Applying Expert Heuristic as an a Priori Knowledge for FRIQ-Learning

Many Reinforcement Learning methods start the learning phase from an empty, or randomly filled knowledge-base. Having some a priori knowledge about the way as the studied system could be controlled, e.g. in the form of some state-action control rules, the convergence speed of the learning process can be significantly improved. In this case, the learning stage could start from a sketch, from a knowledge-base formed based upon the already existing knowledge. In this paper. the a priori (expert) knowledge is considered to be given in the form state-action fuzzy control rules of a Fuzzy Rule Interpolation (FRI) reasoning model and the studied reinforcement learning method is restricted to be a Fuzzy Rule Interpolation-based Q-Learning (FRIQ-Learning) method. The main goal of this paper is the introduction of a methodology, which is suitable for merging the a priori state-action fuzzy control rule-base to the initial state-action-value function (Q-function) representation. For demonstrating the benefits of the suggested methodology, the a priori knowledge-base accelerated FRIQ-Learning solution of the "mountain car" benchmark is also discussed briefly in the paper.