International Railroad Discussion > Driverless Ore Trains in West Australia

Rio Tinto continue to trial driver-less ore trains in West Australia.


https://www.railpage.com.au/news/s/rio-tinto-operates-first-fullyautonomous-test-train

Mike

Michael McInerney
Marion, South Australia,

So this route has zero level crossings and zero trespassers?
Otherwise it's all a matter of time until someone is killed or injured.
Does this mean the personnel in the Ops Centre can be held liable?

In the video there was a crossing. This is just like flying a Predator drone on the other side of the world, and using the missile to take someone out. This is not a bad idea in the long run.

A high proportion of Rio's mine haul-trucks are driver less, run from a control center in Perth 1400km to the south, Komatsu software

The Pilbara is sparsely populated, virtually all of the mine workers/train drivers are FIFO from Perth, single driver operation, high wages and overhead

http://www.abc.net.au/news/factcheck/2013-10-30/are-train-drivers-in-the-pilbara-the-highest-paid-in-the-world/5029012

Far cheaper to have train operators in a control room in the south. Rio have expended over US$500mil on the driverless train project to date, they are serious with regards to reducing their cost structure

PHall Wrote:
-------------------------------------------------------
> So this route has zero level crossings and zero
> trespassers?
> Otherwise it's all a matter of time until someone
> is killed or injured.
> Does this mean the personnel in the Ops Centre can
> be held liable?


Trains with engineers strike vehicles and trespassers all the time. This doesn't make that any better, but I don't see how it makes things worse.

Dave

DavidP Wrote:
-------------------------------------------------------
> PHall Wrote:
> --------------------------------------------------
> -----
> > So this route has zero level crossings and zero
> > trespassers?
> > Otherwise it's all a matter of time until
> someone
> > is killed or injured.
> > Does this mean the personnel in the Ops Centre
> can
> > be held liable?
>
>
> Trains with engineers strike vehicles and
> trespassers all the time. This doesn't make that
> any better, but I don't see how it makes things
> worse.
>
> Dave

at least here in the states, EVERY rail can give dozens of examples of how an alert crew stopped a train BEFORE there was an accident that a crewless train may NOT have stopped for.......

66

goneon66 Wrote:
-------------------------------------------------------
> DavidP Wrote:
> --------------------------------------------------


> at least here in the states, EVERY rail can give
> dozens of examples of how an alert crew stopped a
> train BEFORE there was an accident that a crewless
> train may NOT have stopped for.......
>
> 66

Australia, especially the Pilbara is not the States

You will have also noted the article above (admittedly from a green/left workers collective), the 12 hour shifts and single crew operation

I believe that the US require 2 crew members working shorter shifts?

jmt Wrote:
-------------------------------------------------------
> goneon66 Wrote:
> --------------------------------------------------
> -----
> > DavidP Wrote:
> >
> --------------------------------------------------
>
>
>
> > at least here in the states, EVERY rail can
> give
> > dozens of examples of how an alert crew stopped
> a
> > train BEFORE there was an accident that a
> crewless
> > train may NOT have stopped for.......
> >
> > 66
>
> Australia, especially the Pilbara is not the
> States
>
> You will have also noted the article above
> (admittedly from a green/left workers collective),
> the 12 hour shifts and single crew operation
>
> I believe that the US require 2 crew members
> working shorter shifts?

Not shorter, 12 hours is legal.

How do they handle problems with the crewless trains out in the middle of nowhere? What do they do if a locomotive (s) die on the road, or a drawbar breaks/is yanked out, or air brake problems?

lynnpowell Wrote:
-------------------------------------------------------
> How do they handle problems with the crewless
> trains out in the middle of nowhere? What do they
> do if a locomotive (s) die on the road, or a
> drawbar breaks/is yanked out, or air brake
> problems?

So, having some knowledge on the tech behind it, here's my non-insider view:

The trains are operating over a grade profile that is relatively even without a large variance in conditions or other factors that effect operability. Instead, they're 3-units on the head and a fixed length with consistent load outs that have no real variance as they are unit trains. With that in mind, the physics of the train's movement over the line is pretty easily predictable based on past experience. After all, total variance in load per car for a unit train is <1% in general (got this from a VALE white paper in Brazil), and the loading systems are largely the same in the Pilbara). Furthermore, the unit on the head end is using both a camera and a photo-recognition neural network.

