With support from CUAV Tech, this project received two CUAV NANO-7 flight controllers and two NEO 4 SE GNSS modules for development and testing of an autonomous UAV exploration system.
The project focuses on developing a complete autonomous exploration and navigation pipeline that integrates perception, mapping, planning, local control, and safety filtering with the PX4 flight stack for gps denied environments. The GNSS module were used during outdoor testing and calibration.
The overall architecture is:
RGB-D + IMU → SLAM → Occupancy Grid → Frontier Detection → RRT* → APF → CBF Safety Filter → PX4
The system was developed with both PX4 SITL/Gazebo and physical UAV deployment in mind, maintaining the same general architecture between simulation and hardware.
The visual-inertial odometry from RTAB-Map is fed to the PX4 estimator, enabling vision-based state estimation and GPS-denied navigation.The planner identifies frontiers between explored and unexplored regions and selects exploration targets based on information gain, distance, and approach feasibility. RRT* then generates a collision-aware path toward the selected frontier, while an Artificial Potential Field (APF) controller generates smooth local velocity commands while responding to nearby obstacles.
A Control Barrier Function (CBF) safety filter is placed between the local controller and PX4 to provide an additional layer of collision avoidance. The filter enforces obstacle and frontier safety constraints and uses a closed-form KKT-based QP solution to modify the commanded velocity when necessary.
For simulation, the system uses PX4 SITL, Gazebo, ROS 2, and Micro-XRCE-DDS. For hardware deployment, the same overall architecture was integrated with CUAV NANO-7 flight controllers, with companion computers running the higher-level SLAM, exploration, planning, and safety components. The hardware setup also includes Intel RealSense D435i, and Luxonis OAK-D Pro W cameras.
This simulation-to-hardware approach allowed the exploration planner to be evaluated not only in Gazebo but also under real-world constraints such as sensor synchronization, onboard computation, PX4 estimator configuration, and real-time ROS 2–PX4 communication.
Videos:
Project Repository
The complete project, including the ROS 2 packages, PX4/Gazebo simulation setup, launch files, exploration planner, and hardware configuration, is available on GitHub:
The project demonstrates an end-to-end approach to autonomous UAV exploration, combining SLAM, frontier-based exploration, sampling-based planning, local control, and safety-critical control with the PX4 flight stack.