Combining the principles of fuzzy logic and reinforcement learning for control of dynamic systems

Many complicated dynamic systems use fuzzy inference systems due to its robustness and smooth response. However, they do require the existence of an expert to consider a suitable rule-set. The main challenge is, therefore, to be able to define a suitable rule-set needless of an expert. Studies have been carried out to design fuzzy logic based controllers without the need of an expert’s experience and knowledge. In order to solve this issue reinforcement learning (RL) algorithm is recommended to define fuzzy rules for a fuzzy control system. Reinforcement learning can also be practical in driving the generation of the suitable rule-set based on the interactions with the environment. It can be simply combined with fuzzy logic and provide the relationship between the states and the admissible action, which is the same as creating the uzzy logic if…then engine. In this paper reinforcement learning approach is used to design a real controller for controlling an inverted pendulum modeled in SIMMECHANICS environment while a force is exerted to its cart.