Evolution of Efficient Swimming Controllers for a Simulated Lamprey

Abstract

This paper investigates the evolutionary design of efficient connectionist swimming controllers for a simulated lamprey. Efficiency is defined as the ratio of forward swimming speed to backward mechanical wave speed.Using the lamprey model proposed by Ekeberg (1993) and extending the work of Ijspeert et al. (1999) on evolving lamprey swimming central pattern generators (CPGs) through genetic algorithms (GAs), we investigate the space of possible neural configurations which satisfies the property of high swimming efficiency. Techniques are devised to measure efficiency at various swimming speeds. The measurements are incorporated into the fitness function of Ijspeert's original GA and efficient controllers are evolved. Interestingly, the best evolved controller not only is capable of swimming in a similar manner to the real lamprey, but also with the same efficiency (about 0.8). Moreover, it can exhibit a wide range of controllable speeds and efficiencies.