Uber driverless car in fatal accident

[Brother D]

Here is someone driving the same route with a camera actually relecting real world light levels vs some crappy dash cam released to the police

https://youtu.be/1XOVxSCG8u0

Frankly 99% of Americans put the pedestrian at fault, despite the fact if the person was looking at the road vs watching a film in their lap it would be a non event. But that is another topic, if an American runs a red light and kills a pedestrian, it's still the pedestrian's fault. (source me - having been knocked over by an American running a red light, and hit by cars three times in 5 years - they are that st at driving).

Still frankly shocked that the automated systems did not pick up a collision vector.

[Digger]

A reminder that these EV’s are almost silent and that in the UK I’m led to believe pedestrians have right of way once on the highway.

[RobDickinson]

Digger said:

A reminder that these EV’s are almost silent and that in the UK I’m led to believe pedestrians have right of way once on the highway.

A reminder the Volvo xc90 isn't an ev...

And EU rules, soon us rules will require low speed noise makers, most evs already have them.

And another reminder at this speed, 38mph, the Tyre noise is far louder than any standard engine noise making everything else irrelevant.

[RobDickinson]

I'm also pretty sure if you go wandering around on major roads in the UK you don't have 'right of way'.. Only if they are crossing at marked places..

[Digger]

Pedestrians should always have right of way regardless. It is but a simple fact.

[RobDickinson]

I don't think you mean what you think you mean.

Regardless, UK highway code disagrees.

[Digger]

What do you think I mean?

[RobDickinson]

Digger said:

Pedestrians should always have right of way regardless. It is but a simple fact.

This.

First what you are saying isnt something that can be a fact or not. its an opinion. One that is not supported by the highway code in the UK.

By 'always have right of way regardless' you mean pedestrians should be able to walk out anywhere at any time into the road and force cars to stop and allow them to cross the road. or sit down or whatever..

Thats plainly bks. And not what the highway code states. Pedestrians do have right of way at marked crossings etc, or of they have already started to cross the road before the car turns onto that road and specifically stated they dont at any other place.

Edited by RobDickinson on Friday 23 March 03:10

[JonChalk]

RobDickinson said:

I think we need to understand the video has been processed the codec quite poor and the dark shadow areas crushed.

I think the lady would have been visible a fair amount before you see her in the video.

I'm any similar real world situation I'm pretty sure I'd have spotted them in plenty of time and avoided them.

The lidar equipped car has no excuse.

I agree. Rob's a better photographer than me, but having played with drone, GoPro, dashcam and DSLR cameras, in low light, the human eye does a far better job of handling available light than almost anything you can buy as a consumer.

This footage would not be representative of what any one of us would have been able to see.

That should not be what the system is relying on however, for collision detection - those are independent of light levels (as they provide their own energy - light or radio waves, or both).

Digger, suggest you read:

https://www.gov.uk/guidance/the-highway-code/rules...

before deciding if you want to continue this pointless derailing of this thread.

[Kawasicki]

JonChalk said:

RobDickinson said:

I think we need to understand the video has been processed the codec quite poor and the dark shadow areas crushed.

I think the lady would have been visible a fair amount before you see her in the video.

I'm any similar real world situation I'm pretty sure I'd have spotted them in plenty of time and avoided them.

The lidar equipped car has no excuse.

I agree. Rob's a better photographer than me, but having played with drone, GoPro, dashcam and DSLR cameras, in low light, the human eye does a far better job of handling available light than almost anything you can buy as a consumer.

This footage would not be representative of what any one of us would have been able to see.

That should not be what the system is relying on however, for collision detection - those are independent of light levels (as they provide their own energy - light or radio waves, or both).

Digger, suggest you read:

https://www.gov.uk/guidance/the-highway-code/rules...

before deciding if you want to continue this pointless derailing of this thread.

It’s not just the human eye that is excellent, it is the fact that visual processing/pattern recognition is at the very heart of our evolution and intelligence. Replicating the overall human visual capability is an enormous task.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC41416...

