Scientific American Article: The Truth about “Self-Driving” Cars


#1

The Truth about “Self-Driving” Cars by Steven E. Shladover

The auto industry and the press have oversold the automated car. Simple road encounters pose huge challenges for computers, and robotic chauffeurs remain decades away. Automated driving systems that rely on humans for backup are particularly problematic. Yet in the next decade we will see automatic-driving systems that are limited to specific conditions and applications. Automatic parking valets, low-speed campus shuttles, closely spaced platoons of heavy trucks and automatic freeway-control systems for use in dedicated lanes are all feasible and perhaps inevitable.


Basically, the industry divides car/driver automation into 5 levels, from (1) partial automation like adaptive cruise control to (5) full automation.
The author believes numb 5 is at least 60 years in the future due to the huge advances in software required. There are no existing methods to debug, validate, or “prove” that the system is safe. Research into those areas is just starting.

Number 3 where the human driver takes over in an emergency will probably never be implement due to the possibility of the human being distracted or even asleep.

Number 4 which is full automation but under limited circumstances, such as special highways and parking garages, may be available within 10 years.


#2

Seems like obvious conclusions to me. Technology like this must be developed in stages. You don’t go from 0 to 100 instantly.

There are no existing methods to debug, validate, or "prove" that the system is safe
This I disagree with. Of course there are methods available. However, they are prohibitive in terms of resources and expense.

How do you suppose they proved that launching a rocket with a human aboard was relatively safe? They did it in stages with V&V along the way. They didn’t just build a rocket and stuff some dude on top of it. They first built rockets with sensors that barely punched through the atmosphere and kept improving/expanding upon success (or learned from failure). Over time, they gained enough experience to attempt the “final” goal. This is no different except there isn’t the same level of urgency attached to the effort…


#3

Yes, that is the point, there are no methods available that would work with systems of this complexity.

The aircraft software for a fighter jet take many millions of dollars and years to validate, and that is only 1/10 or 1/100 of the requirements for an automated car. And aircraft can use very expensive redundant computers.


#4

Full automation is FAR more complicated than any of these articles can understand. In well controlled environments, automation does fairly well. Add the randomness of the real world and things become much more difficult. Accidents with human driven vehicles most often happen because one person does the unexpected. The proper reaction can sometimes prevent the accident, or make things worse based on the options for avoidance and/or both operator’s ability to avoid contact. This is a hard thing for a machine to sense, formulate a plan and execute that plan, in real time at the SAME time as another person is doing the EXACT same thing. If it is another machine, communication and control protocols must be developed to prevent contact.

A study of mine into robotic transporters working in a tire plant identified a novel way to avoid the head-to-head stalemate of two machines in each other’s path with a travel path allowing only one robot to pass. The right of way was to the robot with the lowest serial number. #1 was always King. #71, the last was always the lowest ranking robot.

We underestimate the abilities of even the dullest, least skilled, texting drivers to control the randomness of driving.


#5

My question is: What exactly do the Google cars do? My guess is that they require an alert human driver overseeing 100% of the time.

Yesterday I hit a small detour on a city street. There was construction workers controlling the flow, and I had to drive on the wrong side of the street as directed. I cannot see a computer duplicating that with 100% safety and reliability.

edit: the computer would have to recognize a construction worker and the SLOW/STOP sign he was carrying, put that together with the blocked lane, recognize the hand motions of the worker, check for oncoming traffic, and disobey laws to cross the yellow line and drive in the wrong lane. Then recognize when it’s safe to return to the proper lane.


#6

Completely Automated Self-Driving Cars…
What Could Possibly Go Wrong? :neutral:
CSA


#7
The author believes numb 5 is at least 60 years in the future

That’s a pretty long time. To put that into contrast, 60 years after the Wright brothers limped their plane into the air, Lockheed first flew the YF-12, which is a variant of the rather more famous SR-71 Blackbird, and Gordo Cooper orbited the Earth 22 times in the final flight of the Mercury spacecraft program.

Now, I bet if you grabbed a leading aeronautical scientist in 1903 and told him “Hey, in 60 years we’ll be flying more than 3 times the speed of sound and launching people into outer space” he’d have told you to stop drinking.

Even more impressive, the first general purpose computer was Einiac. It was announced in 1946, cost the equivalent of 6.8 million of today’s dollars, weighed 27 tons, was 100 feet long.

60 years later, even the most basic cell phone on the market had over 1,000 times Einiac’s power, and smart phones weren’t even much of a thing yet.

10 years after that, smart phones are ubiquitous - almost all of us carry around in our pocket a device that would have looked almost magical to the scientists working on Einiac.

I find it somewhat difficult to believe that it will take us 60 years to come up with suitable vision systems for computers that allow them to independently navigate (and really, the vision system is the big stumbling block).


#8

My take on the problems is that the vision/navigation/computer systems have to be correct 99.99…% of the time. (add as many 9’s as you think are appropriate).

Because one mistake could result in a death.


#9

Can’t think of anything more boring than sitting behind the wheel/stick/yoke/whatever while the car drives. Whatta ya do, read, adjust the radio, talk on your phone?

