Uber Troubles For Uber

[LordFerret]

3 minutes ago, kerbiloid said:

Afaik, almost all aircraft accidents with numerous victims of last 10-15 years were man-made.
Almost all of them would be by default prevented if AI was ruling the plane.

So, AI is anyway lesser evil than a human crrew.

Aircraft accidents... this comes to my mind.....

 

The thing with aircraft AI is that it's extremely dependent on backup systems and external support... especially the external support. That's why it's as sophisticated and 'safe?' as it is today. Uber cars have no such backup systems or external support.

 

6 minutes ago, kerbiloid said:

If make cars AI, nobody can steal a car.

That's not true. Already I've read news reports and heard discussions about AI cars being hacked. You can't even prevent a damned PC from being hacked... and you expect them to protect a car? Nope.

[kerbiloid]

2 minutes ago, LordFerret said:

Uber cars have no such backup systems or external support.

So, they should be assembled into a network.
But this accident was probably caused by algorithmic problem.

[YNM]

2 hours ago, tater said:

Traffic deaths have dropped for years, almost entirely as a result of automotive engineering improvements, and trauma care.

What, because of those two improvements I can jump onto the path of a lorry on the motorway and be guaranteed staying alive and able to entirely heal ?

 

25 minutes ago, kerbiloid said:

Afaik, almost all aircraft accidents with numerous victims of last 10-15 years were man-made.
Almost all of them would be by default prevented if AI was ruling the plane.

If JFK were to be served by only one runway at any time you'd have more accidents.

[tater]

1 hour ago, LordFerret said:

The thing with aircraft AI

Not AI.

56 minutes ago, YNM said:

What, because of those two improvements I can jump onto the path of a lorry on the motorway and be guaranteed staying alive and able to entirely heal ?

You think if you jump in front of a car driven by a human you will live? Human drivers usually speed, so any hit will be worse (self-driving cars don't exceed posted speed limits generally). A large % are distracted a significant fraction of the time they are driving. Anyone walking out in front of a car such that they could be hit is asking to be hit. I never time jaywalking such that it's even possible to be hit, I know I can get across, or I don't start crossing. I guarantee that if there were some unethical experiment where people cross outside the walk such that the approaching car is required to stop to let them illegally cross and live, a decent % get hit (and all likely get a horn or yelled at).

1 hour ago, LordFerret said:

Before removing the driver, I'd remove the phone. Apparently this is being done, and it's working.

https://www.nhtsa.gov/press-releases/usdot-releases-2016-fatal-traffic-crash-data

I see people on phones literally every single time I drive. Many. My phone never leaves my pocket in the car. And yes, even dialing a phone, ever in the car makes anyone as bad as a drunk IMO.

Note that I'm not defending the safety of this particular Uber system, it clearly has a flaw someplace. That said, if it never saw people, there would be many more crashes, and the self-driving versions are quite possibly less deadly than meat-driven Uber cars. The answer is in the mortality per mile driven. If that number is lower, they are safer, period.

Most all airline accidents are pilot error. Many are CFIG incidents.

That Air France crash was a demo flight. The accident report claimed pilot error. The pilot disputed this, and claimed a problem with fly by wire controls prevented him from throttling up. The engines in fact were throttled up, exactly in line with spec values. Regardless, it was not "AI," and not even really flying itself unless the pilot set it up to (badly). Commercial airliners fly themselves most all the time they are aloft, actually, though the computer is not really doing anything intelligent at all, just maintained set points, it's nothing like self-driving in terms of complexity.

Edited March 21, 2018 by tater

[YNM]

10 minutes ago, tater said:

You think if you jump in front of a car driven by a human you will live?

That's the point ! It doesn't matter who drives, you are dead !

 

... unless if they drive slower, of course...

 

I'm collating some data on road traffic accidents in the US vs the UK. But just for starters, it's looking grim.

