This study proposes a robotic massage system combining visual and tactile inspection for personalized therapy. Equipped with RGB-D cameras, a robot arm, and a force sensor, the system identifies massage areas based on body type via skeleton tracking. Using the kneading exploration method, muscle knots are localized efficiently, reducing location errors by up to 85.7% compared to conventional methods. The system also leverages VR visualization, showing massage actions through a robot arm and human-like therapist avatars. A study with ten participants found that human-like avatars, particularly a boy avatar, received higher ratings for pleasantness than a robot arm visualization, while pain perception showed no significant differences. These results highlight the influence of visual representation on psychological evaluations, enhancing the user experience. Future work will focus on improving interaction methods and integrating advanced physiological feedback to develop an adaptive Human-Robot Interaction system for physical and psychological care.