Multi-agent Distributed Cooperation Based on Incomplete Information Prediction

To solve the problem of multi-agent cooperative decision-making in a partially observable adversarial environment inspired by the learning and reasoning functions of the human cerebral cortex through memory, a new multi-agent distributed cooperative decision-making framework based on incomplete information prediction in a partially observable adversarial environment is proposed. The framework can use support vector regression and other prediction methods to predict invisible information in the environment through historical memory and current observed information and fuse the predicted information and the observed information as a basis of cooperative decision-making; Then, cooperative strategy learning is performed through distributed multi-agent reinforcement learning to obtain the decision-making model of each agent in the team. Multi-agent cooperative decision-making is verified in a typical partially observable adversarial environment using this framework and various prediction algorithms. The results show that the proposed framework is universal to various prediction algorithms and can improve the level of multi-agent cooperative decision-making by 23.4% while ensuring high prediction accuracy for invisible parts.