This talk presents the ground (AGV – Autonomous Ground Vehicle) and aerial (UAV –Unmanned Aerial Vehicle) robot software stacks of an industrial automation project where a mobile manipulator robot (AGV) and a drone (UAV) operate together to support warehouse and palletizing workflows. The presentation focuses on how a practical robotic system can be built by combining robust open-source technologies, reducing development time while keeping the architecture modular, transparent, and adaptable to real-world requirements.
The presentation is split into two complementary parts:
The UAV stack highlights two key open-source repositories that played a central role in the robot architecture: PX4, used as the flight control framework, and NVIDIA Isaac ROS Visual SLAM, adopted to provide visual localization and mapping capabilities in GPS-denied or structured indoor environments. The presentation will show how these components were integrated into the drone stack, why they were selected, and which advantages they provided in terms of flexibility, maintainability, and performance.
The presentation also highlights the perception framework adopted across the robotic system, based on the open-source YOLOv11 network and the Ultralytics framework. This section focuses on how these technologies were integrated into both the AGV and UAV software stacks to provide a common perception layer supporting collaborative operations. The presentation will show how the camera mounted on the robot gripper is used for box detection and pose estimation during palletizing tasks, while the camera mounted on the drone provides information about box presence and approximate location on the target pallet, contributing to a coordinated aerial-ground robotic workflow.
