Development of Advanced Control Strategy Based on Soft Actor-Critic Algorithm

The paper demonstrates the differences between two reinforcement learning methods – the Actor-Critic (AC) and the Soft Actor-Critic (SAC) applied for the benchmark pendulum and double pendulum control problems. The advantages and disadvantages of both methods as well as proposed modifications of the algorithms are discussed. The neural network has been developed using the Python and Tensorflow 2 libraries, and PyBullet Envs was used for environment simulations the case study.