Research on the design of intelligent control strategy for linear primary inverted pendulum based on deep reinforcement learning algorithm with SolidLab’s microcontroller programming

Deep reinforcement learning, as an advanced machine learning method, is capable of automatically learning optimal decision-making strategies in complex environments. The core objective of this paper is to apply deep reinforcement learning algorithms to SolidLab’s microcontroller programming in order to realize the intelligent control of the linear one-stage inverted pendulum system. The study takes the linear one-stage inverted pendulum produced by A Technology Company as the control object, and adopts the model-free control structure of the deep reinforcement learning algorithm to build the controller and conduct virtual simulation experiments. Comparing the experimental effects of LQR and DQN algorithms, the LQR algorithm is better than the DQN algorithm in stabilizing pendulum control of inverted pendulum. Accordingly, a balance controller based on the offline Q learning algorithm is further designed to realize the inverted pendulum stabilization in kind. After optimizing the design strategy, the inverted pendulum system can be rapidly stabilized within 0.9s when it is perturbed by a small angle of about 12°. It shows that the method in this paper can realize the intelligent control of the inverted pendulum system at the linear level.