RISK-SENSITIVE LQR PROBLEMS WITH EXPONENTIAL NOISE

No Thumbnail Available

Date

2024-04-26

Authors

Shortanbaiuly, Olzhas

Journal Title

Journal ISSN

Volume Title

Publisher

Nazarbayev University School of Sciences and Humanities

Abstract

This thesis is about optimal control of Markov Decision Processes and solving risk-sensitive cost minimization and reward maximization problems, specifically, the Linear Quadratic Regulator (LQR) problem with Average-Value-at-Risk criteria. The problem is solved for different risk levels, different random noises (theoretical and sampled), and using different methods: analytical and approximate dynamic programming. The obtained results were analyzed and discussed for the presence of certain patterns and trends. The results show that approximate dynamic programming is a very accurate method for solving risk-sensitive LQR problems with exponential noise.

Description

Keywords

LQR problem, Markov Decision Process, Average-Value-at-Risk, Approximate Dynamic Programming, Exponential Distribution, Type of access: Open Access

Citation

Shortanbaiuly, O. (2024). Risk-sensitive LQR problems with exponential noise. Nazarbayev University School of Sciences and Humanities