[98elise]

Kawasicki said:

havoc said:

The Selfish Gene said:

akirk said:

We need to stop this apologist explanation saying that the human driver could have done very little about it... they absolutely could have avoided it...
Someone died - it is a pretty major negative outcome - there is lots that could have been done to prevent it...
- the driver is still meant to be ultimately responsible - they should have been paying attention
- the driver should have picked up that the lights were not showing enough road for the speed, and taken back control
- the driver should then have braked and avoided killing the pedestrian

instead, the driver let the car make all the choices, including bad choices...
they should be preventing the car from entering a situation where they no longer have time to react... otherwise their role is invalid / pointless

i.e. the driver should have avoided the situation starting - which would have avoided the outcome
we should not be accepting media and others saying - they could do nothing about it because they didn't have time to react - it was their job to only drive in such a way that they would have time to react!

bingo

Blind trust in unproven technology isn't enough. Not by a long shot.

In addition:-
- The car SHOULD have had something like LIDAR/RADAR to pick this up, or even some sort of low-light vision system. BLIND reliance on visible light is piss-poor quite frankly. Be very interested to know whether any of the above was fitted, and why it failed.
- There SHOULD (probably won't be / can't be without true AI) some way of picking up on the subtle clues that an attentive driver would use to highlight potential hazards before they 'jump out' (literally), e.g.
- Signs of kids playing nearby. Toys on the ground / playground across the road / even feet visible under cars
- Indications of animals nearby - parks, unfenced farmland, etc.
- Indications of side-roads joining ahead, e.g. hedgerow/treeling convergence, or even driveways/farm tracks which wouldn't be on any inbuilt map
- Seeing a pub ahead when you're driving around kicking-out time
etc. etc...


Human beings are capable of reacting to all manner of subtle / subconscious inputs that a computer most probably cannot. Not saying every driver does, but even a modestly-competent driver would pick up on environmental cues about the appropriate speed.

Computers are dumb. Either you use amazing sensors and data sharing between vehicles to get around that or you build a computer with as much sense and perception as the driver you intend to replace.

Amazing sensors and data sharing has potential, amazing AI is a long way off. By the time AI is good enough to replace a moderately competent driver we will have much bigger things to enjoy/fear than self driving cars!

Cars don't need AI. It's a rules based environment. People cannot process unexpected things very well, infact when presented with a sudden unexpected event we tend to freeze or panic.

Driving a car can be broken down into a set of rules, and when something is experienced outside of the rules then stop. A computer can do this without getting tired, enraged, looking at Facebook, doing it's makeup etc. It can also have eyes in the back of its head, or eyes that can detect objects outside our range.

[98elise]

rxe said:

It does illustrate an interesting problem.

A lot of the US commentary focuses on the fact that the pedestrian wasn't on a crossing. Uber's defence will probably settle on that, and get away with it. So we will have driverless car technology developed in the US that is completely unsuitable for the UK. If you restrict peds to crossings, then driverless is much easier from a software POV.

I also wonder if elements of the tech simply weren't working that night. Given the miles these things have covered, you'd have thought it would have happened before now if the tech really was that rubbish. This is another issue - if the lidar (or any other of the hundreds of sensors) fails, then the car has to stop. Perhaps this one was still driving but with a bad sensor.

Edited by rxe on Thursday 22 March 20:04

I doubt the software will ever rely on mapping points where people will be on the road. You have to account for people running out into the road, and the car needs to see and recognise what they are. You have to program for the worst case scenario.

[captainaverage]

Digger said:

Pedestrians should always have right of way regardless. It is but a simple fact.

Yes even on motorways. Don't let these fools tell you otherwise. You can walk across the motorway or other high speed A roads and they WILL/MUST give you way. I always give way to the pedestrians on the motorway. Next time you feel like it just do it.

Edited by captainaverage on Friday 23 March 07:38

[Psimpson7]

98elise said:

It can also have eyes in the back of its head, or eyes that can detect objects outside our range.

But in this case seemingly not have eyes in the front of its head....

