Multiagent $Q$-Learning With Particle Filtering for UAV Tracking in Open-RAN Environment

This article introduces a method for target tracking that leverages mobile sensor nodes and uncrewed aerial vehicles (UAVs) within an open radio access network (RAN) framework. The Open RAN is a flexible and standardized architecture that allows open and interoperable components in RANs, promoting efficiency and adaptability. The core methodology involves improving the accuracy and energy consumption tracking in urban areas filled with obstacles and dynamic conditions. Mobile sensor nodes use a particle filtering algorithm to detect and estimate target positions, and this information is relayed to nearby evolved/next-generation nodeBs (e/gNBs), which function as the RAN infrastructure. The e/gNBs manage clusters of UAVs using a specialized xApp integrated into the near-real-time RAN intelligent controller. The UAVs utilize a comprehensive tracking strategy based on received signal strength, a trilateration algorithm, and an enhanced multiagent $Q$-learning (eMAQL) algorithm. This approach enables UAVs to optimize their flight paths while balancing accuracy, power usage, and communication delays. The experimental results show that the system achieves optimal performance with eight discrete actions for eMAQL, with UAVs consuming an average of 90 W of power and maintaining a root-mean-square error of less than 0.5 m for target position estimation. These results highlight the system's effectiveness in providing precise and energy-efficient tracking in complex urban environments.