Selective entropy-fused proximal policy optimisation with federated reinforcement learning for intelligent multi-UAV trajectory and communication optimisation

Description

M Velmurugan and T Velmurugan proposed in 2026 a federated learning system for drone-based wireless networks in the 5G and emerging 6G era. Ground towers are costly and inflexible in remote places. Drones can move coverage where needed, if they can plan routes, assign users, save energy, and hand over connections without draining batteries or leaking data. Their design uses joint reinforcement learning, so drones learn together through a high-altitude platform without pooling user data.

The learning method, called selective entropy-fused proximal policy tuning, drops the replay buffers offline methods need, cutting memory and computing on board. Route planning, user pairing, energy-aware sharing, and handover management run inside one adaptive framework tolerating uneven user distributions. In experiments it improved throughput, fairness, latency, and energy efficiency over baselines. The authors present it as ready for deployment without simulation pretraining, though no headline numbers are given.

Catalog record

Source: https://europepmc.org/articles/PMC13473483