Hamilton–Jacobi–Bellman Enhanced Reinforcement Learning Control for a Linear System

Author(s):

Bradley T. Burchett

Abstract:

Deep reinforcement learning (RL) has recently come to the attention of many researchers in various disciplines. In this note, we explore the application of Deep RL with a supervised learning enhancement from optimal control theory enabled by a physics informed neural network to the control of a linear state space model. Performance is found to be worse than solutions from optimal control theory.