Autonomous Trains

[John]

The PTC already blows the horn for crossings, but we’re not allowed to let it. (And it’s incredibly obnoxious)  We can let it start, but we have to finish the sequence. It automatically adjusts when it starts the crossing sequence based on your speed.

Look, the railroad is WAY ahead of all of us on the technology. What we’re doing now is the tip of the iceberg; updates are rolled out incrementally with new features and functionality added each time. The entire physical plant is already mapped in 3D. Every crossing, no matter how insignificant. Every nuance of grade and curvature. Inward-facing cameras capable of live look-in and can see every corner of the cab. Forward-looking cameras and external mics that can pick up conversations between the engineer and someone standing on the ground. Cameras EVERYWHERE on the property. The dispatcher can pan and zoom, as can local managers.

Lawyers and regulators will hash out the liability issues. Four-quadrant gates, crossing closures, grade separation, fencing. You already can’t avoid hitting something, what difference does it make if no one’s in the cab? Someone monitoring the system can look at the camera footage and see what was hit and stop the train if necessary.

Complete mechanical breakdowns are rare. Sure, they happen, but more often than not the train can keep moving with power from the other units. We already have “rapid responders” who are mechanical department personnel staged with vehicles at strategic locations in case a train has problems. Quite often we are able to bring the train to them for assistance. If not, they can be on site usually within an hour or two.

There’s no avoiding it… every excuse you can think of for why it can’t be done is, or will be dealt with in due time!

It sounds like I’m a shill for the company or a fan of automation. I am not; I’m just a realist! The whole thing makes me sick.

Great perspective ..

[Missaberoad]

The PTC already blows the horn for crossings, but we’re not allowed to let it. (And it’s incredibly obnoxious)  We can let it start, but we have to finish the sequence. It automatically adjusts when it starts the crossing sequence based on your speed.

Look, the railroad is WAY ahead of all of us on the technology....
...There’s no avoiding it… every excuse you can think of for why it can’t be done is, or will be dealt with in due time!

It sounds like I’m a shill for the company or a fan of automation. I am not; I’m just a realist! The whole thing makes me sick.

You guys are way ahead of us on this... We barely have the infrastructure in place and PTC is way off in the future still.
Even our FTO sounds nowhere near as accurate as yours seems to be... Ours struggles with grades and performing consistently.

Perhaps that has to do with how much effort the company invests into it.

[nkalanaga]

~Ian:  I agree about trains and crossings, but in this case, I was referring to automobiles on city streets.  The public won't let them run over pedestrians, even those playing chicken with them.

You're right about the trains.  It doesn't matter who, or what, is in control, someone gets in the way, they will be run over.

[learmoia]

~Ian:  I agree about trains and crossings, but in this case, I was referring to automobiles on city streets.  The public won't let them run over pedestrians, even those playing chicken with them.

Well yeah.. In the case of automobiles, I agree.. I think my Rav 4 will stop itself for a pedestrian...  (not 100% sure, I don't want to test it)

~Ian

[Philip H]

I'm doing a lot of work in marine autonomy these days. Grade crossings are complicated - distinguishing a lobster pot float from a runabout from a navigation buoy is tougher.  The Artificial Intelligence folks are working on it, but true ocean autonomy is probably 5-10 years off.

That said Maersk and their aligned competitors are working hard on full autonomy for open ocean containerships in captive service.  As you might guess the Coast Guard is a bit apoplectic about the whole thing.

[C855B]

... As you might guess the Coast Guard is a bit apoplectic about the whole thing.

As they should be. Automation/autonomy is a great concept, but even with advances in AI and so on, it's the 0.001% failure scenarios that remain the problem. When things like piloting ships and trains fail, they fail really bigly, with very large mitigation and recovery impacts.

Then there's the failure paradox - immediate actions to avert pending doom where all choices result in damage, injury or fatalities. So the AI decision tree has to evaluate which option is least bad, including which decision is least likely to result in litigation.

The mind reels.