- Compared to each respective population (I take UK to be 65.64 million while US to be 325.7 million), you're more likely to die in a road-related accident in the US than the UK, by a factor of 4.

- By the same methods, you're twice more likely to die as a pedestrian in a road-related accident in the US compared to the UK.

Source data for the UK, 2016

Source data for the US, 2016

 

I'm not sure what this talks of each country's characteristics, but it seems the US is slightly lagging behind on road safety.

[kerbiloid]

Absolute majority of accidents are caused not by rare combination of factors, but just by human Courage and Stupidity.

AI can filter out most of them.

[tater]

15 minutes ago, YNM said:

I'm collating some data on road traffic accidents in the US vs the UK. But just for starters, it's looking grim.

- Compared to each respective population (I take UK to be 65.64 million while US to be 325.7 million), you're more likely to die in a road-related accident in the US than the UK, by a factor of 4.

- By the same methods, you're twice more likely to die as a pedestrian in a road-related accident in the US compared to the UK.

So what? Have you been to the US? How fast are cars driving in typical pedestrian vs MVAs in the UK? Distances here mean cars drive faster. The average US work commute drive is 15 miles. The average in the UK? 8.3 miles. About half...

Regardless, I think self-driving cars will rapidly become superhuman in their ability to mitigate collisions.

 

 

[YNM]

7 hours ago, tater said:

How fast are cars driving in typical pedestrian vs MVAs in the UK ?

Well, they have a gracious amount of 20 zones inside cities. The motorways and intercity high-quality dual carriageways is mostly 70 though, a good deal are lowered to 60, 50 or 40. (EDIT : There are also Variable Speed Limits, where the limit is meant to be adaptable to conditions.)

7 hours ago, tater said:

The average US work commute drive is 15 miles. The average in the UK? 8.3 miles. About half...

I'll nitpick : why even cars ? Where's the PT ?

 

 

EDIT :

Spoiler

Here's a reflection. I don't mean for people to just exactly follow whatever's shown - but changing can be a long journey. The US won't change in a snap, I'm sure - but as long as the conscience is there, progress will be made.

 

Edited March 21, 2018 by YNM

[tater]

6 hours ago, YNM said:

I'll nitpick : why even cars ? Where's the PT ?

When there are self-driving cars, the US will have PT.

Take where I live, Albuquerque. ABQ is ~40x40 km, for maybe 700,000 people. The streets are mostly gridded out. People can get to a bus in several hundred meters from most houses, but most bus rides require a transfer (they run N-S or E-W on major streets a mile (1.61km) apart). In the June (late May, even) through Sept it's 35 C, sometimes above 40C. It;s usually sunny, with few clouds (300+ sunny days a year). PT is a non-starter. Other than the cool, early morning, any trip would be incredibly uncomfortable, which is why people don't use the bus. The city is simply too spread out. This is common away from the coasts, and we are not going to retroactively build medieval, European cities. When someone can hail a car with their smartphone that picks them up at their door, drives them, then gets someone else (possibly electric, too!) ABQ will have people using this new form of PT. Short of that? LOL. Not happening.

[YNM]

10 minutes ago, tater said:

When someone can hail a car with their smartphone that picks them up at their door, drives them, then gets someone else (possibly electric, too!) ABQ will have people using this new form of PT. Short of that? LOL. Not happening.

Perhaps one day something will change for good 

[XB-70A]

7 hours ago, tater said:

Most all airline accidents are pilot error. Many are CFIG incidents.

While I agree with this on a global perception, it is also important to note that the sequence of events and decisions also plays, and that embedded systems (as on the ground) also have their share of "responsibility" in most cases. Often due to a simple error or non-consideration of the human factor when designing the system.

 

7 hours ago, tater said:

That Air France crash was a demo flight. The accident report claimed pilot error. The pilot disputed this, and claimed a problem with fly by wire controls prevented him from throttling up. The engines in fact were throttled up, exactly in line with spec values. Regardless, it was not "AI," and not even really flying itself unless the pilot set it up to (badly). Commercial airliners fly themselves most all the time they are aloft, actually, though the computer is not really doing anything intelligent at all, just maintained set points, it's nothing like self-driving in terms of complexity.

