And yet again we prioritize the efficiency of vehicles above all else.
In my city adaptive signals have made walking very unpleasant; there is a main road where the primary direction of car travel is orthogonal to the primary direction of walking. The adaptive input only looks for cars, and so a pedestrian can easily wait two or three minutes for them to get a green cycle. In addition, the logic that triggers a walk signal requires a button to be pushed first, so if you arrive at the crosswalk during a green light for cars in your direction you still need to wait for an entire red cycle because they won’t give you a walk during the current signal.
The worst part is, that according to the data the time saved is about 30 seconds, on a few block drive that has easily taken me an hour and a half in the worst case. For a project that cost half a billion dollars this was a piss poor return on investment, and particularly inappropriate for a busy dense area.
Traffic controller coordination and traffic flow optimization is a very hard problem that many have spent their entire careers working on.
Every intersection acts as it's own problem and also as a node in a much greater problem, then once You think You have everything running smoothly, something like an accident, emergency vehicle or train crossing messes everything up. Then there is the human factor, drivers and pedestrians do strange things that are unimaginable until they do them, and for some reason, they do these "odd" things at the worst possible time.
I have always thought it would make a great basis for a puzzle/strategy video game.
One issue that arises from using AI derived sequencing models, as mentioned in the article, is the difficulty in determining how the program weighs conditions in order to develop it's controler program. This is particularly important to know in traffic control situations for legal and liability reasons.
I wonder — has anyone thought of applying MCTS to this problem, where you create as many possible different scenarios and variations as you can, and then let the system figure out what the best responses might be?
In my city adaptive signals have made walking very unpleasant; there is a main road where the primary direction of car travel is orthogonal to the primary direction of walking. The adaptive input only looks for cars, and so a pedestrian can easily wait two or three minutes for them to get a green cycle. In addition, the logic that triggers a walk signal requires a button to be pushed first, so if you arrive at the crosswalk during a green light for cars in your direction you still need to wait for an entire red cycle because they won’t give you a walk during the current signal.
The worst part is, that according to the data the time saved is about 30 seconds, on a few block drive that has easily taken me an hour and a half in the worst case. For a project that cost half a billion dollars this was a piss poor return on investment, and particularly inappropriate for a busy dense area.