Autonomous cars? I think not

User avatar
bychosis
Posts: 7250
Joined: Sat Jan 14, 2012 1:10 pm
Location: Lake Macquarie

Re: Autonomous cars? I think not

Postby bychosis » Thu Mar 22, 2018 9:59 am

mikesbytes wrote:While not specifically commenting on this incident, I can see another "not my fault" excuse coming to the fray;
1. The Autonomous mode drives conservatively complying with all traffic laws and applying safe practices
2. The occupant wants to get there quicker, not leave a gap to the car in front to prevent others cutting in etc etc
3. To achieve 2. the occupant drives manually
4. An incident occurs
5. The occupant says the vehicle was in autonomous mode
4.5. Hacks the data records in the vehicle to show manual mode not engaged.
bychosis (bahy-koh-sis): A mental disorder of delusions indicating impaired contact with a reality of no bicycles.

Jmuzz
Posts: 631
Joined: Mon Jan 29, 2018 1:42 pm

Re: Autonomous cars? I think not

Postby Jmuzz » Thu Mar 22, 2018 10:38 am

mikesbytes wrote: 5. The occupant says the vehicle was in autonomous mode
The cars will be black boxed up the wazoo, actively designed to protect the company not the owner, so an end user will have no hope of getting away with that.
Smart chips can be secure, Sydney's Opal system is proof of that nobody has cracked it's encrypted section yet or even has full details of what is happening in there.

The company on the other hand can hack their own data and that's a concern.

One of the reasons we are seeing a delay in the data from this crash being made public is that it is not automatically trustworthy.
Ok logs and video exist, but the car was 4G internet connected and had hours to be tampered with.
Not a conspiracy, just sensible police/court process not to contaminate the jury pool with potentially fake data, police have to do their job of checking it is authentic before allowing it to set public opinion.

Corporate hacking in their favour is actually a danger users will have to manage. It may be the case that users will need their own third party dashcam and data sniffing as protection against manufacturer putting the logs into Manual Mode post accident.

User avatar
find_bruce
Moderator
Moderator
Posts: 10579
Joined: Mon May 09, 2011 8:42 pm
Location: Sydney

Re: Autonomous cars? I think not

Postby find_bruce » Thu Mar 22, 2018 10:49 am

road_warrior wrote: You cannot say that based on one incident. Unfortunately you need much more data to determine this empirically.
Howzat wrote:Agreed - but obtaining a more statistically confident sample size of dead pedestrians per million vehicle-miles is not really on the cards. This is the first empirical data we have - and we have to pay attention to it. It's a bit of a fell-at-the-first-hurdle situation. It contradicts the utopian press-release vision of autonomous cars being by definition "safer than human drivers", and it should also indicate that this problem space is harder to solve than your average popular-science prognosticator might think.
Yep. You can't put a drug to clinical trials until you can prove it's safety and efficacy. Good luck getting ethics approval where you say the trial should continue until there is a statistically significant number of deaths.
Howzat wrote:The letter to the editors pages are full of people asking why we're spending money trams and trains when self driving pod cars are just around the corner. Well, that ain't necessarily so.
Self driving trains are already here & auto pilot has been a feature of planes for decades.

Jmuzz
Posts: 631
Joined: Mon Jan 29, 2018 1:42 pm

Re: Autonomous cars? I think not

Postby Jmuzz » Thu Mar 22, 2018 10:50 am

Oh video now, looking very bad for Uber, that's a total failure of the basics.
(Warning may disturb. Also not sure where official release source is so fake is a possibility unless verified. Twitter account seems legit https://mobile.twitter.com/TempePolice? ... frame.html )

http://ktar.com/story/1997617/tempe-pol ... ving-uber/
Last edited by Jmuzz on Thu Mar 22, 2018 11:00 am, edited 1 time in total.

User avatar
bychosis
Posts: 7250
Joined: Sat Jan 14, 2012 1:10 pm
Location: Lake Macquarie

Re: Autonomous cars? I think not

Postby bychosis » Thu Mar 22, 2018 10:59 am

Jmuzz wrote:Oh video now, looking very bad for Uber, that's a total failure of the basics.
(Warning may disturb. Also not sure where official release source is so fake is a possibility unless verified)

http://ktar.com/story/1997617/tempe-pol ... ving-uber/
Assuming it is real. It appears the car has no infra red or heat sensing cameras, or any sort of lidar scanning. The pedestrian was already on the road, but in the dark. Heat sensing ir lidar should have picked up the person on the other lane with time to stop.
bychosis (bahy-koh-sis): A mental disorder of delusions indicating impaired contact with a reality of no bicycles.