I smell something but I’m not sure what it is. Years ago there was a fight between the folks that wanted mass transit like light rail, and those that thought personal vehicles on tracks would be better sociologically. I can’t recall what they were called, personal pods or something. Then the trains won and massive federal funds propelled it. Now the traffic problem is still there and it sounds like some cities would like to see the personal pods too in the form of self-driving cars. Of course it will take massive amounts of federal and state funds to do.

Now back in 1974, we used to commute into downtown Minneapolis on the freeway. Once in a while there would be a back-up but most of the time we sailed right through with no problems. Then somewhere around the mid-80’s there was massive construction downtown of office buildings and so on. Traffic became deadlocked and we all found different routes and side streets to get to and from work.

So to me the traffic problem was caused by excessive downtown development beyond the infra-structure. Then of course you had to boost mass transit to try and compensate which then boosts more development downtown. Just a vicious cycle with no end. As they say, follow the money. Who is making money on the whole cycle. And there is so much undeveloped land beyond the big cities but of course it would cost less and you wouldn’t need a train.


#10

I would argue that they only have to be better than humans.

A self driving car system that screws up and kills 25,000 people in one year sounds horrific, but it’s much better than the 32,000 people who were killed by humans driving cars in 2014.


#11
I would argue that they only have to be better than humans

Sure you would argue that @shadowfax, you are a person with common sense!

Lawyers would advise a MUCH different approach and as such, the engineers assigned to these programs must as well. @BillRussell, I think has the right approach… 99.99% correct and even that 0.01% must be defensible in court to 12 ordinary people.


#12

I’ve worked in the software engineering space for about 40 years. I use to predict the direction and depth of where the industry was going. I stopped trying to predict a couple decades ago because my predictions have almost always fallen well short of reality.

I’ve said numerous times in this forum about this topic and that the technology isn’t there yet, but it’s much farther along then I thought possible just 5 years ago. My prediction is the technology will be ready in 10 years or less. Based on my previous predictions means it’ll be here in about 3.


#13

Yes, I do not understand the prevailing position that these systems must be infallible. Any hazard and risk assessment exercise will be to reduce risk to an acceptable level. What is truly acceptable is up for debate but certainly, if infallibility is your measure, nothing would ever be built or deployed as a consumer item…


#14

The technology is there and it has been for a long, long time. The hold up is public acceptance and allowing such vehicles access to public highways or at least separate travel lanes. The public is not ready to share the road with automated cars, but, depending upon how much you are willing to spend, cars all ready have automated brakes, throttles and steerage and have since abs and peripheral electronic gear was added to cars
.
Your throttle and pedals are mearly suggestions and the steerage for a few cars is already but with mechanical linkage back up.

Computers easily have the capability but frankly, like the rest of you, I am not ready to share the road with computer driven car which is a much better driver then I am. The military first then the airlines industry have used this lifesaving technology for years and passengers have been flying with it.

It’s just about public acceptance.


#15

Btw, texting and holding hands with the wife would take on a new meaning when automated driving takes over. Problem drinkers and illegal drug users would love the option. Imo, it should come with sensors that does not allow users to activate the starters. Of course, I feel that way about all cars once a driver has been convicted of using while driving.


#16

@shadowfax, I think Takata would like to have you sitting on their jury. Their airbags have saved many countless thousands of lives, but have cost the lives of a little over a dozen people.


#17

Dagosa, I disagree, the technology is not there, and will not be there for a long time to come.

Just think of the difficult driving situations you have been in, and think about how a computer could handle those. The detour I mentioned in my last post. An inch of snow, so all the lane markers are obscured. A school crossing with a guard. A cop directing traffic.

These all have to work 99.99% of the time. As a comparison, human drivers handle the above situations 99.99% of the time, and the computer has to do at least as well.


#18

The Google car is capable of handling itself in Mountain View, CA, a mix of urban and suburban driving. I don’t think they have been on the highway yet. To date, there is one accident attributable to the automated car, and these cars have driven over 1.5 million miles. In addition to Mountain View, the self driving cars have tooled around Austin TX, Krikland, WA, and Phoenix AZ. Google may eventually decide to sell them in the open market, but I think their Big Idea is to use them to get street level views wherever cars drive. A truck driver could show up with a trailer full of self driving cars, let them loose in a town near you and collect them after taking pictures of everything. Then it’s on to the next town.


#19

A few bewilderments occurred to me while driving today, based on things that happened on the road.

  1. would self-driving cars always obey the speed limit? Like that 110 year old Buick driver that I was behind today?

  2. what would it do when it came upon a cop car with a pulled-over driver, both vehicles out over the solid white shoulder line and into the travel lane? Saw that a few times today.

  3. would it find a safe and immediate space to pull over when an ambulance came up behind it with the lights flashing? Or would it drive off onto the woods? Or keep going as if the ambulance wasn’t there?

  4. how will it react to complicated and confusing construction areas?

  5. could it find it’s way out of Boston!!!


#20

When a self driving car can find it’s way around Atlanta, GA where I am told there are 32 streets with Peach Tree in the name of some form I will then call them a success.