CENTRALIZED REINFORCEMENT LEARNING FOR CONTROL AND TASK ALLOCATION FOR AN UAV TEAM

This thesis investigates multi-robot task allocation for UAV swarms, focusing on how multiple drones can cooperatively plan, navigate, and execute missions efficiently. It develops a layered system that combines low-level flight control, path planning, and high-level task scheduling within a unified framework. Reinforcement learning is used as a core methodology, and the selected software tools and simulation environments are chosen to support both accurate control learning and scalable mission-level decision making. To reduce computational cost, the work also introduces a surrogate environment that approximates travel times while preserving the main planning features of the full simulation. The proposed approach is extensively evaluated to assess the performance of each component individually and of the overall system in cooperative UAV mission scenarios.