An evolvable virtual ecosystem: Applying genetic algorithms, artificial neural networks, and fuzzy systems to a virtual environment
Date
1998
Authors
Eccles, Jeremy Samuel
Major Professor
Advisor
Dickerson, Julie A.
Committee Member
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Virtual agents can add realism to a virtual environment. This thesis describes a virtual ecosystem that combines elements of genetic algorithms, artificial neural networks and fuzzy systems to create and control the movement and interaction of autonomous virtual agents within a virtual environment. Within the ecosystem, each creature has its own simple neural network that controls its movement based on inputs to its sensors. The neural networks learn to perform food-finding and predator-avoiding tasks through evolution of their connection weights via genetic algorithms. Augmented fuzzy cognitive maps govern the overall functioning of the environment and allow each creature to switch between different neural networks. Real-valued edge connections were determined to produce transitions between states that more natural than those produced by a simple augmented FCM.
Motion loops were designed to move the creatures based on the responses of their neural control structures, and experiments were conducted to determine the appropriate parameter settings for them. Physics models of motion implemented in the motion loops were shown to produce more natural movement of the autonomous agents and often aided in their evolution. Collision detection, competition, and coevolution were also experimentally shown to be helpful in evolving the autonomous agents. The research and experimentation resulted in the development of an evolvable virtual ecosystem that functions in real-time.
Series Number
Journal Issue
Is Version Of
Versions
Series
Academic or Administrative Unit
Type
thesis