Indeed, the case of the flight 296 was particular and was a good example of flying mistake and conception mistake. We can define the choices and actions of M. Asseline as incorrect (as well the absence of "proper" reactions from the F.O.), but Airbus also was responsible for some mistakes made at the time. 

Here are two notices for the flight operations at the time of the accident:

- Operation Engineering Bulletin 19/1, noting that the CFM56-5B could be subject to a slow thrust rising at low altitude and high angle of attack.

- Operation Engineering Bulletin 06/2: potential deficiency of the on-board altimeters leading to erroneous displays at low altitude.

 

Another one is the choice by Airbus, for safety purpose, to have included a control inputs "limiter" on the sidesticks. Although the system seems justified in most cases to avoid stalls, it seems that it also had its share of responsibility in the accident, preventing the PIC to pitch up enough for a very short period of time. If the action could have been done, it might have saved the aircraft and the few passengers who passed through it.

 

However, this flight is a good example that if in aviation the sequence of events is recognized as the main cause of incident and accident, the legal systems (and the manufacturers also) are privileging the choice of a single guilty. After all, if we were dwelling on the responsibility of everyone, we could very well blame a toilet cleaner for not doing her/his job properly, and so have played on the efficiency of the engineer who designed a failing system.

[YNM]

33 minutes ago, XB-70A said:

However, this flight is a good example that if in aviation the sequence of events is recognized as the main cause of incident and accident, the legal systems (and the manufacturers also) are privileging the choice of a single guilty. After all, if we were dwelling on the responsibility of everyone, we could very well blame a toilet cleaner for not doing her/his job properly, and so have played on the efficiency of the engineer who designed a failing system.

... And this is why I ranted about road design and usage. After all, removing the possibility entirely is more safe than just anticipating for it.

[tater]

A reality check:

Quote

A fatal crash involving a self-driving Uber likely was "unavoidable" based on an initial police investigation and a review of video, Tempe Police Chief Sylvia Moir told The Arizona Republic on Tuesday.

The woman was walking a bike across the street at 10 pm, outside of a crosswalk. She walked from shadow directly in front of the moving car, and a human driver would not have seen her, either.

 

Quote

A large median at the site of the crash has signs warning people not to cross mid-block and to use the crosswalk to the north at Curry Road instead. But the median also has a brick pathway cutting through the desert landscaping that accommodates people who do cross at that site.

 

[YNM]

Also reality check (source) :

Quote

Police have said the Volvo had a video camera that recorded the crash. The Volvo was traveling about 40 mph and made no visible attempt to brake in the video, Elcock said.

The speed limit in the area is 35 mph.

 

Also, guess what I found :

http://www.tempe.gov/city-hall/public-works/transportation/traffic-counts

So far the GIS is a bit of PITA though. But dang it is amazing to have data(s) online !

 

EDIT :

Found a different source :

http://data-tempegov.opendata.arcgis.com/datasets/traffic-counts/data?selectedAttribute=CLASS

True the website's a tiny bit broken, but you can obtain them.

Edited March 21, 2018 by YNM

[tater]

Real drivers virtually always exceed posted limits by around 5-10mph in my experience. A car moving at the posted speed would likely be more dangerous than one moving at nominal speeds.

I actually witnessed a (drunk) woman walking into a car on a 4 lane divided road (Tramway Blvd here in Abq). Luckily the light had just changed, since the limit is 45 in that spot, and traffic moves at 50, so the car was going much slower. We pulled over and my wife and another doc who happened to have seen the crash rendered assistance.

Not the driver’s fault, she stepped into traffic from the median. Could not have done better if she was trying to kill herself.

If the Uber blew a crosswalk/light I’d be far more concerned.

