Possibly late here, but why aren't more / all trains automated? The route is static; Speed should be easy to determine; AFAIK they aren't mechanically unreliable enough to warrant a full-time mechanic on-board; LIDAR and image recognition can detect hazards on the track.
It seems silly that there have been such strides in automating automobiles before trains.
Most human-driven trains use coloured circular lights for signalling. Determining the presence of a coloured circle in a known location is about the simplest computer-vision problem known to man, so we could probably practically solve this cheaply tomorrow with an iPhone on a suction cup.
Maybe there are legal issues in many/most jurisdictions getting in the way, but isn't saving "roughly ... the railroads’ total capital spending in a single year" worth a change in regulations?
Why? The regulations are there for a reason. In Netherlands the rail safety system has to be heavily improved before automation is possible. What apparently is the cause of the outdated safety system is that the system was (when built) ahead of its time, but nowadays other systems are much better.
They're now going to test with automated trains on specialized goods-only train lines.
In this case the regulation is there to ensure safety. If improvements cost a lot it'll have to be spread out over multiple years, or e.g. not automate away drivers. What's to me seems utterly strange is immediately jump to just allowing such changes and assume regulation is the problem.
No one has figured out how automation can save a flight in the air if the automation fails. Remote control and communication can also fail. Ground based trains or vehicles can just stop on the ground.
There was a thread about it on reddit, please go and find it. The trains do not need LIDAR or image recognition. Going in straight line on rails is simple. You don't need to even brake (useless, stopping distance is way too big). What happens in stations and what preparations are done, that cannot (e.g. cost efficiently) be automated. Metro on other hand is mostly automated (see Europe)
How much savings would you expect from this? And whose job should be automated?
If you think about replacing one driver per train then probably the cost of automation is higher than his salary. Instead of paying one man a salary you would be forced to pay a maintenance contract to some IT/automation company that would charge you exorbitant rates just because this is industrial automation and you have no idea how to operate it.
some IT/automation company that would charge you
exorbitant rates just because this is industrial
automation
This is shenanigans. This is the kind of game Enron [0] played, gouging prices for irrational reasons, until eventually outside intervention dismantled their operation, to discover actual costs.
A bigger benefit than cost savings with the workforce would be higher reliability of service. With the Tube in London delays with driver changes often cause the trains to be held up significantly.
TfL don't suffer the external costs but the costs of one driver are insignificant compared to five minutes for 500 passengers on a train, day in day out.
Some airports have automated, unmanned trains or trams that take passengers from terminal to terminal within the airport.
The example is illustrative because those are usually either elevated or underground, with virtually no possibility of people being on the tracks, and no crossings with other traffic.
Somewhat unrelated but I have wondered why people are assuming the first use of automated vehicles will be for personal transport on a public road.
It seems a lot simpler to have a part of the road network dedicated to fully automated vehicles. And instead of immediately transporting people (which carries a high risk) it seems a lot more sensible to initially focus on the transport of goods.
The initial set up would be quite expensive, but once you no longer need to worry about human drivers and don't need to worry overly about a million to 1 chance for things to go wrong, you should be able to use the road a lot more efficiently.
The Copenhagen metro is fully automated, but it was built like that from the start. This means that there's no metro driver unions, and the metro stations were built in a way, where it's generally hard to get unto the tracks. It did have some issues to begin with, but it works flawlessly today.
On some lines where the train is automated, the metro operator still keeps someone on the train who has training for an emergency situation, I.e. Evacuating passengers.
So the saving is their salary compared to the train driver's salary.
On big trains, we expect the driver to spot dangers like landslides, fallen trees. These happen extremely rarely, so the driver would be bored and lose concentration without having to drive the train -- within the strict parameters set by the computer, on the most modern high speed lines.
Specifically for the London Underground there's a lot of union opposition. Some lines have mostly-automatic operation (everything except opening/closing the doors) and keep a driver around for safety reasons.
In some places (eg. London), automation is totally possible, but is resisted so hard by workers unions that all other workers on the railway would go on strike until the train operator promised to rip out automated systems and return to manual operation.
On some lines, all the driver does is open and close the doors. Even getting to that level of automation caused lots of strikes.
When it's their whole livelihood, all they've ever known, and their families depend on it? You can hardly blame them. We need to give people other options. A lot of people who work these sorts of jobs don't have the savings that would allow them to retrain. If they lose their jobs, their families are in deep, deep trouble. This is where something like a basic income comes in. It allows workers a floor in the case their jobs get automated away. Gives them a chance to pick themselves up and find something else to do with themselves. Right now, they don't have that, so they -- understandably -- fight tooth and nail against being automated away (to everybody's detriment).
It seemed that way, then I got a train driving simulator. Buy a train simulator (Train Sim 2018 is what I have) and try to drive a train to the timetable. Then play with the conditions (raining, autumn, midnight). Then realise that you need to do this without the map at the bottom of the screen. Then try to understand all the rules and regulations that a train driving simulator doesn't simulate, like how to walk down the track safely (old training videos are on YouTube are very interesting).
Also, in real life, train drivers are responsible for the safety of the passengers on the train, and also have to deal with the aftermath if anyone ends up being in front of the train.
If I was a train driver, I would like to be paid enough to cover all of that!
The average tube driver's pay is ~42,000 pounds. That converts to $56,000 a year. That's not a very high income. That's solidly lower middle class depending on where you're living and whether you're single or the primary wage earner for your family. London is one of the highest cost of living areas in the world.
A family living in London with that income -- even if the second parent is also working -- is unlikely to have any savings. Unless the second family member has a high income information, tech, or finance job.
The Lille (north of France) metro/subway/underground is totally automated - there aren't even any drivers for safety reasons. I haven't heard of any major issues and from using it personally it's always been flawless.
