Location and path planning of dorsal acupoints for a robotic massage system

In order to address the problems associated with manual acupoint massage, a novel integrated robotic dorsal acupoint massage system that combines visual acupoint localization with massage path planning is proposed in this paper. In this massage system, computer vision is employed to locate the dorsal acupoints positions which are then used for robotic path planning, followed by the execution of massage operations via robot end-effector. Methodologically, the primary contributions of this study are outlined as follows. Firstly, an adaptive brightness mapping data augmentation strategy for the training phase that can generate non-uniform illumination data, is proposed to mitigate illumination distribution discrepancies between experimental and real-world environments. Secondly, a state-space network equipped with a flexible multiplicative gating mechanism is developed to enable global receptive field modeling and enhance feature selection capability while maintaining linear computational complexity. Thirdly, a structural alignment loss is put forward to constrain the relative geometric relationships among keypoints, thereby suppressing structural distortion and improving acupoint detection accuracy. Fourthly, a quality-aware adaptive local reconstruction mechanism has been incorporated into the standard Ant Colony Optimization with the objective of generating shorter and smoother acupoint massage paths without additional priors. In this study, the visual acupoint localization and massage path-planning modules are sequentially integrated, enabling the detected acupoint coordinates to directly serve as the input for subsequent robotic path planning. A comparison of the proposed system with representative related works reveals that it achieves higher accuracy and stability in acupoint location and significantly improves path planning quality, demonstrating excellent practical performance.