In this study, an AI robot uses a camera and computer to perform face recognition and uses non-contact image physiological signal measurement technology to predict heartbeat and blood pressure. The predicted heartbeat and blood pressure are displayed on the robot tablet and are transmitted to a cloud database to assist healthcare management. In this study, we use RGB images to extract facial features and points of interest of the face and palm. The changes in vasoconstriction at these points of interest reflect the relationship between the absorption of light by blood, heartbeat, and blood pressure. Using the photoplethysmography (PPG) signal of the green channel in the RGB image through a convolutional neural network (CNN), deep learning technology can predict heartbeat and blood pressure values and even determine whether a subject has arrhythmia. Our results demonstrate that the predicted heart rate and blood pressure errors are 2.6% and 1.7%, respectively. The AI companion robot in this study can obtain the subject’s physical information by a non-contact method, reducing anxiety and the cost of labor in medical care.