It might be useful for implementing the kind of decisions that are required in spring AI (for example: assigning construction units to build tasks).
It's basically extended matrix algebra: There you usually are solving a linear system in the form of Ax = b, but with MLCPs you can have additional constraints on the system such as inequalities, or quadratic terms (known as quadratic programming)
I remember from JCAI that I tried to solve such problems with the usual methods (IE: just start coding and try to take all the factors into account), but I usually failed at that
Once you have a good structure set up the equations for the solver, you can basically use the solver everywhere and focus on building a good AI decision model and collecting information for your model.
Another advantage of using a very generic solver method is that you can make your entire AI multithreaded by making your solver multithreaded (Although right now I have no idea if thats actually feasable)
Chris Hecker has done a presentation about it which is downloadable: http://chrishecker.com/The_Mixed_Linear ... ty_Problem
The presentation is pretty useful to see the advantages of LCPs, but after that comes a shitload of math that needs be learned (Which I'm trying to do now).
Online books:
http://www.princeton.edu/~rvdb/LPbook/onlinebook.pdf
http://ioe.engin.umich.edu/people/fac/b ... y_webbook/
EDIT: Another idea, this lends itself well for having one fast C++ solver and a flexible lua based AI that just sets up the equations.
