A FORCE-SENSITIVE ROBOTIC FRAMEWORK FOR AUTOMATED BANDAGE PEELING

The removal of adhesive dressings from skin is a contact-rich manipulation that remains largely manual despite its clinical importance and the high incidence of medi- cal adhesive-related skin injury (MARSI). Automating this task requires accurate dynamic modeling together with reliable interaction-force regulation in a compliant, uncertain environment. This dissertation presents a model-based framework for robotic adhesive dressing removal by integrating manipulator identification, peeling mechanics, and interaction control on a simple four-degree-of-freedom robotic platform. The manipulator is first characterized through actuator modeling, followed by the development of model-free motion control to establish a platform suitable for dynamic identification. A physically consistent inverse dynamics model is then identified from experimental motion data using a symbolic Euler–Lagrange formulation, a friction and drivetrain structure selected by validation error—load-dependent Coulomb friction and reflected armature inertia—and constrained parameter estimation that guarantees a positive-definite inertia matrix and strictly dissipative friction. The identified model is verified through structural consistency tests and validated against independent experimental trajectories, achieving R² = 0.93 for the gravity-dominated shoulder joint. Together, these results provide the dynamic foundation required for model-based interaction control. To represent the environment, the mechanics of adhesive peeling from soft substrates are synthesized into a reduced control-oriented model that captures the coupled effects of peel angle, peel rate, and fracture energy on the required peeling force. Separate bench experiments characterize two distinct material systems. Tensile tests document the backing response of a representative commercial dressing, while peel experiments characterize the dressing used in the peel campaign. The peel measurements reveal a fracture energy that increases by a factor of 3.4, from 37 to 126 J/m², as the peel rate increases from 1 to 6.3 mm/s. The tensile results are reported as separate material characterization and are not used to assign the peel-environment stiffness. Building on these models, the proposed framework formulates computed-torque, task-space impedance, and passivity-based interaction controllers, coordinated by an adhesion-aware supervisory layer that estimates fracture-energy parameters online and adapts peel rate and angle within a prescribed deterministic force budget. As an initial hardware validation, the computed-torque controller is deployed and evaluated in free space over the full peel trajectory. A calibrated protocol and force-based evaluation quantities are specified for future closed-loop adhesive-peeling experiments. Collectively, these contributions provide a model-based foundation for force-aware robotic adhesive dressing removal.