This work aims to evaluate the recovery of arm mobility in patients undergoing rehabilitation by leveraging artificial intelligence. The proposed system, “Roboterap-IA,” integrates an arm rehabilitation device, a serious game and biosensors to indirectly measure pain and stress levels that are key factors that influence rehabilitation progress. Pain and stress are inherently subjective and challenging to quantify due to individual variability in thresholds. To address this, the system collects significant physiological signals such as galvanic skin response and heart rate to estimate stress and pain levels. These metrics are combined with the patient’s performance in the serious game to train an artificial neural network capable of evaluating mobility recovery. For the training, experiments were conducted with healthy individuals whose mobility levels were simulated using bandages to restrict movement. The objectives of this research are to identify meaningful biosignal patterns related to stress and pain and to develop an AI-based approach to calculate arm mobility recovery. This integration of robotherapy, gamification, and AI has the potential to innovate rehabilitation by providing personalized and objective progress assessments. The implementation of a neural network to process collected data ensures accurate mobility evaluations and more reliable predictions of patient needs, achieving up to 96.92% accuracy.