[font=Verdana, Arial, Tahoma]Hey,

i'm wondering how AI can be used in the future to solve urban problems like increasing traffic congestion. The reason is that I'm working on a research project on smart cities and I thought about an interlinked semantic system in which AI can be used for traffic coordination and the prevention of traffic jams.

I'm also thinking in terms of information sharing between traffic participants to predict traffic flow (which might include thoughts like robotic cars for urban areas). My assumption is that by let's say 2050 we are all so interlinked so that real-time sharing of information is easily possible and so any AI-system can be fed with all the data necessary.

The major question I'm asking myself: is it possible to develop e.g. a semantic system that can analyse all available data (gps positioning etc.) so that it can detect evolving traffic jams and inform all traffic participants who might be affected by that jam? How can AI be used to understand human mobility behavior? If you can think of any solution, just share your thoughts![/font]

[font="Arial"]This kind of problems is most of the time modeled through multi-agent systems: try Googling "multi agent systems traffic", you will find myriads of studies. AFAIK they aren't much semantic in this sort of modelization yet.[/font]
[font="Arial"]
As a side note, Thales has recently announced a new project called "Terra Dynamica" which seems to be close to what you are looking for: http://www-poleia.li...ca/accueil.html (gtranslate it)

http://www.kylotonng...com/serious.php :
The Terra Dynamica project refers to the dynamic animation of the city, of its activities and actors, and appears as a complement of the works already conducted within the Terra Numerica project regarding urban modeling and visualization, with the objective of modeling “life inside the city”: its inhabitants and crowds, its vehicles and traffic.
The Terra Dynamica project aims at making available a solution of representation of the animation of the various actors in the form of components describing and animating the various actors of the city and making them inter-operate in the 3D urban universe, while integrating the temporal aspect.
The objective is to set up the animation elements allowing the creation of a “living virtual city”. This “living virtual city” provides a unique context (unified and shared) of representation and coherent visualization allowing to address several complementary operational fields such as the conception of the city and of its services, the managing of operations or of urban crises, and the new services aimed at helping the citizen.
[/quote][/font]

I thought I read that in some cities traffic management is already controlled via AI. Like the cameras on the lights detect heavy flows of traffic and adjust light timings accordingly.

Depends on what you define as "solving" the problem.

For practical purposes, sure; there are already very intelligent systems in place in various cities etc. which control things like traffic flow. Power and oil distribution networks are also computer-controlled. There are surely many other systems like this with similar characteristics, which I'm simply not aware of.


Mathematically speaking, though, there is no known closed-form solution for such complex flow problems. Even fairly trivial fluid dynamics problems are viciously hard to solve in the general case, c.f. Navier-Stokes etc. We can certainly arrive at improvements and get approximations to a solution which are satisfactory for practical purposes, but I doubt AI will truly "solve" anything of this nature in the foreseeable future.

Terra Dynamica sounds really interesting, thank you very much for the hint! Most multi-agent systems are fed with data from sensors, traffic lights etc. but it would be interesting to have a semantic system and use direct data from traffic participants. Major problem will be ensuring privacy because e.g. using gps data from mobile phones can be "re-tracked" even if the name of the person and other details are deleted. A person going from A to be every day in the morning and from B to A every day in the afternoon obviously travels from home to work. Thus, data can be reassigned to people. So using private data would definitely be highly efficient but there needs to be a solution to the privacy aspect first.

@ApochPiQ: i agree, there is no "perfect" solution. But AI can become close to perfect due to continuous learning which is not the case with current traffic coordination systems. They are not flexible and do not adapt to the needs of the people. Predicting traffic flow throughout the whole city would definitely be too complex, but for the majority of the distance people will probably stick to macromobility (i.e. traditional public transport, especially trains and subway). But as soon as people leave a train station, there ways split up and current public transport does not ensure enough flexibility. If public transport would take everybody to the desired destination, people would not use their cars anymore (this might solve the space problem in cities). The major question is: what sort of vehicle-sharing system which learns from human mobility behavior can ensure enough flexibility at an affordable price for governments? If such systems are first set up on the outskirts, they could be gradually expanded throughout the whole city.

Maybe I don't quite follow but are we talking about people while driving in their cars to actually talk to and or type data into a device which is collected and used to help determine better routes according to time of day, day of week and so forth? If this is the case then I think we are creating yet another reason to take bad drivers eyes and attention off the road. So simply track how a driver gets from point A to point B, monitoring this and finding out how they do it quickest at given times and days and what not might just be a start to a collection of quickest route points.

However to me an optimal progressive step would be to use sensors that the cities are currently using to time lights. Taking data of how many vehicles are passing through intersections, at what speeds and so forth can give you hints as to how fast you would move through that area on that path. Comparing this to the data of other nearby routes that lead to the same destination you can intelligently select a custom path that will lead through the least traffic. However I believe that this technology is already in the works, at least in my area there are already digital road signs that are doing this. When you are on the freeway every few miles there is a digital sign that tells you (I-480 to I-77 6 mins, I-90 to I-77 4 mins) This is just before the ramp that you can choose 480 or 90. However this isn't helping you if 90 puts you on 77 5 minutes farther away from where your getting off then 480 does...

Anyway, point is I think what these signs are starting to do is the breakthrough that will start helping ease traffic congestion. By integrating these flow monitoring systems into GPS devices we can recalculate to avoid the heavier traffic areas, maybe driving 2 extra miles but avoiding a 15 minute traffic jam will actually get you there quicker. This is what I would start looking into personally.