This was a project for the Multidisciplinary course at TU Delft, delivered by our team. The client problem was inspired by the Lely Discovery Collector barn-cleaning robot.

Overview
The goal was to automate barn cleaning with reliable localization, obstacle avoidance (cows, humans, walls), and multi-robot coordination, for maximum coverage and manure detection.

Contributions

• Developed a system with a single exploration and mapping robot for map generation followed by coverage using all robots that use local planners to avoid dynamic obstacles.

• A central planner formulates a routing/coverage problem. Implemented in MiniZinc and solved with OR-Tools CP-SAT for faster valid/optimal solutions.

• The implemented ROS stack uses, navigator node to follow global paths via move_base, mapping_node to launch slam_toolbox, communications_node bridging ROS and local server over MQTT for coordination between multiple robots.

• Developed desktop planning UI to define waypoints/resources and visualize maps/states, a mobile GUI for monitoring, teleop, and E-stop.

Results

• Successfully implemented and coordinated 3 mobile manipulators to navigate a barn with resource-aware route planning, reactive local planning around dynamic obstacles and robust localization with LiDAR.

• Awarded the most robust solution.

Languages and tools used: ROS (move_base, slam_toolbox), C++, Python, MQTT, pygame, NumPy.

My contributions were, designing and implementing the behavior architecture that included integrating mapping, obstacle detection and navigation with a state machine.