First fatality involving a self-driving car

[BernieCarbo]

I learned to drive defensively. I don't do things that can distract from my being able to expect the unexpected. There is no way a computer could ever drive 1/10th as safely as I could, and I'm not buying one bit that this death was unavoidable.

I think most people drive defensively, and even I have driven 51 years without an accident. But there is no way a human being can drive as safely as an automated vehicle, not to mention that the technology is still in its infancy. But in this case, I would have expected more from the technology even though no human could have avoided this either.

A good comparison is to just look at the several deadly train accidents over the last few years, In each case, a human engineer was at the controls instead of using an auto guided system, and they became distracted and lost control.

[kaldaniels]

Computers are pretty well bound to their algorithms.

I wonder if that Uber computer hits that bicyclist 100 out of 100 times? I would think so.

Does a human hit the bicyclist 100 out of 100 times?

I do imagine this is a one-off type of event, where computers drive safer overall, but they need their "blind spots" to be fixed, no pun intended.

[WrongVerb]

Computers are pretty well bound to their algorithms.

While this is true of most programming these days, one feature of AI is that it is distinctly not, in fact, bound to its algorithms. It essentially develops its own algorithms based on its learning. The learning algorithms themselves, can be parameterized, but I think even that is becoming an out-moded method of programming AI, in favor of things like neural nets. That said, do we know whether AI was being used in this self-driving car?

I wonder if that Uber computer hits that bicyclist 100 out of 100 times? I would think so.

Does a human hit the bicyclist 100 out of 100 times?

I do imagine this is a one-off type of event, where computers drive safer overall, but they need their "blind spots" to be fixed, no pun intended.

To me, one thing I am not sure is programmed into self-driving cars (and I think Bernie Carbo alluded to this) is the idea of defensive anticipation. At any moment while driving, you should be anticipating the worst things that can happen and accounting for how you'll manage those should they happen. That includes someone darting into traffic from the shadows and from between parked cars. I doubt self-driving car programming is at that level, though I don't know for sure. And it should be the goal of programmers in the future to get the AI to handle that.

[kaldaniels]

While this is true of most programming these days, one feature of AI is that it is distinctly not, in fact, bound to its algorithms. It essentially develops its own algorithms based on its learning. The learning algorithms themselves, can be parameterized, but I think even that is becoming an out-moded method of programming AI, in favor of things like neural nets. That said, do we know whether AI was being used in this self-driving car?



To me, one thing I am not sure is programmed into self-driving cars (and I think Bernie Carbo alluded to this) is the idea of defensive anticipation. At any moment while driving, you should be anticipating the worst things that can happen and accounting for how you'll manage those should they happen. That includes someone darting into traffic from the shadows and from between parked cars. I doubt self-driving car programming is at that level, though I don't know for sure. And it should be the goal of programmers in the future to get the AI to handle that.

I don’t know anything about whether AI was used. But if that scenario happened exactly the same way 100 times, doesn’t the computer driver hit the bicyclist 100 times (that computer at this point in time).

[WrongVerb]

I don’t know anything about whether AI was used. But if that scenario happened exactly the same way 100 times, doesn’t the computer driver hit the bicyclist 100 times (that computer at this point in time).

As I understand it, an AI system would learn not to do that.

[BernieCarbo]

AI is more of a buzzword that journalists like to throw around than something that will actually exist in the decades ahead. We have been using self-learning algorithms for years, and this is often confused with AU, but even then someone has to give it strict rules to follow. And, the rules are extremely dependent on data, and data acquisition is still the weak link.

For instance, when I grew up, I would spend a few weeks raking blueberries, and then a week raking cranberries. To enable a machine to do the same thing, it would have to be fed an enormous number of parameters. We are extremely far away from teaching a system how to rake blueberries, and then being able to push it in a swamp and expecting it to adapt for cranberries, even though a ten year old kid could do it.

In this case, assuming the reports are correct and 1) it was dark, and 2) it was a multi-lane road, a human would have hit the woman 100 out of 100 times, but I would expect a computer to do better because it can see in the dark and can tell the difference between an automobile and another object that shouldn’t be there. I still stand by my assertion that something in the code was overlooked, but a human couldn’t do better.

[Kingspoint]

While this is true of most programming these days, one feature of AI is that it is distinctly not, in fact, bound to its algorithms. It essentially develops its own algorithms based on its learning. The learning algorithms themselves, can be parameterized, but I think even that is becoming an out-moded method of programming AI, in favor of things like neural nets. That said, do we know whether AI was being used in this self-driving car?

