Swing-up and giant-swing control of Acrobot using reinforcement learning

Abstract

In this paper, we apply the Actor-Critic method, which is one of the machine learning methods, to swing-up problem of an Acrobot. First, we train an agent to reward the energy of the first link of the robot. At this time, we use the model that limits the amplitude of the second link. We confirm by simulations that the energy can be increased from a swing-up motion to a giant swing motion by the agents trained for the several amplitudes of the second link.