Interaction Control for Collaborative Robots: An Adaptive Impedance-Based Approach in Contact-Rich and Uncertain Scenarios

The widespread adoption of collaborative robots in industrial and service sectors demands advanced control strategies capable of ensuring safe and effective physical interaction with unstructured environments. This thesis focuses on the study and development of interaction control frameworks for contact-rich manipulation tasks, with impedance control as the primary methodology to shape the robot’s dynamic behavior at the contact interface. Classical impedance control, however, relies on fixed parameters that must be tuned a priori, limiting its effectiveness when environmental stiffness or task conditions are uncertain or vary during execution. To overcome this limitation, an adaptive impedance control strategy is developed, capable of tracking a desired contact force through online modulation of the stiffness parameter, without requiring prior knowledge of the environment model or task geometry. The adaptation is driven by an update law based on the force tracking error, combined with a smooth activation function that implicitly handles the transition between free-motion and contact phases. To contextualize the contribution, a systematic fixed-parameter tuning approach is also investigated, demonstrating that even optimally tuned static parameters cannot match the flexibility of the adaptive strategy under varying contact conditions. The proposed approach is validated on the 7-DOF Franka Emika Panda collaborative manipulator, following a rigorous sim-to-real workflow: the control architecture is first developed and tested in a physics-based simulation environment and subsequently deployed on the physical robot. Experimental results across multiple trajectory scenarios and contact conditions demonstrate the effectiveness of the proposed approach in achieving simultaneous pose and force tracking, with robustness to environmental uncertainties and generalization across different force references and task conditions.