 

[YNM]

7 minutes ago, tater said:

Real drivers virtually always exceed posted limits by around 5-10mph in my experience. 

Time for a change.

Also

Quote

As far as your speedometer is concerned, the (UK, ed.) law says that it absolutely must not under-read (so it can't show a lower speed than the one you're doing) but it may over-read by 10% + 4km/h. Most manufacturers will calibrate speedos to sit partway between the true speed and that margin of error, so if you appear to be doing 30mph you could actually be travelling at anything between 25 and 30, and a readout of 70 could mean anything between 61 and 70.

This means that enforcement tends to start at a point some way above the number on the signs. Many police forces in the UK share an "enforcement threshold" of 10% + 2mph on top of the limit, below which they will take no action (though some police forces will enforce limits at a lower threshold). There is a second threshold beyond which the offence is considered sufficiently serious that the case will be sent straight to court. However, speeding is still speeding, and even if they decide not to pull you over for doing 55 in a 50 zone, police patrols are still very unlikely to give you a thumbs up as you go past.

(source, emphasis mine).

[tater]

I drive many thousands of miles a year, and pass cops at a few mph above the limit routinely, as does literally everyone else.

The safest speed is actually above ambient traffic speed (the so called “Solomon curve.”

[YNM]

4 minutes ago, tater said:

The safest speed is actually above ambient traffic speed

Safe for you, not safe for the poor sod you hit.

[tater]

8 minutes ago, YNM said:

Time for a change.

Increase the speed limits?

The limit on the fun, curvy hill to my house is 35. LOL. I go fast, but my tires don’t even squeal, so I’m not even close to pushing it (and that’s grossly above 35). Literally the only fun driving in this town.

[kerbiloid]

Spoiler

3 hours ago, tater said:

ABQ is ~40x40 km, for maybe 700,000 people.

 

Wiki says:
area ~490 km2, (i.e. probably ~40x40 is because of uneven shape),
population ~560 k, so 700 k including visitors.

(And 4.7 km² of water.)

Probably mostly 2-storey houses, with ceilings ~ 0.003 km high or so.

3 hours ago, tater said:

People can get to a bus in several hundred meters from most houses, but most bus rides require a transfer (they run N-S or E-W on major streets a mile (1.61km) apart).

So, we can take ~3.2 km between bus routes.

***

Estimated inhabited volume of the city ~= 490 * 2 * 0.003  = 3 km³.

Say, if take edge = 2 miles = 3.2 km, then height = 3 / 3.2² = 0.3 km.

So, if build a 300 m high, 3.2x3.2 km wide, ziggurat (10 km² of land area) near that water (4.7 km²), with a single bus road along perimeter, one can get to the nearest bus stop at least as quickly as now.

If add a diagonal cross of roads, a walk will take ~3 times less.

Let's add such square every 100 m of height (i.e. ground, 100 m, 200 m, roof).
And fast elevators between the 100 m transport layers.

Then longest possible route would be: ~500 m by foot (5 min) + 6 km by bus tram (6 min) + 300 m by elevator (1 min or so) = 10..15 min.

And no cars at all. No road inidents. No exhaust and greenhouse gases. No drunken drivers speaking phone, no robocars AI.
No visible moving transport at all. Trams in horizontal tunnels, elevators in vertical ones.
No road police, no penalties.

And all this beauty right on the lake (or what is that "water area"), surrounded by trees, with BF(R/S) launchpad on the other side of the water.

 

26 minutes ago, tater said:

Real drivers virtually always exceed posted limits by around 5-10mph in my experience. A car moving at the posted speed would likely be more dangerous than one moving at nominal speeds.

It moves backwards relative to normal brave drivers. Down the oncoming side. Pure kamikadze.

Edited March 21, 2018 by kerbiloid

[tater]

Just now, YNM said:

Safe for you, not safe for the poor sod you hit.

