Integrating Real-Time Object Detection with LiDAR Data for Enhanced Robotic Autonomous Navigation

The adoption of autonomous mobile robots in complex environments, such as warehouses, factories, offices, airports, and metropolitan areas, has been steadily increasing in recent years due to technology advancements in fields such as artificial intelligence, edge low-power computing platforms, and sensor systems. The main focus of this thesis was the integration of YOLO (You Only Look Once), a neural network for object identification, within the navigation system already installed onboard a robotic platform to improve its detection capabilities and handling of dynamic obstacles on the road, such as cars and other vehicles. This thesis, developed within the LINKS Foundation in Turin, originated as an extension of an autonomous navigation project initially geared toward indoor mail delivery by TurtleBot robots. To address the challenges posed by outdoor navigation and in the context of last-mile deliveries, the project was subsequently oriented toward the use of Agilex's Scout 2.0, a mobile robot designed to operate in outdoor contexts. The work adopts ROS 2, the next-generation robotic framework, and Nav2, its navigation platform, which goal is to provide the tools for safe and adaptive navigation in dynamic outdoor environments. The available LiDAR data has been integrated with the results of the AI-based objects detection, obtained through YOLO and performed on the camera data, enhancing the Nav2 Stack's cost map for a more accurate and complete representation of the surrounding environment, which is critical for autonomous outdoor navigation. This data fusion overcame some of the limitations of LiDAR in detecting dynamic obstacles at varying distances, improving the system's ability to make quick and safe decisions when navigating complex urban delivery scenarios.