[Maletrain]

~Ian:  I agree about trains and crossings, but in this case, I was referring to automobiles on city streets.  The public won't let them run over pedestrians, even those playing chicken with them.

You're right about the trains.  It doesn't matter who, or what, is in control, someone gets in the way, they will be run over.

My comment about car jacking was aimed at the automobile part of this thread. 

And, imagine driving through Baltimore where the "squeegee kids" "clean your windshield" at multiple traffic lights while you are stopped, and then get really pushy if you don't hand them some money (and it better not be coins).  Some of those confrontations are already scary. Think about what it would mean if they could just stand in front of the cars until they get whatever amount of "payment" they want.

And, then there are the actual car jackers, who want your car and will kill you for it.

It would be just about impossible to make an AI system that could read the intent of somebody standing in the right of way.  So, the driver would automatically be forced to accept whatever "politically correct" action is programmed into the pedestrian avoidance system.

Trains?  Not such a big deal.  Can't stop them fast enough, anyway.  And they start so slowly that anybody standing in front is going to get shoved, not splattered. 

[Ed Kapuscinski]

So, the driver would automatically be forced to accept whatever "politically correct" action is programmed into the pedestrian avoidance system.

Like not killing someone over replaceable property?

[Maletrain]

And, then there are the actual car jackers, who want your car and will kill you for it.

It would be just about impossible to make an AI system that could read the intent of somebody standing in the right of way.  So, the driver would automatically be forced to accept whatever "politically correct" action is programmed into the pedestrian avoidance system.

Like not killing someone over replaceable property?

No, Ed, "PC" like assuming that the guy pointing a gun at you will only take your property and not maim or kill you in the process or just for the fun of it, even if you don't fight back.  You live near Baltimore, Ed, so don't pretend that you have not read about the attacks and deaths.  Heck, there are kids walking up behind strangers and trying to knock them out with one punch from behind, in the tourist areas.  At one point, the people watching an Orioles baseball game were kept in the stadium for an hour after the game because of the riot going on outside in the "nice" area- not some protest of any sort - just a large number of people acting-out in a mean way.

Besides, just because something is "replaceable" does not mean that its owner can afford to replace it or even live without it. So, until the PC police start promptly giving victims full property replacements when criminals take or destroy something, "replaceable" is not really the issue.

But, thanks for illustrating the dangerous PC assumptions that would probably be programmed into computers that have to make a decision for a human.

As I said, the human needs to be able to make the judgement about what the situation calls for.  A machine will never be perceptive enough.  And, any weakness in a machine making decisions will b exploited by criminals.

[Philip H]

According to the Supplemental Homicide Reports from the FBI's Uniform Crime
Reports, about 15 murders a year involve auto theft. It is not possible to
determine how many of these murders meet the definition of carjacking.

https://bjs.ojp.gov/content/pub/ascii/c02.txt

Serious injuries, such as gunshot or knife wounds, broken bones, or internal injuries occurred in about 9% of incidents.

https://en.wikipedia.org/wiki/Carjacking#Studies

Those are national statistics. They are such low probability that the AI can be trained to account for them.  Ed is spot on.

[Maletrain]

Obviously AI cannot be "tuned" to "account" for probabilities in any manner that reflects the detail conditions of a specific situation.  That is just a false statement.  What you are saying is that, according to your statistics, only a few people would be intentionally killed, but more would be saved by preventing accidental deaths.  That is not "tuning" for specific situations, that is "tuning" to an overall body count.  Quite a different concept than what I am talking about. And, so far, autonomous cars have not been fatality free, either.

Further, national average statistics are of no use in certain inner cities.  Even "tuning" for what city the vehicle is in would be totally inadequate for calculating the odds of malevolent intent for a person stopping a vehicle.  People recognize things about the situations that a machine simply cannot take into account.

Finally, the statistics of today are useless for predicting a future where a new vulnerability is created that will change the statistics.

Autonomous cars are quite a different technical, social and ethical situation from autonomous trains.  So, let's drop this argument about cars and get back to the trains.