New AI framework helps drones adapt faster to emergency situations by preventing network connectivity issues during disasters

PE-MAMoE (Plasticity-Enhanced Multi-Agent Mixture of Experts) enables UAVs serving as aerial base stations to quickly restore connectivity after disasters

The system addresses plasticity loss in deep reinforcement learning caused by sudden changes in user mobility and traffic demands

Uses a mixture of experts approach where each drone selects specialized experts per decision step, with a Phase Controller that reinvigorates the learning process after major shifts

Built on multi-agent proximal policy optimization with centralized training and decentralized execution for efficient coordination