Online reinforcement learning multiplayer non-zero sum games of continuous-time Markov jump linear systems

• Published in: Applied mathematics and computation 2022-01, Vol.412, p.126537, Article 126537
• Main Authors: Xin, Xilin, Tu, Yidong, Stojanovic, Vladimir, Wang, Hai, Shi, Kaibo, He, Shuping, Pan, Tianhong
• Format: Article
• Language: English
• Subjects: Coupled algebraic Riccati equations; Markov jump linear systems; Multiplayer non-zero sum games; Reinforcement learning
• ISSN: 0096-3003; 1873-5649
• DOI: 10.1016/j.amc.2021.126537

•A novel online mode-free integral reinforcement learning algorithm is proposed to solve the mutiplayer non-zero sum games.•The online learning is used to compute the corresponding N coupled algebraic Riccati equations.•The policy iterative algorithm is applied to solve the coupled algebraic Riccati equations corresponding to the multiplayer nonzero sum games. In this paper, a novel online mode-free integral reinforcement learning algorithm is proposed to solve the multiplayer non-zero sum games. We first collect and learn the subsystems information of states and inputs; then we use the online learning to compute the corresponding N coupled algebraic Riccati equations. The policy iterative algorithm proposed in this paper can solve the coupled algebraic Riccati equations corresponding to the multiplayer non-zero sum games. Finally, the effectiveness and feasibility of the design method of this paper is proved by simulation example with three players.