Jmuzz
Posts: 631
Joined: Mon Jan 29, 2018 1:42 pm

Re: Autonomous cars? I think not

Postby Jmuzz » Thu Mar 22, 2018 11:07 am

Don't know if they would have good heat sensing. The cheap grade like flir-one don't have any distance capability, they would need to be the serious high end stuff.

There should be Near-IR illumination beyond the headlight (which have restricted side illumination by law) and far better view available.

Perhaps that is just the basic dashcam optical view and there is much better vision not extracted yet.

It should be far better.
Certainly inexcusable obstruction detection if that is the real situation and it missed that and got through closed course testing with impacts in that easy to detect situation.

Operator is also negligent based on that video.

AdelaidePeter
Posts: 1230
Joined: Wed Jun 07, 2017 11:13 am

Re: Autonomous cars? I think not

Postby AdelaidePeter » Thu Mar 22, 2018 11:38 am

bychosis wrote:
Jmuzz wrote:Oh video now, looking very bad for Uber, that's a total failure of the basics.
(Warning may disturb. Also not sure where official release source is so fake is a possibility unless verified)

http://ktar.com/story/1997617/tempe-pol ... ving-uber/
Assuming it is real. It appears the car has no infra red or heat sensing cameras, or any sort of lidar scanning. The pedestrian was already on the road, but in the dark. Heat sensing ir lidar should have picked up the person on the other lane with time to stop.
Holy cr*p.

OK, that's not "stepped suddenly in front of self-driving car" like the LA Times article I quoted yesterday ( http://www.latimes.com/business/la-fi-h ... story.html ). Yes the pedestrian/cyclist clearly wasn't careful enough, and probably a human driver would still have hit her. But she comes into view about a second* before impact, and any alert human driver would have at least braked and slowed down.

And if the self-driving car had lidar or whatever they have, it should have detected her well before she came into view.

I still believe driverless cars will significantly reduce the road toll one day, but this tragedy says to me they're not ready for use in populated areas yet.

* Judging by the markings on the road, she comes into view about 15 metres ahead. 54 km/h is 15 metres/second, though reports are driver was doing 38 mph (61 km/h). So maybe less than a second, but it looks like time for some reaction.

User avatar
Howzat
Posts: 850
Joined: Wed Aug 15, 2012 7:08 pm

Re: Autonomous cars? I think not

Postby Howzat » Thu Mar 22, 2018 12:12 pm

This is bad news for Uber. All of their financing is money betting on the idea Uber spins, namely that futuristic driverless pod cars juuust about to be deployed. Its a big story with big plans, smart money, disruption!, and black swans for everyone.

If that turns out to be hype and hot air, and they can't deliver, then the other side of the business is the taxi-company-with-an-app operation. That side loses money on every single ride. The more sales they make the more money they lose. It's billions.

Meanwhile, actual computer scientists are impressed that they been able to build neural networks that can learn to distinguish between pictures of cats, and pictures of dogs. It turns out doing that is a significant accomplishment for machine vision and AI. I'm not kidding.

AdelaidePeter
Posts: 1230
Joined: Wed Jun 07, 2017 11:13 am

Re: Autonomous cars? I think not

Postby AdelaidePeter » Thu Mar 22, 2018 12:51 pm

AdelaidePeter wrote: * Judging by the markings on the road, she comes into view about 15 metres ahead. 54 km/h is 15 metres/second, though reports are driver was doing 38 mph (61 km/h). So maybe less than a second, but it looks like time for some reaction.
On reflection, if the driver/car could only see 15 metres ahead, that is a major problem. Any responsible human driver would either slow down, or put the lights on high beam so they can see further ahead. So yeah, a responsible human driver would have done much better.

User avatar
Mububban
Posts: 3043
Joined: Wed Feb 20, 2008 12:19 pm

Re: Autonomous cars? I think not

Postby Mububban » Thu Mar 22, 2018 12:52 pm

AdelaidePeter wrote:I still believe driverless cars will significantly reduce the road toll one day, but this tragedy says to me they're not ready for use in populated areas yet.

