Poll & Prediction: Autonomous Car Equipment at 5k by 2019

[Jeff Matthews]

There is an interesting article about how autonomous vehicles react in the event of an emergency i wish i could remember. Basically take this scenario. What if you are riding in your autonomous car and another vehicle loses control for whatever reason blown tire,steering or suspension failure etc and your car has two choices,hit a busload of kids possibly injuring or killing some of them or driving you off a cliff and most likely killing you. Which will it do? Not saying i think this is a bad thing just a lot of stuff to consider.

 

Once the AI is sufficiently-developed, the machines will turn on us.  When the machine makes a choice to preserve itself and disregard human lives, then, we will know it happened.

[oldtimer]

Therefore it wouldn't drive over a cliff.  Sorry kids.

[Mallette]

There is an interesting article about how autonomous vehicles react in the event of an emergency i wish i could remember. Basically take this scenario. What if you are riding in your autonomous car and another vehicle loses control for whatever reason blown tire,steering or suspension failure etc and your car has two choices,hit a busload of kids possibly injuring or killing some of them or driving you off a cliff and most likely killing you. Which will it do? Not saying i think this is a bad thing just a lot of stuff to consider.

That's the same choice a human would have, is it not?  Regardless of the choice the machine will do a better job of minimizing destruction than a human and overall the predictions in the decline of deaths and injuries far outweigh such concerns. 

 

Dave

[Gilbert]

Driver Killed While Using Tesla’s ‘Autopilot’ Feature

 

 

http://www.huffingtonpost.com/entry/tesla-autopilot-accident-crash-death_us_577584afe4b0a629c1a95099

[Don Richard]

I just saw that on TV a few minutes ago. Very sad. Doesn't surprise me though.

[JJkizak]

It's a wonderment to me that people view the "Non reliable state of electronics" as a totally reliable means of control. What happens with a lightning strike near the car? What happens when someone opens a garage door and it interferes with the car system? What happens when one of the hundreds of weather radars interferes with the car? What happens when somebody's modified drone electronics interferes with the car? What happens when a military radar interferes with the car?

What happens when a sloppy ham operator system interferes with a car? What happens when Tropo ducts frequencies 500 miles away that interfere with the car? I wonder if they will use NASA solid state devices to protect against excessive heat and radiation and tropical weather degradation. I would only trust the electronics if Western Electric built and manufactured them.

JJK

[Don Richard]

According to Inside Unmanned Systems magazine Nov./Dec. 2015, a publication for engineers that focuses on engineering, policy, and practice, it would take over 300 billion lines of code to achieve complete autonomy for driverless cars. The article states that, because of the complexity and in order to reduce costs, an open architecture applicable to all brands of vehicles would have to be implemented, with all manufacturers contributing financially to the effort. Middleware like the PolySync platform would interface the vehicle's sensors and systems with the open architecture control algorithms that will be developed. 

 

The idea of centralization is new to auto manufacturers who have preferred in-house solutions to industry wide solutions. In-house solutions create a competitive edge for the manufacturers, but standardization is the only practical option for systems with the level of complexity necessary for fully autonomous control of road vehicles. For example, 50 million miles of testing different scenarios will be required to determine if the system is safe or not. A method of updating the control algorithms will also have to be worked out before public release.

 

No predictions concerning a time table for selling fully autonomous vehicles were in the article, other than saying, "In the future". 10 - 15 years seems about right to me. Time will tell.

Edited July 1, 2016 by Don Richard

[Mallette]

“Neither Autopilot nor the driver noticed the white side of the tractor trailer against a brightly lit sky, so the brake was not applied.”

 

Of course, the ridiculous is emphasized but this is the key statement.  The specific issue of the programming will be fixed, if it hasn't been already.  But how do you fix the millions of drivers who "don't notice" all the time?

 

There will be more of these, and the luddites will seize on them...but they will be the exception rather than the rule and Musk's prediction will come true. 

 

Dave

[Mallette]

"In the future". 10 - 15 years seems about right to me.

 

Give current semi-autonomous systems already are estimated to reduce the accident rate for the unreliable human by 40%, one needs to bear in mind that the path to complete autonomy such that your 4 year old can be picked up at school by the car isn't an all or nothing thing.  Near autonomy is already here...actually, full autonomy is already available if you ignore the maker's warnings as the dead guy above did.  While I'd trust a Tesla on autopilot over any human driver already I certainly would be watching just as hard as I do now for potential issues and be ready to respond.  That is what is expected at the moment until such kinks as this one are located and fixed. 

 

300 billion lines of code?  Given that that with much less human performance is already less than that of the Tesla or even the lesser vehicles with adaptive cruise, auto braking, and lane control I think that's a bit on the fringe.

 

Dave

[Don Richard]

The article said closer to 300 billion than the 300 million lines of code already in some present day vehicles. Apparently a lot of fuzzy logic is necessary to cover all possible conditions that the vehicle would have to handle. The article did predict that in 25 - 30 years the vehicles would be perfected to the extent that wrecks would be a thing of the past.

