The analysis of deep reinforcement learning for dynamic graphical games under artificial intelligence

Area of Science:

• Artificial Intelligence
• Game Theory
• Control Systems

Background:

• Dynamic graphical games present complex challenges for autonomous decision-making.
• Traditional methods struggle with computational complexity and information exchange limitations.
• Optimizing strategies in these environments requires advanced learning techniques.

Purpose of the Study:

• To develop a deep reinforcement learning (DRL) framework for autonomous decision-making and strategy optimization in dynamic graphical games.
• To reduce computational complexity and information exchange through local performance metrics.
• To enable intelligent agents to make decisions using only local information.

Main Methods:

• An online iterative algorithm utilizing Deep Neural Networks (DNNs) and an Actor-Critic framework.
• Implementation of a distributed policy iteration mechanism for agent decision-making.
• Definition of local performance metrics to manage complexity.

Main Results:

• The DRL-based algorithm significantly improves decision accuracy and convergence speed.
• Demonstrated reduction in computational complexity and enhanced scalability.
• Effective performance in optimal control problems within dynamic graphical intelligent games.

Conclusions:

• The proposed DRL approach offers an effective solution for autonomous decision-making in complex graphical games.
• Local information-based distributed policy iteration enhances agent autonomy and efficiency.
• The method shows strong potential for real-world optimal control applications.