Modelling and Simulation of Autonomous Decision - making Process Based on Proximal Policy Optimization
Chen Zhou, Baoran An, Weijian Huang, Bing Yu · 2023
The autonomous decision-making problem in the mixed process of medium-range and close-range confrontation stages of agents is studied in this paper, utilizing reinforcement learning theory. The complexity and intensity of the environment resulting from the integration of information technology and system-of-systems (SoS) support in the agent confrontation scenario, as well as the oversimplification of scenarios in traditional research on autonomous decision-making of agents, are taken into account. Firstly, a simulation platform is constructed based on the Agent-Based Modeling and Simulation (ABMS). Then, the state space, action space, and reward function are designed to adapt to the multi-stage confrontation scenario for agents. Finally, an autonomous decision-making method based on Proximal Policy Optimization (PPO) algorithm for agent confrontation is proposed. The experimental results demonstrate that the proposed decision-making algorithm can learn reasonable confrontation tactics in multi-stage confrontation scenarios, verifying the feasibility of studying the problem based on the above modeling and simulation framework.