Continuous control using Deep Q Learning with Oscillatory Activation functions

Abstract

The capacity of reinforcement learning (RL) to learn from the interaction between the environment and agent provides an optimal control strategy. DQN is a deep neural network structure used for estimation of Q-value of the Q-learning method. The CartPole game is essentially a game in which a stick is attached to a cart and the cart moves along a friction-less track. The goal here is to make the cart move left or right to keep the pole from falling. The system is controlled by applying a force of +1 or -1 to the cart. The pendulum starts upright, and the goal is to prevent it from falling over. A reward of +1 is provided for every timestep that the pole remains upright. The episode ends when the pole is more than 15 degrees from vertical, or the cart moves more than 2.4 units from the center. In this paper, we explore the CartPole game and the effect oscillatory activation functions have on this.