DEVELOPMENT AND VALIDATION OF AN INTEGRATION AND SUPERVISION FRAMEWORK FOR COLLABORATIVE ROBOTS

Medical robotic systems require consistent integration of manipulators, sensing, control, and distributed software before experimentation on physical platforms. This thesis, developed within an industrial research collaboration in biomedical robotics, addresses the software integration and supervision of a KUKA LBR Med14 R820 manipulator through an external ROS 2 environment. The work first analyses the native KUKA Sunrise ecosystem, focusing on motion primitives, event-driven conditions, state machines, reference frames, calibration, and the Fast Robot Interface (FRI). It then consolidates a reproducible ROS 2 Jazzy workspace on WSL2 and Docker, integrating the LBR stack, ros2_control, mock mode, Gazebo, and RViz2, while keeping visualization, dynamic simulation, and future hardware communication conceptually distinct. The main software contribution is lbr_supervisor, a custom ROS 2 package that subscribes to /lbr/lbr_state, stores the latest joint configuration, produces periodic diagnostics, counts received messages, and detects missing or interrupted state streams. Because mock mode and Gazebo expose sensor_msgs/JointState rather than the FRI-specific lbr_fri_idl/msg/LBRState message, a separate JointState-LBRState compatibility bridge was implemented exclusively for simulation tests. This separation preserves the supervisor's compatibility with the interface expected in hardware mode without embedding simulator-specific logic. The infrastructure was verified through an incremental protocol comprising message-type inspection, manual publication tests, mock execution, headless Gazebo simulation, gz_ros2_control and controller verification, activation of the complete Gazebo-bridge-supervisor chain, reversible joint-motion commands, and graphical observation in RViz2. The tests confirmed the functional consistency of the software chain within the simulated scope and showed that the supervisor follows actual state evolution rather than a static configuration. The obtained results do not constitute clinical, safety, metrological, or physical-robot validation, nor does the compatibility bridge reproduce complete FRI semantics. The next stage is therefore a cautious transition to the real manipulator, beginning with network and FRI-session verification and passive read-only state acquisition before any controlled motion. Overall, the thesis provides a modular, reproducible, observable, and progressively validated basis for future application-specific developments involving the KUKA LBR Med platform.