A learning-based non-linear control system with constraint consideration

A learning based control system which can consider control-interfering-constraints is proposed. Reinforcement learning is applied into the system to make it possible to learn to control upon achieving its control objective. Constraints could exist whether in the system manipulate-able parameter or even in non-manipulate-able parameters as in a non-linear control system, thus often causes learning failure when these constraints is not considered. Therefore, reinforcement learning is used to arranged substitute targets that can consider any interfering constraints during the learning process. This control system is applied on an inverted pendulum system and its effectiveness is later confirmed through simulations experiments.