Benchmark example for the Heuristically accelerated FRIQ-learning

Heuristically accelerated FRIQ-learning (HFRIQ-learning) is a Fuzzy Rule-Interpolation based Q-learning method which is able to embed external expert knowledge into the learning system. It usually speeds up the learning process, but in case if the expert knowledge describes an incorrect solution, it could also have a negative effect on the learning performance. In HFRIQ-learning, the Q-function knowledge representation is a state-action-value sparse (fuzzy rule interpolated) rule-base. The expert knowledge usually expressed by production (state-action) rules. This paper briefly introduces the way how the HFRIQ-learning optimises the Q-function rule-base and how it locates and corrects the incorrect expert rules. The paper also demonstrates the effect of the expert knowledge quality on the HFRIQ-learning performance on a popular reinforcement learning benchmark example.