I'd be wary at this stage to proclaim "this is why I'd never trust self driving tech".

A friend of mine gave to consider that the camera has rather high contrast as opposed to the human eye, so had the driver been more attentive he/she should have been able to see the pedestrian sooner.

 

[buttershrimp]

looking at the video and saying "i couldn't avoid it either" is correct because humans need headlights to see and biker literally stepped into the path of the vehicle. now imagine if we could see at night and same scenario. or alternative, imagine daytime and person A walking the bike slowly but across inappropriate place on the road and directly into path in front of person B who is driving at 40 mph. Person B (driver) is responsible to see and avoid an accident. Saying UBER automation (being person B) and seeing like its daytime couldn't avoid this is total BS. now i'm unhappy that police department went out and published "uber unlikely" responsible. shows lack of understand how this technology works.

I should clarify that I don’t blame uber automation either... the technology is not the issue to me.... I know this sounds odd, but it’s the way it was tested and the ramifications of this type of error on public roads. This is a big event from a company that has a reputation for not prioritizing public safety .... the type of thing Ralph Nader would have had an aneurysm over

 

[agtdDelirium]

Lets have some perspective.

From the CDC:
"In 2015, 5,376 pedestrians were killed in traffic crashes in the United States. This averages to one crash-related pedestrian death every 1.6 hours. Additionally, almost 129,000 pedestrians were treated in emergency departments for non-fatal crash-related injuries in 2015. Pedestrians are 1.5 times more likely than passenger vehicle occupants to be killed in a car crash on each trip."

From this data, 15 pedestrians were killed and 353 pedestrians were struck the same day the Uber vehicle struck the woman. Obviously the number of human driven cars far outweigh autonomous vehicles at this point, but one pedestrian killed by an autonomous vehicle should not restrict development of what will almost assuredly prevent deaths in the future. Just like most human driver vs pedestrian accidents, Uber likely had a chance event at a very bad moment in time. You would have never heard about this woman's death nationally if not for being struck by an Uber vehicle.

Pushing the technology in testing is the only way to continue development. This woman's death will likely cause Uber to change something about their system that may prevent the next 15 similar deaths. Safety technology has always progressed this way, similar to seatbelts, airbags, or anything else safety related to our lives.

I am not calloused to this woman's death. It is a tragedy just as the 14 other similar deaths that statistically occurred that day. Her death should aid in the development of this technology which needs to continue. I'm sure Uber will be paying a large settlement to this woman's family. Probably not so for every other victim that day. Those are also the ones we need to remember who will benefit greatly when this technology matures.

 

[lunitiks]

one pedestrian killed by an autonomous vehicle should not restrict development

one pedestrian killed by an autonomous vehicle should not restrict development

one pedestrian killed by an autonomous vehicle should not restrict development

See what happens when I take that sentence out of context and repeat it? I become a politician!

 

[buttershrimp]

I wish we could separate the thread into categories... 1) how to go about developing a technology 2) the technology itself 3) Uber

My issue is with Uber, uber uber. Let’s take a look at this and then ask ourselves how much Uber has seriously screwed the people working on the technology and are enthusiastic about autonomous cars. Remember, this product is in TESTING!!!!!!



Uber gets a ticket
In December 2016, Uber's computer-controlled car was caught on video running a red light, four seconds after the light turned red. Uber said that the violation was a human error since a person was required to sit behind the wheel. The company suspended the driver after California regulators ordered a rollback of Uber’s self-driving cars citing pedestrian safety.

 

[daktari]

@Uncle Paul Car was travelling at 38 mph, equals 17 m/s. Victim crossed 5 lanes, and was almost over - 15 meters at a pace of between 1m/s to 2 m/s.
So car could have been as far as 255 meters away when she started crossing, that is 2.8 football fields. The car hit the pedestrian, not the other way!

@agtdDelirium You talk about research ethics. In medicine, this is probably easier than in autonomous driving. People are sick and are interested in trying new and maybe better treatment. But you don't induce a disease on well being human to try to treat it.
This Uber experiment killed an innocent, that maybe would never be hit by a car and could live long and prosper. She did not care about Ubers experiment. So if Uber, for a period of a year or two, increased the number of deaths by 2-fold, is that OK because they will reduce it again later? Not easy ehtically this experiment....

 

[buttershrimp]

Lets have some perspective.

From the CDC:
"In 2015, 5,376 pedestrians were killed in traffic crashes in the United States. This averages to one crash-related pedestrian death every 1.6 hours. Additionally, almost 129,000 pedestrians were treated in emergency departments for non-fatal crash-related injuries in 2015. Pedestrians are 1.5 times more likely than passenger vehicle occupants to be killed in a car crash on each trip."

From this data, 15 pedestrians were killed and 353 pedestrians were struck the same day the Uber vehicle struck the woman. Obviously the number of human driven cars far outweigh autonomous vehicles at this point, but one pedestrian killed by an autonomous vehicle should not restrict development of what will almost assuredly prevent deaths in the future. Just like most human driver vs pedestrian accidents, Uber likely had a chance event at a very bad moment in time. You would have never heard about this woman's death nationally if not for being struck by an Uber vehicle.

