Behavoir Model Matrix LearningEach wolf contains a table formatted like the one on the left. All the active events (green) are summed up, and the one with the highest score becomes the wolves current state. Which each have behavior as well, (Stalk tries to stay out of site but follow the rabbit). Each state has requirements, for example, Stalk cannot occur if SeeRabbit isn't active.
Events - Reinforced LearningAn event occurs when a wolf sees or smells a rabbit. An event ends when the rabbit is caught, or lost sight of for X seconds. The weights involved in the event are ethier decremented or incremented depending on the results of the event
Genetic LearningThe two most successful wolves stay, and the least two are replaced by mixes of the first two wolves with some random mutations.
|