Design of a ROS 2-based visual trajectory planner with analytical modeling and data-driven friction refinement

This thesis presents a ROS 2-based framework for the generation, execution, and dynamic validation of robot trajectories in an industrial robotics context. The goal is to ensure continuity between interactive planning, simulation-based verification, real-robot execution, and experimental analysis. The first contribution is a Visual Planner based on RViz2 and MoveIt 2, which allows an operator to define waypoints interactively and convert them into fully timeparameterized trajectories. The generated motion can be represented in both joint space and Cartesian workspace and is exported as a CSV file, providing an explicit and reusable trajectory description for simulation and execution. The second contribution is a Trajectory Executor capable of replaying pre-sampled references through the ROS 2 control framework. Unlike controller-side interpolation approaches, this solution preserves the timing and smoothness properties defined during trajectory generation and enables the acquisition of simulated motion quantities, including positions, velocities, accelerations, jerk, and efforts. The main contribution of the thesis is the simulation-to-real validation of the dynamic model. Torques obtained in Gazebo from the URDF/Xacro-based robot model are compared with torque feedback measured on the physical robot during the execution of the same trajectory. Since the simulated torques mainly reflect rigid-body dynamics, the residual is interpreted primarily as the effect of friction. An offline refinement procedure is therefore introduced, first with a dynamic friction term and then with an additional static component. The coefficients are identified on one trajectory and evaluated on a second one. The results show that the proposed workflow provides a coherent link between planning, simulation, physical execution, and model refinement, significantly improving the agreement between simulated and measured torques.