Transfer Learning Method from Parameter Scene to Physical Scene Based on Self-Game Theory

Abstract

Due to the inherent challenges in meeting the comprehensive training requirements of combat scenarios within a digital environment, this paper proposes innovative solutions. Firstly, a parametric scene training method is introduced, aiming to enhance the adaptability of the training process. Secondly, the utilization of the digital environment as a test environment is suggested, which can significantly improve the efficiency of iterative virtual-to-real conversion. Additionally, employing digital simulation as a pre-test environment for the physical setting can effectively reduce deployment costs and enhance the safety of virtual reality migration experiments. The proposed approach involves constructing a training model based on the parameterized environment and implementing a feedback loop utilizing the digital environment agent model. This framework enables continuous verification of the environment’s parameters, ensuring the accuracy of decision-making strategies and evaluating the transferability of the decision-making model.