* Judging by the markings on the road, she comes into view about 15 metres ahead. 54 km/h is 15 metres/second, though reports are driver was doing 38 mph (61 km/h). So maybe less than a second, but it looks like time for some reaction.
Watching that footage was sickening. I kept looking for a sign of the victim and saw NOTHING until she just appears in the headlights and then gets hit.
“It’s very clear it would have been difficult to avoid this collision in any kind of mode (autonomous or human-driven) based on how she came from the shadows right into the roadway,” the police chief told the outlet.
Given normal human reaction times, I think any driver would have had just enough time to go "oh ****!" and take their foot off the accelerator, but even if they got their foot on the brake, the car would not have slowed down at all.
Her white sneakers come into view at 3 seconds, her blue jeans and bike frame/wheels come into view at 4 seconds, and impact happens just before 5 seconds on the video counter. Short of swerving, that car was never going to stop in time :cry:

Strangely the victim wasn't looking at the approaching car at all, like she didn't know the vehicle was approaching? Or perhaps misjudged the closing speed and how much time they had to safely cross.

I think self driving cars will see accident and fatality rates plummet, but there will always be incidents. Machines are still programmed by fallible humans.
When you are driving your car, you are not stuck IN traffic - you ARE the traffic!!!

User avatar
Tequestra
Posts: 193
Joined: Tue Feb 27, 2018 2:12 pm

Re: Autonomous cars? I think not

Postby Tequestra » Thu Mar 22, 2018 1:01 pm

Image

The driver was asleep at the wheel. If my 94 yo Grand Mother was driving that Volvo, that poor cyclist would have walked away.

http://www.abc.net.au/news/2018-03-22/s ... ed/9575624
Viva le Tour Electrique' !!!

User avatar
g-boaf
Posts: 21320
Joined: Mon Sep 26, 2011 6:11 pm

Re: Autonomous cars? I think not

Postby g-boaf » Thu Mar 22, 2018 1:23 pm

bychosis wrote:
Jmuzz wrote:Oh video now, looking very bad for Uber, that's a total failure of the basics.
(Warning may disturb. Also not sure where official release source is so fake is a possibility unless verified)

http://ktar.com/story/1997617/tempe-pol ... ving-uber/
Assuming it is real. It appears the car has no infra red or heat sensing cameras, or any sort of lidar scanning. The pedestrian was already on the road, but in the dark. Heat sensing ir lidar should have picked up the person on the other lane with time to stop.
The footage is real, this is a disaster. I can't see how that car didn't pick up the person and stop. Even BMW 7 series and S-Class Mercedes from a couple of generations ago had this technology to be able to see people or animals in the dark and warn the driver or otherwise do an emergency stop if needed.

User avatar
Howzat
Posts: 850
Joined: Wed Aug 15, 2012 7:08 pm

Re: Autonomous cars? I think not

Postby Howzat » Thu Mar 22, 2018 1:48 pm

Mububban wrote:I think self driving cars will see accident and fatality rates plummet, but there will always be incidents. Machines are still programmed by fallible humans.
Machines can't avoid what they can't see. The essential problem is that they can't see very well. Compared to humans, these things are vision-impaired.

The challenge is a combination of very tough engineering problem for the sensing technology and very tough computational problem for the recognition of the signal.

Many people have a hard time beleiving that something that's easy for people could be hard for a computer to do. But such is the case.

User avatar
Alex Simmons/RST
Expert
Posts: 4997
Joined: Fri May 02, 2008 3:51 pm
Contact:

Re: Autonomous cars? I think not

Postby Alex Simmons/RST » Thu Mar 22, 2018 2:03 pm

The technology is pretty advanced but not everyone is using the same technology.

This is Google's from 3 years ago and they will be further along than this by now:
https://youtu.be/tiwVMrTLUWg?t=744

That's one particular example of driverless system seeing and anticipating what humans don't. Go back and watch the whole thing for some background.

User avatar
bychosis
Posts: 7250
Joined: Sat Jan 14, 2012 1:10 pm
Location: Lake Macquarie

Re: Autonomous cars? I think not

Postby bychosis » Thu Mar 22, 2018 2:06 pm

Howzat wrote:Machines can't avoid what they can't see. The essential problem is that they can't see very well. Compared to humans, these things are vision-impaired.
While the machines are currently vision impaired, humans are attention impaired. People can see better, but they can't concentrate any where near what a machine can do.

Also once the machines can see better all car should be able to learn from one car's mistake with a near instantaneous software update. Out of the box they'll also be better drivers, not having to learn from scratch like humans do.
bychosis (bahy-koh-sis): A mental disorder of delusions indicating impaired contact with a reality of no bicycles.

