A ROS 2 autonomous robotic system for hospital navigation — localizing with AMCL, reading AprilTags to find the target room, opening doors, and classifying patient state with a custom YOLO model under a central Mission Manager.
This project is a ROS 2 based autonomous robotic system designed to navigate through a hospital environment, identify a target room using AprilTags, enter the room, navigate to a patient, and determine the patient's condition using a custom YOLO object detection model.
Goals included combining autonomous navigation, localization, AprilTag detection, door control, and AI-based patient-state detection under a central Mission Manager; running a full mission from localization to patient classification; and detecting three patient states — Alive (standing), Alive (in bed), and Collapsed (on floor).
Built a Mission Manager ROS 2 node that sequences the full hospital challenge: localize with AMCL, navigate to the pinboard, read the AprilTag for the target room, verify the room tag, open the door via service call, then move to the patient observation pose.
Integrated Nav2 for path planning in a mapped hospital world in Webots on TurtleBot3. Used AMCL for localization against a prebuilt hospital map, and AprilTag detection to resolve room identity from the pinboard and doorway tags.
Trained and deployed a custom YOLOv8 model (Ultralytics) to classify patient state in a single detection pass at the observation pose. The model reports three classes — standing, in bed, or collapsed on the floor — and feeds the result back into the mission sequence.
Coordinating localization, navigation, AprilTag perception, door control, and vision in one reliable mission state machine.
Getting AMCL and Nav2 stable enough in the Webots hospital map for repeatable room entry.
Training a YOLO model that can tell standing, in-bed, and collapsed patients apart under simulation conditions.
End-to-end hospital mission runs under a single Mission Manager: localize → pinboard tag → room verify → door open → patient observe → YOLO report.
AprilTag-based room selection and Nav2 navigation worked together for autonomous room access in simulation.
Custom YOLO detection delivers one of three patient states for the mission outcome without manual intervention.
The system was developed as the Autonomous Robot and Mobile Systems final project: Webots hospital simulation, ROS 2 packages for localization and navigation, AprilTag nodes, Mission Manager logic, and a custom YOLO patient-state model. Launch flow: bring up the Webots TurtleBot hospital world, then start the mission controller launch that wires AprilTag, localization, navigation, and the mission manager.
Localize, read the pinboard, enter the room, observe the patient — one Mission Manager owns the full sequence.
Code and docs: github.com/atharva2406/ARAMS-hospital-challenge. Future work could include multi-room missions, richer patient taxonomies, or transferring the stack onto physical TurtleBot hardware.