Application of Reinforcement Learning in a Real Environment Using an RBF Network

Abstract

Abstract. The application of reinforcement learning algorithms in the context of robot behaviour learning is a poorly explored and very promising area of research. In the present work the learned strategy which resulted from simulation has been applied in a real world envi-ronment. To achieve good results in the real world it was necessary to build a simulation environment which mirrors the reality up to a prac-ticable degree. We use a radial basis function network to approximate the action-value function. To enforce a robot to learn a desired behav-ior a special online reward model has been developed. The approach reality-simulation-reality has been used to optimise the learning pro-cess in the simulation and apply the method in reality afterwards. Additionally the advantages and disadvantages of the application of the RBF-features over coarse coding with binary features have been examined. 1