Scintilla
Posts: 325
Joined: Sun Feb 25, 2018 5:36 pm

Re: Autonomous cars? I think not

Postby Scintilla » Thu Mar 22, 2018 2:07 pm

Alex Simmons/RST wrote:
Scintilla wrote:
march83 wrote:In the entire history of the world, 1 person has been killed by an autonomous car and it's a failure. There were probably hundreds killed by human drivers just in 1 day yesterday.
THAT ^^^ is not the point.

These vehicles are not even generally available for sale nor in public use as a general 'thing', yet they are currently failing to identify and avoid key things. Like people :roll:

They have technology that is fine for coping with a closed-circuit racetrack, or maybe restricted access freeways (even here I kinda doubt their safety with stray and unexpected hazards), but they simply are not, and are not likely to be in the foreseeable future, able to deal with public access roads, with people walking, changing conditions, roadworks, detours, bicycle riders, kids and dogs or other animals about. Scrap 'em!
They already are. These are cars on the roads in a regular US city.

Did you read the latest information on what happened? Police reviewed the video. Not even the superior reaction time of a car's systems could have avoided this, let alone a human driver.
Yep, read all that. It may well prove to be the case that the circumstances of the pedestrian's action presented a truly unavoidable collision. However I have been hearing too much of the detection-failures of these vehicles when presented with *normal* road scenarios. As for being already on the roads - they are still only MINOR in number, in the US mainly, and basically in test-bed form. There are great concerns to me in their overall spread on our road system.

The 'autonomous-will-solve-it' expectation is just setting ordinary pedestrians and bicycle riders up for more motoring-abuse, just that the blame will be cast upon the vehicle. Try crossing roads in Italy or France and you will see how the pedestrian is king. I cannot believe that the autonomous parameters will be set to avoid all pedestrians crossing the roads, at the expense of faster progress.

Scintilla
Posts: 325
Joined: Sun Feb 25, 2018 5:36 pm

Re: Autonomous cars? I think not

Postby Scintilla » Thu Mar 22, 2018 2:11 pm

find_bruce wrote:Self driving trains are already here & auto pilot has been a feature of planes for decades.
Self--driving trains and autopilot on planes are a walk-in-the-park compared with the mayhem of dealing with a typical urban road environment.

User avatar
Tequestra
Posts: 193
Joined: Tue Feb 27, 2018 2:12 pm

Re: Autonomous cars? I think not

Postby Tequestra » Thu Mar 22, 2018 2:16 pm

Howzat wrote:Many people have a hard time believing that something that's easy for people could be hard for a computer to do.
We do tend to take the complexity of what happens between our own ears for granted, out of habit I suppose. The network of different neural pathways that would be used in the process of driving around the block might resemble a satellite video of a small city in real-time. It is quite an amazing biological computer.

I was just thinking after seeing that video from the Volvo, that mammals were indeed evolving long before the dinosaurs were conquered. I don't know how I could possibly guess how much longer than 65 million years ago it might have been that the structure inside your head first began to evolve into its modern guise, but it was a very long time ago, and your brain has passed the test of time because your genetics survived it.

It maybe seen as a tad hubristic for Science to think that it can reverse-engineer such a grand design as the human mind, which has survived through so many different geological and climatic, and other environmental and social conditions that we could not even reliably be sure to even dream about its potential with our very best imaginations.

The mind can do things that modern science does not necessarily understand the reasons for, such as swerving left a few degrees to save the life of biological organism.

R.I.P. Elaine Herzberg.
Viva le Tour Electrique' !!!

User avatar
mikesbytes
Super Mod
Super Mod
Posts: 22159
Joined: Thu Dec 01, 2005 11:42 pm
Location: Tempe, Sydney
Contact:

Re: Autonomous cars? I think not

Postby mikesbytes » Thu Mar 22, 2018 2:47 pm

Tequestra wrote:Image

The driver was asleep at the wheel. If my 94 yo Grand Mother was driving that Volvo, that poor cyclist would have walked away.

http://www.abc.net.au/news/2018-03-22/s ... ed/9575624
Enough time/distance to swerve
If the R-1 rule is broken, what happens to N+1?

Jmuzz
Posts: 631
Joined: Mon Jan 29, 2018 1:42 pm

Re: Autonomous cars? I think not

Postby Jmuzz » Thu Mar 22, 2018 3:04 pm

The video contrast and dynamic range is terrible, which is normal for a cheap dashcam.
There were street lights and headlights (which are good in this car by all reports) so real vision would be as good as it gets at night, not a big dark vortex like the video.

