My proposal is now complete and submitted!
It is quite different that the document I had last week but only for the better. I received feedback from my peers but especially my project supervisor which made me reconsider my approach and make some drastic changes.
The most significant one is limiting my scope, to what was before several poorly implemented prototypes of different AI techniques to one single, hopefully better implemented one. So ditching the previous approach I have redefined it to look something along these lines:
That has resulted in a part of the proposal that I would consider it's worse part, that is no definite direction in to which technique will be used. Instead I have listed several popular ones and a choice between them will be made after some trail &error and a literature review to try and determine which of the techniques would suit best and give promising results.
Some of the possible choices would be neural networks, a type of Learning AI that is designed to mimic the way a brain uses information and learns from success and failure mainly used when a problem has known inputs and outputs or genetic algorithms, a type of heuristic algorithms inspired by genetics and natural selection. They are typically used for optimizing functions by using the “survival of the fittest” approach where the most optimal solutions produce even more optimal solutions. However this is not a definitive list as this will be expanded during the review.
What comes next is the feasibility demo, which will probably manifest as two similar versions of crowd simulations compete with performance testing and a small survey. This is intended to showcase how the project will go ahead rather than have any definitive results. It will most likely be done in Unity.