Machine-Learned Crowds
During the second term at BU, we had a personal inquiry unit which gave us freedom to learn or research anything we wanted. I chose to look into behavioural animation. I investigated some behavioural models proposed in research, played around with Golaem and stuff like that. After a while, my eye fell on some research papers that described machine learning techniques that came up with behaviours for crowds automatically. I decided to implement something similar to what Craig Reynolds did with the game of tag.
The above video was part of my exhibition stand during the Personal Inquiry exhibition. As you might be able to tell, I implemented a system to automatically come up with crowd behaviours using genetic programming. Everything was made using python, and as an exercise/learning experience, I made sure I was able to use the generated behaviours in Golaem (a crowd plug-in for Maya), using the Golaem python api. The behaviours were then combined with other Golaem behaviours to have some animation when character were moving. Obviously, this project was more about the technical and programming aspects than the actual results!
