Decentralized coordination of intelligent system of systems under partial observability

Limited by the physical constraints of the weapon platform equipment, such as cameras and sensors, it is only capable of observing local information in its immediate vicinity, particularly within high-confrontation and high-interference battlefield environments. Consequently, this hinders the effective realization of decentralized coordination between platforms within the combat system of systems (SoS), thereby impeding efficient execution of combat tasks. To enhance the efficient utilization of combat resources for the construction of task communities, enabling platforms to decentralized coordination in executing combat tasks based solely on local information, this study proposes an approach utilizing the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm that leverages partial information for the construction of task communities. By engaging in continuous interaction with the environment, the platform can enhance its decision-making capabilities and independently generate optimal solutions based on local information. Furthermore, we propose an information sharing mechanism to enable the platform to obtain a wider observation area, thereby enhancing the accuracy of its task resource allocation. The evaluation results demonstrate that the proposed method significantly enhances platform coordination efficiency and resource utilization, even when operating with limited information. In comparison to other baseline methods, the task satisfaction degree can be increased by approximately $15\text{\%}\sim 20\text{\%}$ with only partial information.