RoboCup Rescue Bot
1st-place national winner at RoboCupJunior Rescue Simulation 2026. Fully autonomous robot built in Python on the Webots/Erebus simulator — navigates unknown environments, detects victims via YOLO, maps terrain with occupancy grids, and reports findings to the competition system in real time. Qualified for World Championship 2026.
Built for RoboCupJunior Rescue Simulation 2026, this autonomous robot operates across four integrated software modules: perception, navigation, mapping, and communication. The sensor suite — LiDAR, GPS, IMU, colour sensor, and a dual lateral camera configuration — feeds a pipeline that runs continuously as the robot explores. Navigation uses DFS to guarantee full coverage of unknown areas and BFS to compute optimal return paths between nodes; heading is held by a proportional controller driven by IMU and GPS fusion. Victim detection runs a YOLO model in real time on the camera streams, while hazardous materials are identified through deterministic chromatic analysis. All findings — coordinates, victim classifications, and occupancy maps — are automatically transmitted to the competition system via the referee API.
Exploration efficiency and victim recall had to be maximised within a strict time budget, with no prior knowledge of the environment. Fusing GPS and IMU data reliably under simulation noise required tuning the proportional heading controller to remain stable without overshoot on sharp turns. Integrating YOLO inference into a tight control loop without introducing latency spikes that disrupted navigation demanded careful frame-rate throttling and asynchronous inference queuing. Building a consistent occupancy grid from LiDAR sweeps while the robot was moving — without a fixed origin — required continuous re-registration of scan data against the growing map, which amplified any heading error. Synchronising the four modules so that mapping, perception, and path planning never blocked each other was the central systems integration challenge.
1st place at the national RoboCupJunior Rescue Simulation competition, earning qualification to the RoboCup World Championship in Incheon, South Korea, scheduled for July 2026. The project deepened expertise in autonomous robotics, AI-driven perception, real-time computer vision, and the system-level integration required to make all four modules operate reliably under competition conditions.