A learning-based control system by knowledge acquisition within constrained environment

Knowledge acquisition is important in order to build a knowledge based system. One of the methods used in acquiring knowledge is reinforcement learning. Reinforcement learning is commonly defined as a try and error style learning that occurred in episodes. This is difficult to ensure a real control object safety condition since a control object is restricted to its environmental constraints. Therefore, a control system which can acquire knowledge within constrained environment, and use it to achieve a control objective in continuous time is proposed. This control system is applied on an inverted pendulum control system and its effectiveness is confirmed through a series of simulations.