Autonomous Uber kill vehicle (Dash Cam)

[DashcamDPR]

As long as there are autonomous cars....

...cars with human drivers should all have dashcams!

 

[kamkar]

Well we should strive for zero deaths, no doubt about that.
But history also dictate that it will probably be pretty naive, and with a major thing like traffic which are all around us as soon as we step out the door.
Then i am sure a lot will die in situations where that should not really happen, just like it do now in regular human operated traffic, it just take a split second of lost situational awareness and something go wrong.

Only "good" thing about it is that there are plenty of us humans to go around, i know its a stone cold thing to say, and what if it was one i cared for that died ?
And i can not rule out if some one by way if neglect or plain stupid killed one i care for, and the judicial system not doing anything near what i think should be done, well then i might do something about it myself.
At my age i no longer have the same love / need for life, and a lengthy jail sentence would mean death to me.
And i would not mind dying for a person or cause i love / respect, this little thing is what take care of the manpower needs in our wars.

 

[M---]

Somebody on Reddit took a crappy hand-held cellphone video while driving past the crash site. Somebody else took a screenshot and compared it to a screenshot of the Uber video... The lighting situation is markedly different.
Hand-held cellphone vs Uber screenshot

Does anybody here live in Austin? I'm super-curious to see what the street looks like with a proper dashcam.

 

[dash riposki]

The infrastructure is too far behind for this stuff to work. We're talking decades away, and what eventually happens is probably nothing we can picture, now.

I think the logical progression ends up with personal autos being a thing of the past for most people. The cargo transport stuff doesn't have to move at 100 kph to be efficient, just 24 hours a day, safely. (Which brings us to the question why the hell aren't there automated long distance cargo trains?)

 

[Matt_V939]

Check this reconsideration about the uber crash
https://www.theguardian.com/technol...-uber-death-woman-failure-fatal-crash-arizona

"..The footage “strongly suggests a failure by Uber’s automated driving system and a lack of due care by Uber’s driver”, Bryant Walker Smith, a University of South Carolina law school professor and autonomous vehicle expert, said in an email. He noted that the victim is visible about two seconds before the collision, saying: “This is similar to the average reaction time for a driver. That means an alert driver may have at least attempted to swerve or brake.”....

“I really don’t understand why Lidar didn’t pick this up,” said Ryan Calo, a University of Washington law professor and self-driving expert. “This video does not absolve Uber.”

Even though the video appeared dark, King said there was likely more visibility than the footage suggested and noted that the darkness should not affect the car’s detection abilities.
“Shadows don’t matter to Lidar,” added Cummings. “There is no question it should have been able to see her.”

 

[Nigel]

Check this reconsideration about the uber crash
https://www.theguardian.com/technol...-uber-death-woman-failure-fatal-crash-arizona

"..The footage “strongly suggests a failure by Uber’s automated driving system and a lack of due care by Uber’s driver”, Bryant Walker Smith, a University of South Carolina law school professor and autonomous vehicle expert, said in an email. He noted that the victim is visible about two seconds before the collision, saying: “This is similar to the average reaction time for a driver. That means an alert driver may have at least attempted to swerve or brake.”....

“I really don’t understand why Lidar didn’t pick this up,” said Ryan Calo, a University of Washington law professor and self-driving expert. “This video does not absolve Uber.”

Even though the video appeared dark, King said there was likely more visibility than the footage suggested and noted that the darkness should not affect the car’s detection abilities.
“Shadows don’t matter to Lidar,” added Cummings. “There is no question it should have been able to see her.”

Agree with most of that.

"and King noted that the exact section where Herzberg entered the street is a common area for pedestrians to cross near a local park."

Before autonomous vehicles are let loose from their testing areas, they are going to need these knowledge bases that humans have about where accidents are likely to happen and lower speeds should be used rather than just traveling at the speed limit. Most human drivers would have known or realised extra care was desirable at that location, an autonomous vehicle needs to be told. Easy to create the knowledge base, just collect the accident statistics, just part of the infrastructure that [U]dash riposki[/U] mentioned.

"“I really don’t understand why Lidar didn’t pick this up,” said Ryan Calo, a University of Washington law professor and self-driving expert."

Probably nothing to do with the Lidar, Uber are one of the companies trying to solve the problem with artificial intelligence instead of mathematics. Their systems are trained to recognise danger instead of logically calculating risk. It's a shortcut where nobody actually needs to understand how the system works, they just train it until they get the right result in the testing. The result is a system that can't be proved safe and may not get things right in circumstances it hasn't been trained for. Artificial intelligence is always going to be risky and shouldn't be used in a safety critical situation like this.

 

[Gibson99]

Probably nothing to do with the Lidar, Uber are one of the companies trying to solve the problem with artificial intelligence instead of mathematics. Their systems are trained to recognise danger instead of logically calculating risk. It's a shortcut where nobody actually needs to understand how the system works, they just train it until they get the right result in the testing. The result is a system that can't be proved safe and may not get things right in circumstances it hasn't been trained for. Artificial intelligence is always going to be risky and shouldn't be used in a safety critical situation like this.

thing is though, basic collision detection is most likely at the foundation of all AI cars. even if the car's headlights were off, LIDAR should have seen the cyclist unless it wasn't working, was disabled, or maybe it had dead bugs on it or something... but assuming lidar WAS broken, what kind of backup system is there? if the optical camera is anything like the apparent $20 taxi cam they used as a dashcam to show inside and out, then that explains a lot. you can't skimp on something so critical to safety, even if it isn't the primary sensing system.

