From biomechanical data to robotic execution: design and control of an intelligent Tuina system

Introduction Traditional Chinese medicine (TCM) Tuina (massage) relies heavily on practitioners' physical stamina and subjective experience, leading to operational variability that hinders therapeutic standardization and scalable rehabilitation service delivery. To address this limitation, this study proposes an intelligent TCM Tuina robotic system integrating biomechanical data-driven design and hybrid control strategies. Methods A 6-DOF manipulator equipped with a biomimetic end-effector is developed, and a CMPGA-Corr neural network (Convolutional Multi-branch Prototype-Guided Architecture with a Channel-Correlation Branch) is constructed to decouple the spatiotemporal features of nine Tuina manipulations from 25-channel tactile data. Robotic execution is implemented within a Robot Operating System 2 (ROS 2)-based control framework. The framework combines RRT-Connect path planning, S-curve speed profiling, and RBF-compensated admittance control. PID, conventional Admittance, and RBF controllers were compared in complete-process simulations using five repetitions per manipulation. Results The RBF controller produced the lowest mean RMSE and MAE for all six evaluated manipulations. Physical experiments on a silicone back phantom further demonstrated bounded and repeatable force-tracking behavior under the tested conditions. Discussion The proposed framework bridges TCM practice with modern robotic engineering and provides a scalable and standardized solution for precision rehabilitation robotics.