[Mallette]

40% improvement of currently available technology works for me, and I don't believe for a minute it will take more than a few years to get to 90%.  That is after about 4 years of work.  While generally these things accelerate with the constantly improving data from users (and accidents), let's just say we continue at 10% improvement per year.  That get's use to 90% accident reduction in 6 years.  I'll be OK with that.

 

Every addition ads data.  This isn't like a moonshot.

 

Dave

Edited July 1, 2016 by Mallette

[DrWho]

The issue here is the moral responsibility in situations like this.....was the feature ready? How do we not have the death of an individual caused by a rushed financial pursuit?

 

The difference about a human making the mistake is that the morality is much more clear. 

 

The thing is, the rest of the auto industry has been studying autonomy for arguably much longer than Tesla, and they've chosen a much less cavalier approach of slowly and subtly bringing in the automatic aids.

[Mallette]

The issue here is the moral responsibility in situations like this.....was the feature ready?

 

Agreed, and Tesla very clearly informed users it was not ready for them to take a nap.  The guy ignore that, and in a very unlikely situation paid the price.  The rest of us gain from his death.  How many people actually gain from the thousands of similar human caused accidents that happen every year?  At least this one is software correctable.

 

But everyone agrees you can't fix stupid.

 

Dave

[oldtimer]

If an early death was a criteria, then autos and even horses would have been outlawed long ago.  Still, I enjoy driving.  if only everyone else on the road were at least as responsible....

[Travis In Austin]

There is an article in this month's Scientific American, written by THE guy in this field, he founded PATH at Cal Berkeley and has been working on autonomous vehicle technology since 1986.

The article discusses the five levels of AV as set forth by SAE. He says level 2 is currently available commercially. He discusses how the media blows things out of proportion and call thing autonomous when they are not even close.

Level 4 and 5 he says are driverless, and are probably "decades away" on a commercial basis. Apparently pedestrians and rain are a major problem.

Thought it was a great article, trying to find a link.

[Don Richard]

40% improvement of currently available technology works for me, and I don't believe for a minute it will take more than a few years to get to 90%. That is after about 4 years of work. While generally these things accelerate with the constantly improving data from users (and accidents), let's just say we continue at 10% improvement per year. That get's use to 90% accident reduction in 6 years. I'll be OK with that.

 

You don't seem to appreciate the magnitude of the problem concerning making a self-driving car possible. It's quite a bit more complex than making an autonomous vacuum cleaner for the home, for example. Auto manufacturers are just beginning to realize the implications. The lawyers are beginning to get involved. Getting lawyers involved is a sure way to slow progress .

 

 

 

Every addition ads data. This isn't like a moonshot.

 

A moonshot is a walk in the park, computationally, by comparison. The Apollo computers were 2.0 MHz 16 bit processors that used 64k of magnetic core memory. That's it . 64 kilobytes of memory, total. No bulk storage. 3 of them onboard for redundancy.

 

There was no traffic or pedestrians to contend with up there either.

Edited July 6, 2016 by Don Richard

[Mallette]

The Apollo computers were 20 MHz 16 bit processors that used 64k of magnetic core memory.

 

Nope.  The Apollo computers were the most advanced general purpose computers in the known universe and saved the day from the one you described above which was trying to land them on a boulder.

 

Dave

[Mallette]

Many of the naysayers here are harping on "autonomous."  Right now the only denotatively autonomous device on the planet is a human.  One of the scientists quoted above is very right...but he is using the scientific definition of autonomous.  I get that.  But the average educated person doesn't use it that way.  Autonomy in the commonly understood sense was demonstrated over a year ago by the cross country run in which the occupants had to take over only once.  Autonomy, as generally meant, is already on the road.  Full autonomy, as meant by your vehicle leaving your work parking, picking up your child at school, then picking up you wife at work, taking them home and returning to pick you up without intervention is certainly a decade or more away.  My belief, based on experience in the computer industry that once a concept is mooted, is useful, and in demand the development proceeds FAR faster than "experts" predict, is that the vast amounts of run time data already being gathered will generate a Moore's Law effect in the drive towards full autonomy.  My own estimates have been overly conservative so far.

 

Dave

[Don Richard]

 

The Apollo computers were 20 MHz 16 bit processors that used 64k of magnetic core memory.

 

Nope.  The Apollo computers were the most advanced general purpose computers in the known universe and saved the day from the one you described above which was trying to land them on a boulder.

 

Dave

 

 

I was talking about the computers on board the spacecraft itself, like we were talking about the computers for a driverless car. If you have better information on those Apollo devices I would be interested in learning more. I think the processors were from Motorola but not sure about that. (actually used Fairchild ICs)

 

The specs I quoted don't seem impressive today but for back then, not too shabby:

 

https://en.wikipedia.org/wiki/Apollo_Guidance_Computer

Edited July 6, 2016 by Don Richard

[Gilbert]

Level 4 and 5 he says are driverless, and are probably "decades away" on a commercial basis. Apparently pedestrians and rain are a major problem.

 

 

That's consistent with my research on the subject back in 1992 -93...... maybe by mid 2030's at best.  Dave and I'll will likely be pushing up daisy by then, but if we're lucky, a fully autonomous hurst will carry the casket...... though I would prefer a horse and carriage.

Edited July 2, 2016 by Gilbert