Coalition Game of Radar Network for Multitarget Tracking via Model-Based Multiagent Reinforcement Learning

Task assignment is crucial for multitarget tracking of the radar network and is mainly solved by centralized optimization methods, which results in the issues of robustness deficiency, high computational and communication costs, and inflexible adaptability in complex environments. To overcome these issues, task assignment of the radar network for multitarget tracking is formulated as a multiagent decision-making and learning problem where each radar node in the network acts as an intelligent agent that can make the tracking decision according to its task preference and interacts with other agents for the sake of the best network utility. To describe the decision preferences of radar nodes for various tasks, the criterion for the design of the utility function is presented, and a utility function under this criterion is devised based on the quality of service framework. Then, the coalition game with transferable utility is developed for task assignment where the utility of the coalition is completely transferred to all coalition members. The existence of the stable coalition partition of the developed game is analyzed theoretically, and the model-based multiagent random Fourier features reinforcement learning algorithm is proposed to solve the optimal solution to the game in the high-dimensional state space, which is proven to be converged at a Nash-stable coalition partition. Some numerical simulation results are provided to illustrate the effectiveness of the proposed algorithm in terms of tracking performance and resource conservation.

Automated summary

Task assignment for multitarget tracking in radar networks is formulated as a multiagent decision-making and learning problem, where each radar node acts as an intelligent agent that makes tracking decisions and interacts with other agents. A utility function is designed to describe the decision preferences of radar nodes, and the coalition game with transferable utility is developed for task assignment. The stable coalition partition of the game is analyzed theoretically, and a model-based multiagent reinforcement learning algorithm is proposed to solve the optimal solution. Numerical simulation results show the effectiveness of the algorithm in terms of tracking performance and resource conservation.