Robust Optimal Control of Continuous Time Linear System using Reinforcement Learning

The paper explores the area of Reinforcement Learning (RL) that is emerging as an elegant tool in solving the control problems that require optimality and robustness simultaneously. The task of stabilization of an inverted pendulum system with known/unknown internal dynamics is discussed to reveal the advantages of RL approach over conventional approach and is demonstrated using simulations. Also discussed are the algorithmic challenges faced by the designer in using the RL approach for online robust optimal control of unknown systems.