Defence Emin Martirosyan: "Optimal Control and Reinforcement Learning Algorithms for Inverse Dynamic Games"

Promotors: 1^st promotor: Prof Ming Cao, 2^nd promotor: Prof Jacquelien M. A. Scherpen

Abstract: In this study, we address the inverse problem in linear-quadratic dynamic games, aiming to learn the cost function such that the given or observed tuple of feedback laws constitutes a Nash equilibrium. We examine several cases of linear-quadratic dynamic games. For continuous-time games, we develop gradient-based algorithms, while for discrete-time games, we propose an iterative algorithm to solve the inverse problem using modified algebraic Riccati equations. The study considers various payoff and information structures. The developed algorithms are implemented in both model-based and model-free approaches. To further illustrate potential solution characterizations for the inverse problems, we demonstrate a method to generate an infinite number of equivalent games without the need to rerun the entire algorithms repeatedly.

Dissertation