I like @M---'s post above showing the difference the camera can make. Austin isn't far for me (not that i'd go there just for this - i might if i was already in the austin area - it's about 2 hours from me), but the crash happened in Tempe, Arizona.

 

[DT MI]

...Maybe it was one of these silent electric vehicles and the cyclist didn't hear it coming?...

I read somewhere it was a Volvo XC90.

 

[M---]

I like @M---'s post above showing the difference the camera can make. Austin isn't far for me (not that i'd go there just for this - i might if i was already in the austin area - it's about 2 hours from me), but the crash happened in Tempe, Arizona.

Whoops, my memory was out! Tempe AZ vs. Austin TX... They're 1000 miles apart from each other! How on earth did I get them mixed up?

But yeah, if the LIDAR couldn't recognize that pedestrian as a hazard, then I've got to wonder what other shortcuts they've taken (hardware and software) to come up with their self-driving cars. Driving safely (defensively) involves a lot more than just driving in accordance with the right of way!

 

[Nigel]

I read somewhere it was a Volvo XC90.

Well it wouldn't be the diesel version in USA so probably the plug in hybrid - may well have been running silently on electric power.

Mind you, 401 bhp is a lot of power to put in the hands of a computer that is still being tested

 

[Nigel]

thing is though, basic collision detection is most likely at the foundation of all AI cars. even if the car's headlights were off, LIDAR should have seen the cyclist unless it wasn't working, was disabled, or maybe it had dead bugs on it or something... but assuming lidar WAS broken, what kind of backup system is there? if the optical camera is anything like the apparent $20 taxi cam they used as a dashcam to show inside and out, then that explains a lot. you can't skimp on something so critical to safety, even if it isn't the primary sensing system.

Part of the problem may have been that the lidar, and radar if they have it, couldn't see the bike approaching, or receding, because it wasn't, so to predict the collision required recognising the moving object and its shape and tracking its path across the field of view which is a lot more complex than just detecting an approaching vehicle where the pixels on the lidar will show steadily decreasing values. Should have seen the bike as an obstacle once it was directly in front though, both on lidar and 3D visual cameras.

The Google cars just use visual cameras, they reckon that the roads are designed for human eyes which are visual so stereo cameras are enough. I've always wondered what happens when you get multiple cars with lidar / radar sensors, do they interfere with each other and cause crashes?

It is a bit surprising that they don't check their in car videos regularly and sack any drivers not paying constant attention, presumably they consider that his behaviour is acceptable?

 

[DT MI]

...so probably the plug in hybrid - may well have been running silently on electric power.
...

Possibly, but given the hybrid version only has a range of 14 miles in 'electric only' mode it's likely very unlikely that was the case.

 

[jokiin]

Considering Volvo don't sell any autonomous vehicles yet it's a bit presumptuous to assume the rest of the vehicle is as per a production model, maybe better to wait until they release more details, it's all internet guess work based on limited information at this stage

 

[M---]

Vehicle noise: at parking lot speeds, EVs and hybrids are a real problem for pedestrians. My car's a hybrid, and I can silently cruise through parking lots, which does startle pedestrians (or more often, they startle me by stepping out when they really shouldn't). Cruising at ordinary road speeds, tire noise is the dominant sound from my car, as it is for almost all cars.

The pedestrian in this crash should've seen the headlights, or heard the tire noise.

 

[sludgeguts]

This couldn't possibly be the pedestrian's fault though?
I didn't get to this ripe old age by simply wandering across roads in front of moving traffic

 

[Nigel]

This couldn't possibly be the pedestrian's fault though?
I didn't get to this ripe old age by simply wandering across roads in front of moving traffic

Clearly the pedestrian has some responsibility, but it is still not acceptable for the car to kill the pedestrian, it should have attempted to avoid the accident, or at least slowed down to avoid the death. The pedestrian was occupying that bit of road first, the front of the car hit the pedestrian rather than the front of the pedestrian hitting the car.

 

[flank]

Volvos are suppose to have some crash avoidance built in. Some even have pedestrian airbags. But its pretty useless if the car is going to just mow them over.

 

[jokiin]

Volvos are suppose to have some crash avoidance built in. Some even have pedestrian airbags. But its pretty useless if the car is going to just mow them over.

the car they used in that didn't actually have the auto braking function

 

[dash riposki]

Name sound familiar?
Another 'expert'..................

https://www.popsci.com/whos-responsible-when-self-driving-car-crashes

 

[old4570]

0 Deaths = 0 Stupid people ....
Not sure how Ur going to achieve that ? > Breeding Programs ? - Culling programs ? - Dont see it ...
Education Programs ? ( Probably the best last great hope > but what gooberment is going to upskill bad / dangerous drivers ) = Seems to be a serious reluctance to actually doing something more than paying the problem LIP service !