Real-Time Collision Avoidance in Autonomous Drone Fleet using Decentralized Priority-Based Model Predictive Control

This thesis presents the design and implementation of a decentralized collision avoidance system for Unmanned Aerial Vehicles (UAVs). The approach relies solely on local information and is optimized for real-time execution. A priority-based Model Predictive Control (MPC) strategy is adopted, where UAVs are assigned priorities: lower-priority drones predict the trajectory of higher-priority ones and adapt accordingly, while higher-priority drones ignore others. This reduces communication overhead and computational cost. Drones navigate toward a target through velocity commands computed by an optimizer, which selects the best intermediate goal and velocity scaling factor by minimizing a cost function. The cost accounts for proximity to obstacles, distance to the final target, and deviation from nominal speed. Trajectories are simulated within a fixed time horizon, both for the drone itself and surrounding obstacles. The system is validated in a simulation framework based on ROS2, Gazebo, and PX4. The results of the tests demonstrate effective obstacle avoidance while preserving smooth motion and safety margins. The proposed method proves robust and scalable, suitable for fleet applications with further extensions to more complex scenarios.