Train drivers do more than just drive a train. They also have to deal with dropping off and picking up cars from the consist. They also need to monitor track conditions that aren't necessarily easy to pick up from track circuits, such as deep snow (https://www.youtube.com/watch?v=Yja2VmZOfdA shows you how some trains deal with snow).
In general, the costs of upgrading track to be able to support autonomous driving aren't worth the savings you get from eliminating driver. The main counterexception to this is Western Australia, where everything of value is 1,000 miles away from where anybody lives (and is transported by trains to the coasts for export), and so railroads face tremendous costs actually getting their drivers to their trains. Unsurprisingly, the mining companies in Australia who own/run their own railroads have invested heavily in autonomous trains.
The massive dump-trucks and drill rigs are going autonomous, with anywhere between 10% and 100% (depending on the site) of the heavy vehicles are now driverless.
The routes can be easily predefined, they are relatively short and you can add additional beacons / markers that make things much easiser.
I think the problem with trains outside of say closed loop underground trains that you have a lot of tracks in varying conditions, you can have track abstractions you have a multi agent environment with crossing and that at least as far as cargo trains go the engineer does more than just drive the train they handle many more operations.
Mine site are open cast, mine large tonnage and have long mine lives justifying the large initial investment.
Australian mining operations are also very dry, which helps a lot. Rain introduces a lot of variables.
Large scale automation is particular to the Australian mining industry at this point.
South African deep level gold mines have been working on this for many years with little success, and I think they have mostly given up.
> Less than a month after the system's opening, on October 2, 1972, an ATC [automated train control] failure caused a train to run off the end of the elevated track at the terminal Fremont station and crash to the ground, injuring four people.
In a bit more detail, they had trackside something-or-others that gave speed information to the train. One train mis-read the speed instruction as calling for 67 MPH instead of 20-something, at a time that it was heading into a dead-end station. When that happens, if there's no human there to say "wait a minute, that's not right", the mistaken command gets obeyed, and the train crashes.
Now: Does that happen more often with automated systems? Or less often? Arguably, less. But to this day, BART has people in the cabs of their trains, at least to monitor the automated system.
Something like the DLR normally operates autonomously, there's a member of staff trained to drive it on every train but it's actually slower and more dangerous if they have to. But that's a completely grade separated system, cows on the DLR is about as likely as cows up the Empire State. And still the first train every morning is manual because maintenance teams are forgetful and a passenger sat at the front doesn't want a metal pry bar in the face when some idiot left it on the track despite checking it off as recovered.
Case in point: https://en.wikipedia.org/wiki/2008_Stud%C3%A9nka_train_wreck - train hits bridge at 90 kph instead of 135 kph; a 2/3 speed, but less than 1/2 the kinetic energy.
(or maybe tech has caught up in the last few decades and they detect obstructions from further away, but that would spoil the story)
http://divine.vic.gov.au/lifestyle/health-and-wellbeing/road... has an account of a collision with a road train:
"The road train driver had slowed down to make a left-hand turn at a crossroads and Scotty had slammed into him at over 100 kilometres an hour. According to the autopsy, Scotty was over the legal alcohol limit. The driver of the road train kept going once he had completed the corner. I was later told he thought he felt a shift in the load as he slowed the vehicle to negotiate the corner."
Anyone wanting to make autonomous trucks perfect should remember the bar is actually pretty low...
https://en.wikipedia.org/wiki/List_of_automated_urban_metro_...
If you are talking about intercity trains instead of metro systems - although metros typically handle many times the number of passengers - China is currently trialling driverless technologies on its Fuxing class HSR on the Beijing-Zhangjiakou route with expected entry into service by 2022.
I think unions will get in the way as much as they can, unfortunately. As an example, a 'fireman' rode along on diesel and oil trains LONG after they were rendered obsolete by the fact that there was no longer any coal to shovel.
https://www.perthnow.com.au/business/mining/derailed-bhp-tra...
> "The driver had stepped off the train..."
It wasn't a driverless train (except in the technical sense that the driver wasn't onboard) and the quote comes from CFMEU (for non-locals, that's the Construction, Forestry, Mining and Energy Union, a massive workers' union) which has a vested interest in pushing back against automation.
Edit:
So it looks like even though they've had the technology going for a while, Rio only just delivered their first 'real' loaded autonomous ore train last year : https://www.riotinto.com/ourcommitment/spotlight-18130_25692...
More on the AutoHaul system: http://www.calibregroup.com/projects/project/autohaul-projec...
Apologies for the mixup
It's not. Always assume that whatever limit can be exploited to maximize economic return is pursued to the maximum extent by everyone involved at nearly all times. If you do so you will be right more often than you are wrong. An occupation that is conceptually "easy" is made demanding through optimization.
Heavy freight trains that navigate hilly or mountainous terrain can rip themselves apart when they are not operated with precision. Passenger train operators can be held to schedules where deviation in either direction is measured in seconds regardless of weather, faults, mistakes, uncooperative passengers, etc.
When large amounts of capital are involved always assume that the expectations will be extraordinary and, therefore, difficult to automate.