Robotic Animal Herding

An investigation into the use of neural networks and genetic algorithms for herding animals. Parts of a previous Robot Sheepdog Project are reimplemented in a modular, object-oriented style in Java, and new algorithms are considered for controlling the Herder entity in the simulation. A simple recurrent neural network architecture is implemented, and trained using backpropagation and genetic algorithms. Backpropagation is found to be difficult to apply effectively to simple recurrent networks, but some success is achieved with genetic training.