Accelerating Fuzzy Actor–Critic Learning via Suboptimal Knowledge for a Multi-Agent Tracking Problem

Multi-agent differential games usually include tracking policies and escaping policies. To obtain the proper in unknown environments, agents can learn through reinforcement learning. This typically requires a large amount of interaction with environment, which is time-consuming inefficient. However, if one an estimated model based on some prior knowledge, control policy be obtained suboptimal knowledge. Although there exists error between guided will avoid unnecessary exploration; thus, learning process significantly accelerated. Facing problem optimization for multiple pursuers, this study proposed new form fuzzy actor–critic algorithm knowledge (SK-FACL). In SK-FACL, information about environment that abstracted as model, calculated Apollonius circle. The combined algorithm, improving efficiency. Considering ground game two pursuers evader, experimental results verified advantages SK-FACL reducing error, adapting to sudden changes made by evader compared pure algorithm.