Swing Up and Balance of an Inverted Pendulum Using Reinforced Learning Approach Coupled With a Proportional-Integral-Derivative Controller
Amit Pal, Tamara N. Nestorovic · 2022 International Conference on Electrical, Computer, Communications and Mechatronics Engineering (ICECCME) · 2022
Classical approaches in control system theory have been developed for decades, yet further improvement and development of advanced control strategies still represents a challenge, depending on different implementation requirements. An inverted pendulum has been established as a benchmark and in this work it is used for testing new control strategies. Due to improvements of hardware, particularly with respect to improvement of the computational time, as well as due to an increase of data available through acquisition, machine learning and artificial intelligence record success in applications in the field of control. In this work, a control strategy which relies on the techniques from Reinforcement Learning (RL) is used to swing up and balance an inverted pendulum. The technique is implemented in combination with a ProportionalIntegralDerivative (PID) controller used subsequently to keep the pendulum balanced and make the system more robust to disturbances and noise. Laboratory experiment has been carried out to verify the proposed control strategy and it was shown that it is successful in swinging up and balancing the inverted pendulum in real-time even in the presence of disturbances. The implementation shows the effectiveness of combining an RL algorithm with a PID controller in real world applications.