Bayesian Learning of Impedance Control for Force-Constrained Contact Interactions: An application to Compliant Robotic Systems

This thesis investigates the design, implementation, and experimental validation of an impedance controller for the Franka Emika Panda collaborative robot, with a specific focus on contact-rich tasks where interaction forces act as constraints on the task execution. The work addresses advanced parameter tuning strategies aimed at ensuring both accurate motion tracking and compliant behavior during physical interaction with the environment. Two main approaches to impedance parameter tuning are explored. The first approach is based on Bayesian Optimization, where Gaussian Process models are employed to automatically tune the impedance parameters. The objective is to achieve accurate tracking of a circular trajectory while maintaining the contact force within a predefined desired range. This method allows for a data-efficient and systematic exploration of the parameter space, balancing trajectory tracking performance and force regulation. The second approach focuses on a dynamic tuning strategy in which impedance parameters are adapted online to enforce a constant desired contact force along a linear trajectory that is not precisely defined. This setup is designed to emulate interaction with discontinuous or uncertain surfaces, highlighting the controller’s ability to adapt to environmental variations and modeling uncertainties. All proposed control strategies are implemented and evaluated on a real robotic setup using the Franka Emika Panda platform. Experimental results demonstrate the effectiveness of the proposed tuning methods in achieving stable and compliant contact interactions while satisfying force constraints, confirming their applicability to real-world collaborative robotic tasks.