Evolving neurodynamic controllers for autonomous robots

The creation of architectures for controlling the behavior of autonomous systems is a difficult challenge. Evolutionary robotics uses neurally inspired models, rather than explicit symbolic systems, to evolve controllers for robots. Most approaches in evolutionary robotics have used abstract ANN or spiking single neuron models to evolve control architectures. In this paper we apply the evolutionary approach to creating a controller for an autonomous robot based on the aperiodic K-set neural population model. We introduce a discretization of the basic K-set units. We then demonstrate that the evolutionary approach evolves effective controllers for navigation tasks using the basic discrete units.