A*-based Collision Avoidance for UAVs with ROS 2 and PX4 Integration

This thesis, developed in collaboration with MAVTech s.r.l., presents a modular architecture for collision avoidance in UAVs designed for outdoor applications. The system is implemented in ROS2 and integrated with the PX4 Autopilot for flight operations. The primary objective is to enhance safety while maintaining operational continuity: avoidance maneuvers are triggered only when necessary, ensuring no alteration of system nominal conduct. The approach combines an A* local planner operating on a dynamic costmap—fed by on-board sensors—with a Bug-like reactive behavior as a fallback strategy when local scheduling does not locate safe steps. A context-aware mode management system determines which algorithm to activate and when to apply its output to vehicle control. The architecture emphasizes a clear separation between perception, planning, and control, promoting portability, reusability, and future scalability. Development follows a two-phase strategy. In a first phase, the solution is implemented on a terrestrial mobile robot to consolidate the avoidance logic in a platform-agnostic context, without autopilot dependencies. In a second phase, the approach is transferred to the drone, favoring the optimization of A* for flight, while the integration of Bug behavior is postponed to future work. In this scenario, a state system is introduced to manage mission flow: it allows PX4 mission commands to proceed uninterrupted unless obstacles are detected, in which case the A* planner intervenes to ensure safe navigation. Integration with PX4 is achieved in Offboard mode, with careful attention to reference frame consistency and command interface alignment. The evaluation covers the entire spectrum of scenarios in simulation, including challenging ones; tests on a real platform take place in simplified and controlled environment, in compliance with operational and safety constraints. Overall, the work proposes a flexible and reusable solution for integrating collision avoidance into UAV platforms—compatible with real-world missions—while maintaining seamless integration within the ROS 2 and PX4 ecosystems.