Genetically Optimized Neural Controller (GONC)

My undergraduate thesis project with Jesse Burstyn.

Stated goal: "To test the viability of using an artificial neural network as a controller for an agent in a three-dimensional simulated physics environment, specifically for the purpose of generating realistic humanoid animation."

The neural network outputs are used to control the agents 'muscles' and were trained using a genetic algorithm.

Coded in C++ and uses the Bullet Physics library for rigid body simulation. Genetic algorithm and neural network code all developped from scratch.

Googlecode wiki