You seem to be under the impression that this is a small road in a village in the UK, with people crossing between shops. Look at the map of where the Uber crash happened. It was like crossing a highway on foot.

Ped vs MVAs here are usually drunks—the pedestrians, not the drivers.

[YNM]

6 minutes ago, tater said:

You seem to be under the impression that this is a small road in a village in the UK, with people crossing between shops. Look at the map of where the Uber crash happened. It was like crossing a highway on foot.

Ped vs MVAs here are usually drunks—the pedestrians, not the drivers.

So what, you can kill drunk people ?

(well I would actually, though that still sounds a bit of a waste)

 

Also, did you not read the "controversy" ? (the "standard" source)

Quote

In 1968, Julie Cirillo conducted a similar study of 2,000 vehicles on interstate highways that addressed speed variation’s impact on crashes that involved two or more vehicles.^[8] The Cirillo data produced a U-shaped curve similar to the Solomon curve.^[9] The Research Triangle Institute conducted a study in 1970 where data was collected on 114 crashes involving 216 vehicles on a state highway in Indiana to address these concerns by (1) combining automated, embedded speed-monitoring stations with trained on-scene crash investigators, and (2) distinguishing data on vehicles slowing to negotiate a turn from vehicles moving slowly in the flow of traffic.^[10] Reporting on these results in 1971, academics West and Dunn confirmed the findings of Solomon and Cirillo,^[11] but found that crashes involving turning vehicles accounted for 44 percent of all crashes observed in the study and that excluding these crashes from the analysis greatly attenuated the factors that created the U-shape of the Solomon curve.^[12]

It should be noted that excluding accidents "due to turns" would be fallacious for freeways and highways with limited access, where there are no turns available to disprove the Solomon curve.

In 1991, Fildes, Rumbold, and Leening collected self-reported crash data from 707 motorists in Australia with fewer than 200 reporting that they had been in a collision but, unlike Solomon and Cirillo, the researchers found no relationship between slower speeds and increased crash involvement.^[13]Notwithstanding the many studies over the years, in testimony before the Ohio Senate Highways and Transportation Committee on June 10, 2003, Julie Cirillo, Former Assistant Administrator and Chief Safety Officer for the Federal Motor Carrier Safety Administration (FMCSA), testified that "up to the present time there has been no evidence to alter Solomon’s original finding that variance from the mean operating speed is a major contributor to accidents".^[14]

In July 2001, Kloeden CN, Ponte G and McLean AJ of the Road Accident Research Unit, Adelaide University quantified the relationship, "... between free travelling speed and the risk of involvement in a casualty crash in 80 km/h or greater speed limit zones in rural South Australia" using a case control study design. They found, "..the risk of involvement in a casualty crash increased more than exponentially with increasing free travelling speed above the mean traffic speed and that travelling speeds below the mean traffic speed were associated with a lower risk of being involved in a casualty crash." Outlining past research in this area, they suggest that, in the Solomon research, "Both the numerator (number of crashes in a particular speed band) and the denominator (number of vehicle-miles travelled in that same speed band) may have been quite inaccurate for relatively low speeds." ^[15]

 

_____________

@kerbiloid Well, you know.

 

Edited March 21, 2018 by YNM

[tater]

6 minutes ago, YNM said:

So what, you can kill drunk people ?

You cannot avoid hitting someone who walks in front of your car within some range.

Human reaction time is at best ~1-1.5 seconds. Add that to the time it takes to brake to 0, and you get a boundary for avoidable ped crashes. 

At 35 mph (~16 m/s), we are then in a certain hit zone within maybe 20m? (Assuming the driver instantly sees the ped)

[YNM]

7 minutes ago, tater said:

At 35 mph (~16 m/s), we are then in a certain hit zone within maybe 20m? (Assuming the driver instantly sees the ped)

Farther than that.

(source)

[tater]

Note that for in-town speeds, driverless cars will vastly improve this, since the thinking time goes from seconds to milliseconds, and at lower speeds, reaction time dominates.