SCHUMACHER, J.W.; LAMMERT, A.C.; LONG, JR., J.H.; Vassar College; Vassar College; Vassar College: Evolutionary Robotics: Exploring the Origins of Early Vertebrates using Biomimetic Swimmers

Evolutionary robotics (ER) is an established technique used to automatically optimize autonomous robots. Agents are selected based on efficacy of task performance and offspring are created using functions analogous to mutation and random mating. Since we are interested in the evolutionary origins of early vertebrates, we use ER to test the hypothesis that vertebrae evolved in response to selection for improved swimming performance. Our �early vertebrate� is a surface-swimming robot whose one-eyed phototactic behavior is based on that of ascidian tadpole larvae. We built three identical tadpole robots, each with a digital microcontroller that computes a lateral tail offset in response to a given light intensity transduced at the single light sensor. These robots detect and seek light sources and then orbit around the source. To the robot�s flapping, propulsive tail we added a vertebral column whose structure we can vary and hence select and evolve. With a compound fitness function, we select for increased swimming speed and maneuverability and decreased path length and time taken to the light source. We predict that selection for enhanced phototactic navigation will create vertebral columns with decreased length and increased number of vertebrae, a combination that increases the tail�s flexural stiffness. Preliminary results are intriguing, with the ancestral notochord, lacking vertebrae, scoring high relative fitness. Supported by the National Science Foundation (BCS-0320764).