This neural network is effectively a computer program that modifies itself over time based on inputs. Take it over the road enough times, and it will learn what conditions of the train, track and other elements are normal and based on watching the logged inputs of the engineer it will learn how to adapt. With that, the train is able to recognize obstacles, broken rails, lineside obstructions and other things that may affect operations.

In the end, the tech itself is fairly basic and off-the-shelf. The cameras can be bought easily, the computers are nothing special, and neither is the communication tech. The secret sauce here is a fairly straightforward program that has been trained over thousands of trips and now has a very good understanding of what to do in many situations, and given that can easily adapt and/or alert a remote operator to issues.

THAT SAID:

This isn't the same as running a mixed-manifest train with two sets of helpers in the snow over Tennessee Pass or something. This is the self driving equivalent of having a Prius drive around a closed test track that is flat with long sweeping curves at low speed. It ain't rocket surgery, folks - but the more difficult lines are waaaaaay more difficult.

BUT

The tech is advancing rapidly and I would expect to see this over time become more prevalent and eventually make its way to the rest of Australia (for the long super freighters first) and then Brazil and the US in the next 20 years.

I need to go snap photos of manned trains while I can...



Edited 1 time(s). Last edit at 10/05/17 13:55 by zorz.

zorz Wrote:
-------------------------------------------------------
> BUT
>
> The tech is advancing rapidly and I would expect
> to see this over time become more prevalent and
> eventually make its way to the rest of Australia
> (for the long super freighters first) and then
> Brazil and the US in the next 20 years.
>
> I need to go snap photos of manned trains while I
> can...

I would caution against a statement like "the tech is advancing rapidly."

https://www.trainorders.com/discussion/read.php?1,4109858
https://www.trainorders.com/discussion/read.php?6,4109861

Organizations have a habit of "forcing success" of a project when massive amounts of money have been spent. I suspect Rio Tinto won't (and shouldn't) disclose the technical nature of this project, but I suspect some "simplifying" of the software happened while some hardware was added (extra sensors, more manual interventions from remote operators, etc.) to get themselves back on track.

You've already mentioned the strict parameters of the ROW (unit train, "predictable" terrain, etc.) for this particular application. And you are correct that the basics of neural networks have not changed for several years.... so "advancing rapidly" has to come from somewhere else. Instead of machines doing "driver assist," maybe the remote operators are "human assist" of the machine? :-) Not so different from People Movers at airports?

Keep in mind that data obtained from "machine learning" can't be transplanted to a dissimilar machine, or dissimilar environment - you might know that already, but an average TO reader might not know that. So if Rio Tinto decides to buy EMD instead of GE, the EMD has to learn from scratch, even in the same environment, unless the sensors and behavior of both units are near-identical.

I'd refrain from making a lofty prediction of "20 years from now." Let's wait and see what REALLY is going on.

zorz Wrote:
-------------------------------------------------------
> lynnpowell Wrote:
> --------------------------------------------------
> -----
> > How do they handle problems with the crewless
> > trains out in the middle of nowhere? What do
> they
> > do if a locomotive (s) die on the road, or a
> > drawbar breaks/is yanked out, or air brake
> > problems?
>
> So, having some knowledge on the tech behind it,
> here's my non-insider view:
>
> The trains are operating over a grade profile that
> is relatively even without a large variance in
> conditions or other factors that effect
> operability. Instead, they're 3-units on the head
> and a fixed length with consistent load outs that
> have no real variance as they are unit trains.
> With that in mind, the physics of the train's
> movement over the line is pretty easily
> predictable based on past experience. After all,
> total variance in load per car for a unit train is
> <1% in general (got this from a VALE white paper
> in Brazil), and the loading systems are largely
> the same in the Pilbara). Furthermore, the unit on
> the head end is using both a camera and a
> photo-recognition neural network.
>
> This neural network is effectively a computer
> program that modifies itself over time based on
> inputs. Take it over the road enough times, and it
> will learn what conditions of the train, track and
> other elements are normal and based on watching
> the logged inputs of the engineer it will learn
> how to adapt. With that, the train is able to
> recognize obstacles, broken rails, lineside
> obstructions and other things that may affect
> operations.
>
> In the end, the tech itself is fairly basic and
> off-the-shelf. The cameras can be bought easily,
> the computers are nothing special, and neither is
> the communication tech. The secret sauce here is a
> fairly straightforward program that has been
> trained over thousands of trips and now has a very
> good understanding of what to do in many
> situations, and given that can easily adapt and/or
> alert a remote operator to issues.
>
> THAT SAID:
>
> This isn't the same as running a mixed-manifest
> train with two sets of helpers in the snow over
> Tennessee Pass or something. This is the self
> driving equivalent of having a Prius drive around
> a closed test track that is flat with long
> sweeping curves at low speed. It ain't rocket
> surgery, folks - but the more difficult lines are
> waaaaaay more difficult.
>
> BUT
>
> The tech is advancing rapidly and I would expect
> to see this over time become more prevalent and
> eventually make its way to the rest of Australia
> (for the long super freighters first) and then
> Brazil and the US in the next 20 years.
>
> I need to go snap photos of manned trains while I
> can...

