Research on Inverted Pendulum Control Strategy Based on Deep Reinforcement Learning

Aiming at the control planning of inverted pendulum task, the planning strategy based on Soft Actor-Critic (SAC) algorithm was studied. An agent based on Actor-Critic framework is designed, which takes inverted pendulum state as input and planned target position as output as controller. The agent has a total of 5 neural networks, namely, action network actor, 2 current evaluation networks and 2 target evaluation networks. Actor network outputs the next planned position according to the current inverted pendulum state, and the current Critic network and target Critic network respectively output the value evaluation of the current action and the next action according to the reward feedback from the environment. The PD control method is used to convert the planned target position of the mobile network output into the control voltage value, and actually control the inverted swing. Simulation results show that the proposed method is effective and efficient.