Meeting Abstract

P1.95  Thursday, Jan. 3  Exploring Morphospace through Computer Simulation: The Evolution of Navigation Behavior in Digital Fish CUMMINS, Megan A.*; LIEW, Chun W.; LONG JR., John H.; ROOT, Robert G.; Lafayette College liew@cs.lafayette.edu

We use artificial life technology and evolutionary computer techniques to investigate the hypothesis that navigational behavior can be improved by evolving mechanical traits that affect the tail�s ability to generate thrust. Specifically, we consider Young�s modulus and tail length, which, in combination, determine the spring stiffness of the tail. Increased tail stiffness is hypothesized to increase swimming control and to provide elastic energy storage. The evolution of vertebrae from notochords may have improved navigation in response to environmental sensory stimuli. To explore this, digital fish of variable tail stiffness compete and breed in an artificial selection environment using a genetic algorithm (GA). Up to ten parameters are variable. The digital fish are comprised of a disc-like body with a keel and a line segment tail hinged at the base. One off-center eyespot detects light, allowing the fish to swim up the light gradient on the surface of the water created by a fixed light placed above the water. Fitness is determined by a combination of swimming speed, ability to maintain direction and station keeping. The GA works to optimize swimming performance as the swimmers evolve and enables the efficient exploration of a large space of configurations and their effects on behavior. Unlike robotic swimmers, the digital swimmers allow many different parameter combinations to be studied free from the constraints of robotic construction. Fish behavior and results are compared to biomimetic robotic counterparts for physical validation. The results show that tail stiffness is related to navigational abilities and indicate that navigational abilities may have been an important feature of the selection environment of early vertebrates.