Zorz: While you may be familiar with the tech behind this software, I would like to ask if you're familiar with this part of Australia that you can judge the running profile of the land and grades over this area. I often read of comments out of the USA that this will not be relevant to their operations because of implied beliefs that Australia is "flat". There is only one Tennessee Pass, one Donner Pass, etc. but the proportion of these or similar such grades is minimal to overall running of remaining track in your country.

First, I would like to dispel the common belief of people who have never been there, that Australia is flat. This area does cross mountains and operations are relevant to technology which can and will be applied to your country. Grades are grades and you do not need to have the Rockies or canyons around the tracks to make those grades more significant. If you think your deserts or prairie cannot have this applied to it you're mistaken. The mining railroads in the Pilbara do lead the advancement of tech for world railroads which are commonly applied to American operations eg. electronic braking, DPU where the world record for train length was recorded, and now driverless.

Second thing to point out is you are missing the real reason why this may have difficulty of application to your country; that is your Union strength. Your "engineers" (a term which is incorrect - they're "drivers") are heavily unionised. The Pilbara drivers are not. In fact, as has been pointed out, they earn executive salaries pushing $200,000 pa with no union presence. It follows that there will be little issue in modernising their operations which I believe would not be the case in your country. Rio has pumped a lot of effort and dollars into this tech. They had deferred their investment during the course of my time of working in the region which at that point (2009) their spend was $200 million. Now, apparently, the tag is over $500 million and they have succeeded in running the train. It certainly seems they want to make it work and the precedent has been set in driverless trucks in the pit. (aside: those poor guys were earning as much as $135,000 pa).

I believe that Rio tested this without DPU. They are commonly committed to running single sets (165 cars of 100 tonnes). It is BHP-B who runs the DPU sets (and that changes every time some snotty nosed new MBA comes along and switches the operation back to what they used to do but "are going to do it better this time").

lynnpowell:
To answer your question, there is no doubt that there will be local technicians and drivers up there for any fail incidents. They will likely be in Paraburdoo, Tom Price or at the coast. The entire line has an access road beside the tracks so attendance can be made at any point. This is no difference to what they've done in the past when there is a drawbar or air brake. I am not certain if they are there yet with the implementation but electronic braking is supposed to significantly reduce brake issues. Electronic failures, one would expect, will be far less common than air line failures.

Rick

There is currently a significant push in the maritime world for crewless (autonomous) ships for transoceanic voyages, as well as for warships. These vessels are under design, they are not fantasy. They run sea-buoy to sea buoy, and there are even autonomous mooring systems under design as well. "crews" will be remotely located, in most cases with cameras, remote displays of the electronic charts the ship is navigating to, remote radar screens, VHF/UHF radio access, etc. Tankers will include the ability to monitor cargo and associated systems remotely....as noted above pilot-less fighter jets are a military goal...as we already have drones.

http://www.rolls-royce.com/~/media/Files/R/Rolls-Royce/documents/customers/marine/ship-intel/rr-ship-intel-aawa-8pg.pdf

Trains may be the easiest thing to automate as they run on a fixed guideway and save for derailments and someone tampering with switches, it may be pretty bulletproof. Having said all that, I think transport executives are nuts to entrust millions and millions of dollars in cargo and equipment to a data link - but look at how NASA was able to deliver the Pluto probe way out into the solar system, controlling it from literally billions of miles away... including troubleshooting a last day camera problem and fixing it - and putting it right on target.

Bob

I think NASA spent $3 billion dollars on that one. No railroad would pour that amount of money into driver less trains.

I'm dying to railfan this operation. With a remote drone, of course. 

tomstp Wrote:
-------------------------------------------------------
> I think NASA spent $3 billion dollars on that one.
> No railroad would pour that amount of money into
> driver less trains.

No railroad has millions of miles to travel either.