Integration of facial expressions recognition and navigation output styles into a fuzzy logic emotional model for a mobile robot assistant

The pursuit of a more natural and empathetic human–robot interaction represents one of the key challenges in contemporary social robotics. This thesis presents the design and implementation of an emotional control system for the TIAGo robot, developed as an extension of previously proposed fuzzy‑logic emotional frameworks and carried out under the supervision of a psychologist. The system integrates real-time facial expression recognition with a fuzzy-logic emotional model that dynamically modulates the robot’s navigation parameters. The goal is to enable the robot to exhibit adaptive and lifelike behaviour by adjusting its motion according to the emotions perceived in the user and its own internal affective state. An initial version of the emotional control system was first implemented and tested in a pilot experiment with 8 participants, which produced preliminary evaluations and revealed some technical limitations (e.g., latency in the real‑time image acquisition and processing pipeline, and navigation parameter updates that were too smooth to be clearly perceivable). Based on the results obtained, targeted improvements were introduced to develop an optimized system, which was then validated on a larger sample of 28 participants, ensuring higher statistical robustness and reliability of the results. The proposed system defines two internal emotional dimensions for the robot, mood and alertness, which evolve in response to the emotions detected on the user’s face. These internal states, processed through two fuzzy state tables, influence two corresponding navigation outputs: speed (linked to mood) and tuning (linked to alertness), the latter regulating angular velocity and proximity to obstacles. In this way, the robot’s movement becomes an external manifestation of its internal affective state, producing behaviour that more closely resembles that of a living being. Two distinct robot personalities, “shy” and “intense”, were implemented by differentiating the fuzzy rules and the amplitude of the emotional reactions. The shy configuration is characterised by smoother and slower responses, while the intense one reacts with faster and more marked behavioural changes. The entire system was developed within the ROS framework and integrated on the TIAGo platform at the Universidad Politécnica de Madrid. Real-time emotion detection was achieved using a convolutional neural network (CNN) trained on facial datasets, interfaced with the robot through dedicated Python and ROS nodes. The experimental validation involved human participants who interacted with the robot under two conditions: neutral mode (without emotional modulation) and personalised mode (with the emotional system active). Participants then completed the Godspeed Questionnaire (Animacy and Likeability scales) and the EVEA self‑report emotion scale. Statistical analyses using mixed ANOVA showed significant improvements in perceived animacy and likeability for the personalised mode, confirming that integrating emotional dynamics enhances the perceived naturalness and pleasantness of TIAGo. Furthermore, qualitative data revealed that participants perceived the shy robot as more delicate and empathetic, while the intense one appeared more expressive and engaging. The proposed system allows for intuitive and modular adaptation of robot’s personality traits, laying the groundwork for future developments in emotionally adaptive and socially intelligent robotics.