Sensor Fusion for Autonomous Driving

In a context where robotics and autonomous vehicles are becoming increasingly central in industry and research, this thesis presents the development of an autonomous driving system for Yahboom ROSMASTER X3 educational robot. The main objective is the implementation of a rover capable of autonomously following a line and recognising road signs by integrating sensor fusion techniques between LiDAR and a depth camera. This combination allows the robot to obtain a more accurate perception of its surroundings, enhancing navigation adaptability to complex scenarios. The system is based on ROS2 and exploits a hardware platform consisting of an NVIDIA Jetson Nano for image processing and autonomous navigation and Mecanum wheels that provide omnidirectional mobility, improving the robot manoeuvring capabilities. Computer vision plays a key role in traffic sign recognition, implemented via a MobileNetV2 SSD deep learning model, suitably trained on a customised dataset. The software pipeline includes the implementation of navigation algorithms, MQTT communication for remote management, and integration with ROS2 to ensure modularity and efficiency. After the assembly and configuration phase of the rover, the system was subjected to several tests to evaluate performance in real navigation scenarios for signal recognition, stability in line tracking, and responsiveness to environmental variations. The results obtained demonstrate that the integration of sensor fusion and machine learning within a ROS2 architecture allows a significant improvement in the robot autonomous capabilities. This work represents a contribution to research on robotic navigation systems, highlighting the potential of advanced perception technologies for real-world applications in dynamic environments.