Abstract: In human-robot interaction scenarios such as medical rehabilitation and cosmetic care, a robot must maintain a stable contact posture normal to the skin surface. The highly compliant, non-uniform contours of human tissue—together with posture changes and real-time deformation during operation—severely limit conventional attitude-tracking performance. To address this challenge, we establish multiple auxiliary coordinate frames on the force-motion interaction system, describe and analyze the kinematics as well as the force/torque relationships during contact, and construct the corresponding transformation matrices. By combining Hertzian elastic contact theory with a biomechanical adhesion–friction model, we develop a normal-vector relationship model for a rigid spherical end-effector interacting with soft tissue, based on six-axis force measurements, and the model solves in real-time to obtain the current normal attitude. To ensure the accuracy of the six-axis force data, a dual compensation scheme—secondary gravity compensation and periodic torque-error compensation—is implemented. The proposed method enables real-time estimation of the unknown surface normal of soft human tissue through force-sensor feedback. Experiments on a facial model, tracking the trajectory from the glabella along the nasal dorsum to the tip, demonstrate that the normal-attitude error remains within 1.12°–3.2°under an impedance controller that regulates a compliant normal force. These results validate the effectiveness of the control strategy and enhance the robot’s adaptability in unstructured human-interaction environments.

Key words: unstructured human environments,  , normal posture tracking, force feedback