Policy Gradient Convergence in Potential Linear-Quadratic Games with Decoupled Dynamics
From MaRDI portal
Publication:6437568
arXiv2305.13476MaRDI QIDQ6437568FDOQ6437568
Authors: Sara Hosseinirad, Giulio Salizzoni, Alireza Alian Porzani, Maryam Kamgarpour 
Publication date: 22 May 2023
Abstract: This paper explores the potential game properties of the non-cooperative finite-horizon general-sum Linear Quadratic (LQ) game. We show that a general LQ game is potential if and only if it is an identical interest game. Based on this result, we restrict the class of LQ games to those with decoupled agent dynamics and information structure. For this restricted subset, we derive conditions under which the game is potential. Furthermore, for the identified subset, we prove the convergence of policy gradient to a Nash equilibrium.
This page was built for publication: Policy Gradient Convergence in Potential Linear-Quadratic Games with Decoupled Dynamics
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6437568)