[Kawasicki]

98elise said:

Cars don't need AI. It's a rules based environment. People cannot process unexpected things very well, infact when presented with a sudden unexpected event we tend to freeze or panic.

Driving a car can be broken down into a set of rules, and when something is experienced outside of the rules then stop. A computer can do this without getting tired, enraged, looking at Facebook, doing it's makeup etc. It can also have eyes in the back of its head, or eyes that can detect objects outside our range.

Yes humans are not very good at processing unexpected things, but computers can't process them at all.

I think the set of rules required to drive as well as a competent driver is pretty large, and the exceptions to these rules are numerous. To avoid stopping more than a typical passenger would find acceptable the system needs to be pretty intelligent or have access to extensive local spatial data.

[DonkeyApple]

Fastpedeller said:

I find it difficult to understand how the Chief of Police has stated the car wasn't to blame - How much is he being paid by Volvo or Uber? Is it reasonable that he even makes such a statement? Victim blaming is easy when the victim is deceased. RIP.

A sherif is an elected position but chief is chosen by the mayor’s office. It’s a position that supports the mayor. In Tempe the mayor dropped regulation to bring Uber in. Given how Uber operate all around the world then I would near guarantee that there have been speedboat tokens flying around Tempe and in recent days various conversations reminding people about the tokens and what they are for.

And this is going to be one of the long term issues with autonomous cars. The companies working on them have very well established and deep set corporate dishonesty values from the bottom to the Board.

[untakenname]

Uber should be ordered to provide the raw footage as the footage they have produced looks like its from a cheap dashcam circa '14.
Have been told that uber may have unwittingly employed a felon as the driver changed sexes and names before being employed, think there maybe some truth to this as its certainly a very ugly female.

[DonkeyApple]

RobDickinson said:

I don't think you mean what you think you mean.

Regardless, UK highway code disagrees.

But you know what he means and highlighting the very big cultural difference between pedestrians on the road in the US versus Western Europe is a very valid point.

From my own personal experience of driving around New York, Miami, Denver and a few other US cities, pedestrians are remarkably compliant and have jaywalking laws to help. A truly extreme example being Vancouver where other pedestrians actually try to stop you crossing anywhere other than at designated places (very weird city but a great window into how all society will be probably be when the robots take over. ) Conversely, drive in London, Paris and many other cosmopolitan European cities and you notice a very start difference. There are side roads that you cannot turn into due to endless streams of pedestrians crossing unless you drive at them to help manage their descision making. There are junctions controlled by traffic lights where the pedestrians don’t stop for the lights but only when the traffic moves directly at them.

These are big cultural differences and it will be interesting to see how autonomous cars are programmed to navigate these types of junctions and turns in environments where pedestrians have human rights.

[rxe]

98elise said:

Cars don't need AI. It's a rules based environment. People cannot process unexpected things very well, infact when presented with a sudden unexpected event we tend to freeze or panic.

Driving a car can be broken down into a set of rules, and when something is experienced outside of the rules then stop. A computer can do this without getting tired, enraged, looking at Facebook, doing it's makeup etc. It can also have eyes in the back of its head, or eyes that can detect objects outside our range.

I think they do need AI, and that’s why it is so hard.

A human can drive a car safely with two (or sometimes one) biological video cameras and a brain. No need for Lidar, all round cameras, ultrasonic sensors and whatnot. The biological video cameras aren’t really very good in terms of accuracy, but they have high dynamic range and and brain has a truly extraordinary capability to interpolate missing data. The brain is also very good at working out strategies to keep the driver safe. Occasionally the brain gets tired, pissed or angry and then bad things happen.

Treating a car as a rules engine is going to lead to failure because it is impossible to codify every single driving rule in a list - the list will never, ever, be complete.

Let’s take an off the wall scenario. You’re driving along, and 400 yards away you see a helicopter descending rapidly with smoke pouring from its engine. Without any training at all, the human will say “that looks bad, it’s going to land on the road, I’ll stop”. The rules engine will say “road is clear, I’m within the white lines, proceed at the speed limit”. That’s an extreme example, but every time you get in a car, less extreme examples happen all the time. Cyclists do stupid things, you know there is a shop that school kids pile out of, there are objects in the road. We deal with it easily, rules engines don’t.