In the US, most of the train technology is old and not cross compatible, making each installation difficult and expensive. Looking at the NY metro area trains and driveways, there are frequent mechanical issues, both on the trains themselves and on the train instructions such as switch problems, sometimes necessitating reversing back a few stations in order to switch tracks.
* Many trains are old (at least in the UK), and tend to be very slowly upgraded, so the trains on the network were built before it was considered practical to automate them.
* Signalling systems are old, UK example: the warning system (AWS) is built on electromagnets, clever and cheap, but not enough for automation.
* While it's possible to design a system that works well for a normal journey if nothing goes wrong, when things do go wrong you'd need a person (for example, I once was on a train and a door was stuck, the driver got out and fixed it so we could get off), also what about level crossings?
* Once you're paying someone as a failsafe, then they already know how to drive the train, then the extra automation is just extra cost, except when the railway is busy (so London Underground has partial automation, as cost of driver + automation is worth it to run all the extra trains).
* Places with automated systems tend to be isolated (see the DLR, it's isolated from mainline trains, and also has no level crossings).
* Maybe a way to keep people to blame if things go wrong, if an automated train goes wrong, who is to blame?
* Who in their right mind would want to lose their job? From what I understand people enjoy it, and it pays well, and it's not exactly a skill that can be transferred (while in an ideal world someone could get another decent job, the fact is we're not in an ideal world, so if someone loses their job they're going to have a hard time and not have a way to magically get something new, unless they're lucky).
Level crossings aren't an issue. It's easy to detect the presence of large objects blocking a level crossing if the barriers fail to come down and signal to the train to stop before it reaches the crossing.
Drivers tend to earn much more money than guards do because the work is more skilled. If all you're doing is opening doors then you can be a min wage worker who is easily replaceable and thus not very unionisable. So there is benefit even in systems that still need guards.
The real issues are deeper. Yes, union opposition is a big one, but there are technical problems too. Full moving block signalling i.e. trains that track how far they are from other trains and know to slow down or speed up depending on distance was being discussed in the 1990's but the only implementations anywhere are on metro lines. For above-ground heavy rail there are no implementations anywhere in the world, as far as I know.
As observed by the OP this cannot be entirely a technical issue. If Google can make a driverless car that can navigate California, a driverless train is surely far easier. The real problem is that train automation is not a problem that attracts the best engineers. Google can hire out entire academic departments of the top machine vision researchers in the world to work on self driving cars because self driving cars are something that appeals to the whims of billionaires and techies alike. Also because the market size is effectively limitless.
But when was the last time you heard about an automated train startup? When did you hear about someone leaving their job at Facebook to go work on train automation? Trains do not appeal to rich billionaires or American tech workers in the same way that cars do, so funding for them comes entirely from the capex budgets of mostly government funded rail operators, and is spent on a handful of large engineering conglomerates. There are I think only about 4 companies in the world that can make automated train systems of any kind. Each engagement is a massive activity that always requires very large and expensive customisations to the base system. Because there are so few competitors they're huge public contracts and it only takes one or two to be ruled out or refuse to bid and you're down to a single potential supplier.
Projects like this can fail and when they fail they fail very expensively and with huge political fallout. They are the epitome of huge expensive government IT projects. The Jubilee line automation project started installing equipment in 2006 and didn't activate until 2011, with loads of line closures and problems that made major news.
There are also some more direct technical issues. Trains have huge stopping distances. They need to know where other trains are blockages are far in advance of being able to physically see them. They also spend a lot of time in tunnels. Therefore camera based approaches like what self driving cars use are not so trustworthy. This means upgrades to the track so all trains can know where they are at all times, and to get the benefits means reliable communication from a central control room rather than all trains autonomously deciding things like how fast to go (remember they cannot see slow trains on the track ahead). These things in turn mean line closures and engineering works, which are very expensive. And remember there's no money for upgrades because they're all government subsidised.
There's an article about some of the issues and the smoother rollouts London is seeing these days here:
Well... The Victoria Line was a form of automatic since it was made, so it seems to be more practical to have a new system created as an automatic one rather than upgrade an old system. Even with the new systems on the Underground, the driver still needs to know how to drive the train, for when things go wrong, and unlike a car, other trains can't move around a stuck train.
I guess one of the big problem with automating trains is that while it's easier to automate the normal operation of a train, there is less tolerance for any issues (a train stopping blocks all other trains on the line, there are also a LOT of people on a train at a time, stopping distance is huge, etc...).
> But when was the last time you heard about an automated train startup? When did you hear about someone leaving their job at Facebook to go work on train automation? Trains do not appeal to rich billionaires or American tech workers in the same way that cars do, so funding for them comes entirely from the capex budgets of mostly government funded rail operators, and is spent on a handful of large engineering conglomerates.
Cars can stop much more easily than trains. Trains can take one or two miles to stop. That's hard to do with LIDAR.
There have been failures of automation. BART was fully automated. Then one train misread a speed command from a trackside gadget, and sped up to over 60 MPH when it was supposed to be slowing down to 20 to prepare to stop at a station. Unfortunately, the station was the end of the line (Fremont), and the train ran off the end into the parking lot.
On the other hand, that was in the 1970s. Some new systems have been fully automated. The Las Vegas Monorail is fully automated. At the end of the line, it asks you to exit, but there's nobody there to make you, so I didn't. Then it runs out on this dead end to switch over to the other line, where it will fall off 20 or 30 feet if it doesn't stop in time. It's a bit scary, but it worked...
I guess that Train technology is old enough that regulations about drivers, signals, and safety surely assume that there is a phisical person in charge of driving.
Maybe the first step towards automation is beign able to drive the train remotely. Technically feasible; but is it legally feasible?
It seems silly that there have been such strides in automating automobiles before trains.