Slashdot Mirror
Making Trains Run on Time (economist.com)
Plamen Angelov of the University of Lancaster, in Britain, has an idea that he hopes will make train delays rarer. From a report: Often, Dr Angelov observes, the problem is not the inefficiency of operators but the behaviour of passengers -- the "platform-train interface", to use railway parlance. When trains arrive, passengers crowd around the doors waiting to board, restricting the flow of those getting off. When they are about to depart, people often hold doors open, delaying that departure. (A recent study by Japan's Railway Bureau found that passengers attempting to board trains after their scheduled departure times accounted for almost 50% of delays.) Passengers also frequently stand too close to the carriages for safety. Waiting for malefactors to move back behind the yellow safety line on a platform might hold a train up for less than a minute. But over the course of a journey those minutes add up. Even a slight delay is compounded if it causes a train to slip out of its running order and be held at a subsequent station, or be required to follow a slower service.
Dr Angelov thinks that applying artificial intelligence to the problem might help. And that is what he and his team are doing. Using images from the cctv cameras already mounted in carriages and on platforms, their system employs algorithms that have been trained to detect objects such as people, luggage, pushchairs and bicycles. It then measures the movements and positions of these objects relative to areas such as the train doors or the yellow safety line and uses this information to predict problems. The cameras in the carriages detect how busy particular doors are getting as passengers leave their seats and gather next to the exits when the train approaches a station. At the same time, the station cameras monitor the numbers waiting for the train to arrive, whereabouts they are standing along the platform, and how encumbered they are. The two sets of data can then be compared, providing warning of likely areas of congestion. This permits passengers -- particularly those on the platform -- to be directed to doors that will be less busy.
Dr Angelov thinks that applying artificial intelligence to the problem might help. And that is what he and his team are doing. Using images from the cctv cameras already mounted in carriages and on platforms, their system employs algorithms that have been trained to detect objects such as people, luggage, pushchairs and bicycles. It then measures the movements and positions of these objects relative to areas such as the train doors or the yellow safety line and uses this information to predict problems. The cameras in the carriages detect how busy particular doors are getting as passengers leave their seats and gather next to the exits when the train approaches a station. At the same time, the station cameras monitor the numbers waiting for the train to arrive, whereabouts they are standing along the platform, and how encumbered they are. The two sets of data can then be compared, providing warning of likely areas of congestion. This permits passengers -- particularly those on the platform -- to be directed to doors that will be less busy.
The solution has existed for 32 years. Automate the trains.
It works perfectly in Vancouver, not a single accident due to the computer. No, everything is certainly a suicide/accident when someone falls into the track as the train comes into the station. Computers have better reaction time than humans, which is why the track sensors work best when people aren't creating a crowding situation.
What can put an end to all delays, even in the automated system is PED's, or Platform Edge Doors. However most rail systems are a mish-mash of different rail cars of different door positions, so for Vancouver, this doesn't work because the 6-car mk 1 train, 2x2 car Mark 2/2.5 and 4-car Mk 3 train's all have different door spacing. Platform Edge doors would have to be over engineered to work like venetian blinds. eg, by being pulled down when the train is about to open the doors.
Computer AI here simply is a technical solution to a problem that has already been solved. The Skytrain runs "on time" because it has no schedule, there's a train every 75 seconds at absolute perfect conditions, and every 6 minutes during slow off-peak times. How this can be applied in other places requires a signalling upgrade, removing the driver from the calculation, and then running the trains by forcing the doors to close a-la elevators, where they stay open for a minimum dwell time, and then attempt to close a second time, where the alarm will sound. If they hit an obstruction they re-open completely only once. Any attempt to hold the door open will alert station attendants or transit police that someone is holding up the train and they can be ticketed if they hold the doors while the alarm sounds. Issue enough tickets and people will quit being stupid.
The safety problem can only be solved by not cutting costs in the first place. Build PED's from the begining, build platforms big enough for an entire train to load and unload (use the Spanish solution if that's what's efficient, open the exit doors first, and then the entrance doors, setup turnstiles to only allow entry or exit.