Robots become constantly more important for industry and the life of individuals. Fitting their gait to miscellaneous terrains and changing tasks is crucial for meeting given requirements. This calls for machine learning techniques. An approach to automatic gait optimization is studied within this work. Based on Bayesian optimization using Gaussian processes as a prior probability distribution over functions, the gait of a robot prototype is optimized. Experiments are done in simulation, as well as on the actual robot. The robot model is a six-legged machine, a so called hexapod, derived from an insect’s morphology. The underlying controller for basic locomotion control, is a parametrized central pattern generator. The approach is evaluated through experiments in simulation, using MuJoCo as physics engine, and experiments on an actual prototype, using Augmented Reality tracking.