For 15 years, Subarus have used "learning" AI to memorize the habits of how it "the car" is driven. It then makes adjustments over time to how it takes care of itself mechanically. If the vehicle changes drivers, this throws it off, of course, as it assumes it's all the same driver.

Never put chains on a Subaru, btw, as the chains destroy the computer's responses to it's AWD system. If conditions are so bad, you have to put chains on a Subaru (ice, for example), nobody should be driving. The fake AWD systems (don't engage into AWD until trouble has begun with varying degrees of power shifting to the rear wheels, normally only about 20-25%) found in Hondas, Toyotas, Nissans, GMC, Ford, Chevy, and everyone else not named Audi require chains, where Subaru and Audi are exempt in Oregon and most States, and any other vehicle that is AWD all the time, weighs 2500+ pounds and has M&S tires on them (sorry WRX owners).

So, yes, AI computers have learning abilities, but the example given of the bicycle probably isn't one of them. It could be, but what a way to learn!

I see empty spaces that could potentially be hiding anything and I expect that there might be something that in the last second can jump out at me, and adjust my driving accordingly.The self-driving cars need to do this. They are already handicapped because they aren't Subarus or Audis.

I don't play music while driving. I need to be able to hear every noise my car makes and every noise of cars near me. Hearing has saved my life too many times to count from idiot drivers (most of them, and they keep getting worse because cars keep getting better at correcting their mistakes to where tail-gating is the norm thanks to incredible tires and incredible braking systems) to the vehicle I'm driving having an issue I've pulled over to discover.

Obama's trade-in-your-old-car program got rid of a lot of bad cars that were death-on-wheels, and that helped. Today's cars are practically self-driving as it is because less than 10% of drivers actually drive safely. The vehicle is taking care of all the safety violations of the driver and correcting them. They are oblivious to it and don't even know it, as they think they are driving safely. They are not. The vehicle is constantly fixing their mistakes.

My father-in-law just wrecked his brand new Outback and it wouldn't have happened if he'd had a self-driving car. It also wouldn't have happened had he got the model with the auto-brake safety system. He hit a UPS truck at 5 mph, but because the Subaru is designed to spread the impact around to where the passengers are the safest, the airbag, safely, did not deploy, $6K in sensors brought the repair to $9,500. The UPS truck was fine.

[Kingspoint]

There's no way I would have hit that woman because I never would have had my vehicle in that position in the first place. I also would have been driving a lot safer vehicle, one that responds to emergencies a 1000 times better.

- - - Updated - - -

TEMPE, Ariz./PITTSBURGH (Reuters) - When Uber decided in 2016 to retire its fleet of self-driving Ford Fusion cars in favour of Volvo sport utility vehicles, it also chose to scale back on one notable piece of technology: the safety sensors used to detect objects in the road.


That decision resulted in a self-driving vehicle with more blind spots than its own earlier generation of autonomous cars, as well as those of its rivals, according to interviews with five former employees and four industry experts who spoke for the first time about Uber’s technology switch.

Driverless cars are supposed to avoid accidents with lidar – which uses laser light pulses to detect hazards on the road - and other sensors such as radar and cameras. The new Uber driverless vehicle is armed with only one roof-mounted lidar sensor compared with seven lidar units on the older Ford Fusion models Uber employed, according to diagrams prepared by Uber.

In scaling back to a single lidar on the Volvo, Uber introduced a blind zone around the perimeter of the SUV that cannot fully detect pedestrians, according to interviews with former employees and Raj Rajkumar, the head of Carnegie Mellon University’s transportation centre who has been working on self-driving technology for over a decade.

The lidar system made by Velodyne - one of the top suppliers of sensors for self-driving vehicles - sees objects in a 360-degree circle around the car, but has a narrow vertical range that prevents it from detecting obstacles low to the ground, according to information on Velodyne’s website as well as former employees who operated the Uber SUVs.

Autonomous vehicles operated by rivals Waymo, Alphabet Inc’s self-driving vehicle unit, have six lidar sensors, while General Motors Co’s vehicle contains five, according to information from the companies.

Uber declined to comment on its decision to reduce its lidar count. In a statement late Tuesday, an Uber spokeswoman said, “We believe that technology has the power to make transportation safer than ever before and recognise our responsibility to contribute to safety in our communities. As we develop self-driving technology, safety is our primary concern every step of the way.”

