Hybrid force–motion control with learning-augmented surface-normal estimation for contact-rich robotic swabbing
This paper presents an operational-space framework for regulating contact forces on geometrically unknown surfaces by integrating online surface-normal estimation with hybrid motion–force control via task-space admittance. Cartesian motion is decomposed into normal and tangential components defined with respect to an estimated local surface frame. Surface normals are computed using a combination of force–velocity-based estimation with friction compensation, local quadratic surface fitting, and a regression-based prediction method. The estimated normal is used to define the force-control direction and to align the end-effector orientation during contact. The approach is validated in simulation on a cylindrical surface and experimentally on a UR5 manipulator equipped with a wrist-mounted force/torque sensor. Results demonstrate sub-$3^\circ$ mean angular error under ideal conditions and consistent reductions in angular misalignment and RMSE when geometric and regression-based refinements are incorporated, while maintaining stable force tracking. The proposed framework enables robust surface interaction for contact-intensive tasks such as robotic swabbing.