Pushing the technology in testing is the only way to continue development. This woman's death will likely cause Uber to change something about their system that may prevent the next 15 similar deaths. Safety technology has always progressed this way, similar to seatbelts, airbags, or anything else safety related to our lives.

I am not calloused to this woman's death. It is a tragedy just as the 14 other similar deaths that statistically occurred that day. Her death should aid in the development of this technology which needs to continue. I'm sure Uber will be paying a large settlement to this woman's family. Probably not so for every other victim that day. Those are also the ones we need to remember who will benefit greatly when this technology matures.

Let’s also consider the important question now that this incident has occurred: how many more deaths will there be due to the red tape that will likely occur to implement technology? How long will other manufacturers technology implementation be delayed because of Uber not testing this scenario out in an Uber development facility?

 

[animorph]

Absolutely a problem with Uber. No way a slow moving pedestrian in the clear, even in the dark, should be hit. The tech was inadequate.

Some degree of fault with all three though, pedestrian jaywalking in front of a moving car, an autonomous car that can't stop for pedestrians, and a backup driver not ready to react quickly. A better response from any of the three could have prevented this.

 

[run-the-joules]

Absolutely a problem with Uber. No way a slow moving pedestrian in the clear, even in the dark, should be hit. The tech was inadequate.

Some degree of fault with all three though, pedestrian jaywalking in front of a moving car, an autonomous car that can't stop for pedestrians, and a backup driver not ready to react quickly. A better response from any of the three could have prevented this.

Indeed. I'll still put primary blame on the pedestrian though because there's a basic thought process that she failed to follow:

1. Do cars drive in the place I want to walk?

If you answered "yes" to #1:
2. Are there any protections and/or indicators for pedestrians if I want to walk there, such as a crosswalk, lights, or railings?


If you answered "no" to #1 and wish to proceed:
Keep an eye out for things that can kill you.

If you answered "no" to #2 and wish to proceed:
Keep an eye out for things that can kill you.

If you answered "yes" to #2 and wish to proceed:
Keep an eye out for things that can kill you.

 

[Cirrus MS100D]

Not to further derail this thread, but how many accidents did an Uber AV (or similar) prevent? I have no idea and the answer could be zero, but if it prevented one, does that mean it's back to even? If it prevented 2, are we still ahead?

 

[buttershrimp]

Just came across this from other thread... felt it was significant to cross post.


EXCLUSIVE: 'Safety driver' of self-driving Uber which killed pedestrian had string of traffic offenses as well as a felony - but was given OK by Uber to be part of high-profile pilot scheme

• Rafaela Vasquez was the 'safety driver' of the autonomous Uber that hit and killed Elaine Herzburg, 49, in Tempe, Arizona on Sunday
• DailyMail.com can disclose Vasquez had been hit with a string of moving violations such as failing to stop at a red light and speeding, in recent years
• She was cited for driving with a suspended license in 2008 and again in 2009
• Uber applies same standard for self-driving car hires as for regular Uber drivers of no more than three minor moving driving offenses in last three years
• Vasquez was also revealed to have had felony convictions for attempted armed robbery for which she served more than three years in prison in 2001
• Uber had not disclosed her lengthy history of driving offenses in its public statements about the death
• The company issued a statement referring to its hiring policy stating, 'Everyone deserves a fair chance'

Read more: Operator of self-driving Uber had a history of traffic violations | Daily Mail Online
Follow us: @MailOnline on Twitter | DailyMail on Facebook

 

[agtdDelirium]

@agtdDelirium You talk about research ethics. In medicine, this is probably easier than in autonomous driving. People are sick and are interested in trying new and maybe better treatment. But you don't induce a disease on well being human to try to treat it.
This Uber experiment killed an innocent, that maybe would never be hit by a car and could live long and prosper. She did not care about Ubers experiment. So if Uber, for a period of a year or two, increased the number of deaths by 2-fold, is that OK because they will reduce it again later? Not easy ehtically this experiment....[/QUOTE]

I agree and disagree with your statement at the same time. You cite medicine and this is a good example to use. Many medical therapies are developed with trials on people with the intent on improving the outcome for those patient's, but until trialed that outcome is not known. Numerous new medical treatments and devices that are thought to have benefit have been found to be detrimental when trialed. This often leads to new/changed treatments that are then found to improve outcomes. Most medical trials are consented to by those patients involved but not all. Some require immediate care decisions that do not allow consent. All these studies do get rigorously reviewed by IRB's (Institutional review boards) before they are allowed to proceed. There have definitely been "innocents" harmed in the progress to save the larger masses. One can debate whether more oversight should be made for all manufacturers of autonomous driving systems. Government oversight will slow this development to a crawl and in my opinion cost many more lives in the long run.

Let’s also consider the important question now that this incident has occurred: how many more deaths will there be due to the red tape that will likely occur to implement technology? How long will other manufacturers technology implementation be delayed because of Uber not testing this scenario out in an Uber development facility?

I doubt any development facility can test every scenario that will occur in the wild. I would find it hard to believe that any of these companies is deliberately putting what they know to believe unsafe vehicles into situations "just to see what happens". Uber may be a bad company overall, and they may be out for the $ more than safety, but deaths will come with this type of technology. We have to be able to accept some in development that will improve the lives of all. The "disease" is already there. I'm willing to let good intentioned treatments be tried for the cure.