RoboBirds

This thesis presents a model for the simulation of sedentary and migratory birds using parameters and values from biological findings. Furthermore, we evolve artificial neural networks (controllers) for these birds that mimic sedentary and migratory behavior. The emphasis is on the comparison of sedentary and migratory neural networks. The birds are simulated using a robot simulator and the resulting artificial neural networks are analyzed for structure and function. Real birds' cognitive abilities like sight and orientation are modeled as sensors for the simulated birds. The results show that i) not all available sensor input is utilized as expected, rather the birds find alternatives to percept the environment ii) the evolved migration behaviors are based on simple net structures iii) the larger the energy capacity of a bird, the simpler its net structure tends to be.