A New Paradigm for Control Theory: A Comparison of Novel and Status Quo Control Algorithms

Abstract

In this paper, we seek to develop the field of Online Nonstochastic Control as described by Hazan and Singh [12] by testing Agarwal, Hazan, Majumdar, and Singh’s Iterative Gradient Process Controller (IGPC) algorithm [2] against the current status quo Iterative Quadratic Linear Regulator (ILQR) [14]. These algorithms have applications not only for robotics, where they are commonly used, but also when working with all manner of nonlinear dynamical system. We conclude, through testing on the CartPole environment, that IGPC outperforms ILQR, and we consider other ways of testing the algorithms against each other. We then discuss the algorithm’s other potential applications [17].