Uber referred questions on the blind spot to Velodyne. Velodyne acknowledged that with the rooftop lidar there is a roughly three meter blind spot around a vehicle, saying that more sensors are necessary.

https://uk.reuters.com/article/uk-ub...-idUKKBN1H337L

[Kingspoint]

Volvo is also the least "responsive" vehicle manufacturer in the World. I've test-driven every Manufacturer's "AWD" system on the St. Louis Speedway in dry conditions, wet conditions, through cone-courses, braking under various conditions, and off-track in dirt and steep-angled climbing conditions to test power distribution. Also, on special equipment to place each wheel under independent surfaces to see power distribution.

In a class by themselves, Volvo makes the most dangerous vehicles to others because of their poor responsiveness. Sorry, Volvo owners, but your vehicle sucks. It's built like a tank and you won't get hurt. You'll get in more accidents, though, and hurt more people and their property. You might get 300K miles on it because it's built like a tank, but thank goodness you have good insurance because you'll be paying a lot of claims. Volvo is bad in a class by themselves.

[Kingspoint]

So, someone said this Self-driving issue is at it's infancy and simply because of time alone, it should get better.

Well, this article proves that if you give a company such as Uber more time, they willmake teir vehicles less safe out of pure greediness for the buck.

They were already using one of the most dangerous cars in the World to a driver in the Ford Fusion, so they switch to the most dangerous car in the World to everyone and everything in the path of the Volvo to give the impression that the passenger will be safer in the Panzer of the automobile world, and then they remove for cost reasons because one Volvo is the same cost as 3 Ford Fusions, safety sensors at $2000-$4000 each that makes the poorly responsive Volvo's even less responsive.

The Volvo will save them money in lawsuits as it's passengers won't have as many fatalities, and the increased accidents from the Volvo will save them money because everyone on the jury will find it easy to say it was unavoidable, just like everyone is doing here.

Hogwash!

That lady should be alive. Her death was 100% avoidable. They shouldn't be using Volvos (or Fusions) in the first place. More sensors should have been added instead of having them removed. The vehicle should have been driving slower under difficult conditions. More could have been done to prevent this murder.

[Kingspoint]

There's so much more that can be done here. The manufacturers and operatorsof these vehicles don't care one iota about safety. They only care about lawsuits and the steps that need to be taken to lessen their total dollar amount. You don't use a Ford Fusion or Volvo. That should be Step A. You use the fastest reactive vehicle in tge World, the Audi, but a Subaru costs half as much and would suffice.

The article above mentions the Velodyne sensor that sees 360 degrees. But, that by itself, means nothing. Subaru, for example has it's breaking system communicate with all four wheels simultaneously. Most systems have to go past one wheel to get to another, which takes more time. These nano-seconds all add up when brakes are working with the gas pedal, all four tires, computers, transmission, and other parts all the time the wheels and axles are spinning.

I know for a fact that anyone who would use a Volvo or Ford Fusion in the first place isn't concerned with how long it takes for the vehicle to identify a situation and respond to it. They are not thinking safety first, but costs first. A self-driving car should be thinking safety first.

They should not be allowed on streets for testing. That is the most obvious thing that shoukd have been done to prevent this murder of an innocent person, and the next murder that will surely occur.

Her death was 100% avoidable.

Uber should be sued and prevented from getting into the self-driving industry for 20 years.

[Redsfaithful]

I'm also surprised at the brand of cars.

Google has used Lexus and Apple's efforts do too I think. I think Subaru would also be good. I'm biased I guess since these are cars I have owned.

I have an Outback now that has the adaptive cruise control, self braking, eye tracking, etc. and it's amazing. It's by far the safest feeling car I have ever been in.

[Kingspoint]

I'm also surprised at the brand of cars.

Google has used Lexus and Apple's efforts do too I think. I think Subaru would also be good. I'm biased I guess since these are cars I have owned.

I have an Outback now that has the adaptive cruise control, self braking, eye tracking, etc. and it's amazing. It's by far the safest feeling car I have ever been in.

Lexus would be a huge upgrade over the other two choices Uber made.

[BernieCarbo]

The Fusion and Volvo are not dangerous cars. That’s just hyperbole. And Subaru not using AI, unless you stretch the definition to cover every piece of software ever written for a computer. What you describe is called adaptive learning, and is has been used extensively throughout automated systems for 30 years. Without it, automation will have very limited applications.

We don’t know enough about the situation to know if a human could have avoided that woman. Based on the police report and the video, it looks impossible. We don’t know how fast she was walking, or if she was just standing there and then started walking, or if she had been riding her bicycle and then got off. Pedestrians are hit on a regular basis, and it’s usually because they were where they were not supposed to be. No one can say this was 100% avoidable.

[kaldaniels]

“Murder” is a tough sell for me on this one.