Post by Oriol Arroyo i Font

Final Projects from Advanced Robotics at Escola Universitària Salesiana de Sarrià - EUSS, School of Engineering In the Advanced Robotics subject, students work through the full stack of modern robotics: ROS 2, computer vision, artificial intelligence, database management, automation pipelines, and 3D reconstruction. The course is structured around building real systems, not just theory. Each block is introduced with hands-on exercises that progressively build up the skills needed for the final project. At the end of the semester, students design and implement their own autonomous perceiving system from scratch, combining as many of the course blocks as they can into a single working solution. Here are 3 of the projects built this semester: 1st. Real-Time PPE Compliance Monitor with ROS 2 & AI Alerting by Niels Breunig & Diego Casella A five-node ROS 2 (Jazzy) pipeline that captures a live camera feed, detects people with YOLOv8n, and checks head protection with a custom YOLO model labelled on Roboflow and trained via Google Colab. Each person gets a persistent track ID, demographics are estimated with DeepFace, and every compliance event is logged to a SQLite database. Non-compliant detections trigger a Telegram alert through an n8n automation workflow. Runs entirely on a single laptop under Ubuntu 24.04. 2nd. Vision-Guided Robot Arm with Automatic Movement - ABB IRB120 by Darius Cristian Vasilache & "Amalia Fara - https://lnkd.in/emQzXvrm" A camera sensor integrated into a Gazebo simulation of the ABB IRB120 - the same robot model we have in the lab. Using ROS 2 Humble and MoveIt 2, the system detects colored objects on a table, converts pixel coordinates into 3D world coordinates, solves inverse kinematics to compute joint angles, and commands the robot to move autonomously above the detected target. 3rd. Drone 3D Reconstruction with AI-Powered Tennis Ball Highlighting by Pau Marset & Marc Muñoz Gallego Drone footage of an outdoor scene with tennis balls, processed into a 3D model using Agisoft Metashape. Frames are split into two parallel branches: grayscale conversion for the final texture appearance, and HSV color segmentation to isolate the balls. Both are fused so only the balls keep their original color. Metashape aligns on the original color frames for accuracy, then the fused images are swapped in at texture generation, delivering a full 3D reconstruction in grayscale with the tennis balls rendered in color. #Robotics #ROS2 #ComputerVision #ArtificialIntelligence #3DReconstruction #YOLO #MoveIt #Gazebo #EUSS #Barcelona #Engineering