Regardless the real sensors should be IR illuminated and be seeing everything. Plus the other sensors.

There were possible situations where it wasn't possible to react in time. This video is not any of those situations.

There are ANCAP rating tests with this exact scenario, production collision avoidance systems pass that test. This didn't, meltdown, failure at the most basic level.

Given the failure shown it is justified to suspend autonomous tests worldwide until a third party review has explained how this vehicle passed the qualification tests, or which part burnt out at exactly the wrong time.
It's just not a situation which should be possible.

User avatar
silentC
Posts: 2442
Joined: Mon May 05, 2014 5:24 pm
Location: Far South Coast NSW

Re: Autonomous cars? I think not

Postby silentC » Thu Mar 22, 2018 3:23 pm

Was she a poet by any chance?

(Sorry just my sick sense of humour).
"If your next bike does not have disc brakes, the bike after that certainly will"
- Me

fat and old
Posts: 6179
Joined: Sat Aug 30, 2014 12:06 pm
Location: Mill Park

Re: Autonomous cars? I think not

Postby fat and old » Thu Mar 22, 2018 3:25 pm

Hmpf....and I was canned.... :lol:

On another note, any other employers or supervisors here? I for one look forward to the new excuse “sorry, I will be late/won’t make it. I forgot to update my car last night” :lol: It WILL happen....

User avatar
WyvernRH
Posts: 3179
Joined: Tue Nov 06, 2007 6:41 pm
Location: Newcastle NSW

Re: Autonomous cars? I think not

Postby WyvernRH » Thu Mar 22, 2018 3:31 pm

mikesbytes wrote:While not specifically commenting on this incident, I can see another "not my fault" excuse coming to the fray;
1. The Autonomous mode drives conservatively complying with all traffic laws and applying safe practices
2. The occupant wants to get there quicker, not leave a gap to the car in front to prevent others cutting in etc etc
3. To achieve 2. the occupant drives manually
4. An incident occurs
5. The occupant says the vehicle was in autonomous mode
Ah but... the autonomous vehicle will have a log in the computer somewhere (probably only available to authorized access) which will record everything the car does, when it did it, why it did it and who did it if an external command. So in the above case interrogation of the log by the constabulary would reveal the user to have ordered the car to hand over control using their personal command code or token (whatever). Probably also going to be used if authorities want to know where you have been and at what times.
A bit like an aircraft's black box.

Richard

User avatar
WyvernRH
Posts: 3179
Joined: Tue Nov 06, 2007 6:41 pm
Location: Newcastle NSW

Re: Autonomous cars? I think not

Postby WyvernRH » Thu Mar 22, 2018 3:51 pm

Howzat wrote:
Mububban wrote:I think self driving cars will see accident and fatality rates plummet, but there will always be incidents. Machines are still programmed by fallible humans.
Machines can't avoid what they can't see. The essential problem is that they can't see very well. Compared to humans, these things are vision-impaired.
While I can't say what sensors the Uber car was equipped with the above statement is completely wrong. Humans have crappy eyesight compared with a lot of animals and we can only receive data in a restricted range of wavelengths. A well designed sensor array leaves us humans way behind. A good infra-red and lidar driven overlay on a heads up display has to be seen to be believed in low light/smoke/cloud conditions.
Interpreting and taking action on what you are receiving is the tricky bit. Humans can be bad at that too. I have been in a car where a roo jumped out and the driver just froze (in indecision?). Dead roo and big repair bill only that time fortunately but it could have been a child.

Richard

User avatar
Tequestra
Posts: 193
Joined: Tue Feb 27, 2018 2:12 pm

Re: Autonomous cars? I think not

Postby Tequestra » Thu Mar 22, 2018 4:14 pm

WyvernRH wrote: Interpreting and taking action on what you are receiving is the tricky bit.
This is puzzling looking at the still photo above, because there are bright building- and street-lights in the background, and it seems that Ms Herzberg's bright white shoes and socks belong with the lights above, if you imagine it in two dimensions.

I reckon that the best way to incorporate autonomous vehicles into the current and historical set of road rules would be to insist that a licenced driver always be behind the wheel and that driver is entirely responsible for what the vehicle inflicts on the community, assuming that there was no fault with the vehicle's conventional manual apparatus.
Viva le Tour Electrique' !!!

Who is online

Users browsing this forum: No registered users