Rules engines are also really, really bad at handling conflicting information.

IMO the “loads of sensors + rules engine” is an evolutionary dead end for driverless cars. It will get us a long way, but it won’t be good enough.

[DonkeyApple]

rxe said:

98elise said:

Cars don't need AI. It's a rules based environment. People cannot process unexpected things very well, infact when presented with a sudden unexpected event we tend to freeze or panic.

Driving a car can be broken down into a set of rules, and when something is experienced outside of the rules then stop. A computer can do this without getting tired, enraged, looking at Facebook, doing it's makeup etc. It can also have eyes in the back of its head, or eyes that can detect objects outside our range.

I think they do need AI, and that’s why it is so hard.

A human can drive a car safely with two (or sometimes one) biological video cameras and a brain. No need for Lidar, all round cameras, ultrasonic sensors and whatnot. The biological video cameras aren’t really very good in terms of accuracy, but they have high dynamic range and and brain has a truly extraordinary capability to interpolate missing data. The brain is also very good at working out strategies to keep the driver safe. Occasionally the brain gets tired, pissed or angry and then bad things happen.

Treating a car as a rules engine is going to lead to failure because it is impossible to codify every single driving rule in a list - the list will never, ever, be complete.

Let’s take an off the wall scenario. You’re driving along, and 400 yards away you see a helicopter descending rapidly with smoke pouring from its engine. Without any training at all, the human will say “that looks bad, it’s going to land on the road, I’ll stop”. The rules engine will say “road is clear, I’m within the white lines, proceed at the speed limit”. That’s an extreme example, but every time you get in a car, less extreme examples happen all the time. Cyclists do stupid things, you know there is a shop that school kids pile out of, there are objects in the road. We deal with it easily, rules engines don’t.

Rules engines are also really, really bad at handling conflicting information.

IMO the “loads of sensors + rules engine” is an evolutionary dead end for driverless cars. It will get us a long way, but it won’t be good enough.

I think you’re right with regards to full blown autonomy where your car looks after itself, taking itself off to be cleaned, fuelled, parked and where it will take you wherever to demand it to and in a quicker and safer manner than 99% of humans can manage.

But, I think there are simpler scenarios where a rule based, non AI system will do a perfectly good, if not better job. I am thinking of automated buses and some trucks in city centres. A bus route is typically the easiest of routes in a city centre as it isn’t trying to navigate the more complex, smaller roads. Plus, you see far fewer pedestrians chancing it against a massive red bus down Oxford St than you do against minicabs so there seems to be a better ability to control pedestrians. Loading one up with sensors probably makes such a vehicle safer than relying purely on the driver. It wouldn’t surprise me if these sensors found their way more and more into buses anyway and we started seeing the driver only being able to move away if the sensors all tell him he can. Similar with some other vehicles such as spoil trucks in city centres. They always appear to be driven by half blind maniacs with little regard for law or etiquette. It doesn’t surprise me that they have such a high track record for killing cyclists and pedestrians. I can see those being safer automated on a rule based system.

Uber actually has the advantage here over other companies in that it can role out its product city by city unlike PLG vendors. PLG vendors need their products to work in all cities in a country but Uber just needs some. All cities vary in the complexity of their road networks, some will be easy to use rules based autonomous vehicles in and others it will never happen.

So Uber can role out a product in one city and keep using human slaves in another. The nature of minicabs is that they spend their life in the same city. PLGs on the other hand don’t. How does VW sell a car that only works in one city on a continent to private individuals? It can’t. It’s a corporate market in reality such as local hire cars and minicabs.

Autonomous cars aren’t going to be allowed to drive everywhere. They will only work in towns and cities that approve them. So your private vehicle will need an automation program for each of those towns or cities and anywhere in between it will be a conventional car. That’s not a problem, just a cost and a loss of convenience compared to commercial transport that is geographically defined such as minicabs.