Modeling the Learning Ability of Fish in an Aritificial Fish Simulation

Abstract

Simulating the collective behavior of animals in a virtual environment is an intriguing and complex open research area. Flock of birds, schools of sh, a group of working termites, and group of foraging ants are some natural behaviors that based to conduct many researches in Arti cial Life eld. Agent based animal models that exhibit collective intelligent behavior are used for various purposes as same as virtual reality applications like lms, games, medical applications, military applications, etc. There are a number of models in the eld of arti cial life that simulate the collective behavior of sh. But the common limitation of these models is that they do not consider the learning ability of the sh. Fish shows a considerable amount of learning in tasks like foraging and defense. So in order to create more accurate models, these learning abilities should be considered. The purpose of this particular research is to nd out the ways of designing a sh behavior model that contains these learning abilities. The design of this model consists of two main areas, the agents and the environment. The agents are responsible of simulating the behaviors of sh, including the learning abilities. The environment represents the natural environment which the actual sh are behaving in and is responsible in aiding the agents to behave and learn. The simulating the learning abilities of the sh, which is the main consideration of this research is carried out using the temporal di erence learning which a reinforcement learning algorithm. It is also an algorithm in uenced by theories of animal learning. Implementation of the research was carried out using OpenSteer C++ open source library component. The prototype was built on a machine with