GP+Echo+Subsumption = Improved Problem Solving

Real-time, adaptive control is a difficult problem that can be addressed by EC architectures. We are interested in incorporating into an EC architecture some of the features that Holland’s Echo architecture presents. Echo has been used to model everything from cultures to financial markets. However, the typical application of Echo is a simulation to observe the dynamics of the modeled elements such as found in control problems. We show in this paper that some aspects of Echo can be incorporated into Genetic Programming to solve control problems. The paper discusses EAGP (Echo Augmented Genetic Programming), a modified GP architecture that uses aspects of Echo, and subsumption. We demonstrate the usefulness of EAGP on a robot navigation problem.