This week has been focused on finding a good method to test my research question and get clear results. Just to recap, the current version of the question looks something like this:
"Can crowd simulation exhibit a more realistic behaviour by using academic AI principles while retaining a reasonable level of performance?"
I'm starting with this project with the assumption that I will not be getting better performance than the one you find in more traditional crowd simulation. This is a safe bet considering many of the implementations out there are already geared for efficiency whereas in that will not be this project's goal. For this projects success I will accept any results that are "playable". To specify, it need be something that would run on at least 30 frames per second with no significant spikes either way.
While I will be using standard performance values, taken from popular crowd simulations, my first step is to implement a prototype using a traditional approach that can be found in games (Boids) and use it not only as a basis of comparison but also as a basis for advanced AI prototype as well. For that to work smoothly I will ensure that it will be quite a modular project, being able to change the AI behind without needing to interfere with the rest of the project.
One of the most important steps of the project is doing the necessary research to narrow down which techniques would lend themselves to the project. The initial scope would be to implement all promising leads however due to the timescale given it is likely that will be reduced to only implementing the most suitable one.
After both are complete I will be logging performance times using in-built testing capabilities (either provided by the framework used or programmed by me) by running different scenarios meant to stress the simulation.
However, this alone would be insufficient for a proper comparison. In order to verify which of the versions will be superior I have created a user survey that will be given to members of the public after they have played through all the simulation versions. An image of the survey draft can be found here: https://drive.google.com/open?id=0B7xVRHFTM_KWS0dCSWExaVVnTlk