DEEP REINFORCEMENT LEARNING IN CONTROL

Abstract

The objective of this research is to understand and implement Deep Reinforcement Learning in Control domain. Deep reinforcement learning has become a topic of great interest in recent years attracting significant attention from both all type of industries starting from automation to healthcare. In this research, various control tasks which include both discrete and continuous action spaces have been solved using various algorithms like Deep Q Network (DQN) and Twin Delayed Deep Deterministic Policy Gradient (TD3) to perform tasks like inverted pendulum control. A novel reward function is proposed to solve the task of balancing the pendulum in inverted position over a cart. This reward function enhances performance of TD3 over the baseline methods in terms of output control. Exploration enhancement evolution strategy like entropy regularization is hybridized with TD3 and the resulting algorithm shows a positive response by improving the reward curves.