BEMA: Behaviour and Emotion Matrix Assistant

BEMA: Behaviour and Emotion Matrix Assistant BEMA is a facial emotion recognition and robot behaviour mapping framework developed for human–robot interaction applications. The project combines a Vision Transformer facial emotion classification model with a configurable matrix that maps detected human facial expressions and robot interaction states to suitable robot responses. The facial emotion recognition model identifies 5 classes: Angry Happy Neutral Sad Surprise BEMA includes the following robot visual and behavioural states: Neutral Happy Sad Angry Surprised Listening Speaking The Listening, Speaking states are controlled by the robot application and are not direct outputs of the facial emotion classifier. How BEMA Works The typical system workflow is: Human face image → Image preprocessing → Vision Transformer classification → Predicted emotion and confidence score → BEMA behaviour and emotion matrix → Robot expression or behavioural response The robot does not always directly imitate the detected emotion. For example, when an angry expression is detected, the robot may display a neutral or concerned response instead of showing anger. Model Source BEMA uses the pretrained model: dima806/facial_emotions_image_detection The model is based on: google/vit-base-patch16-224-in21k Potential Applications BEMA can support: Social robots Wayfinding robots Reception robots Customer service robots Educational robots Interactive kiosks Digital avatars Emotion-aware interfaces Human–robot interaction research Limitations Predictions may be affected by lighting, camera angle, image quality, facial obstruction, cultural differences, and expressions outside the supported classes. The system should not be used for medical diagnosis, psychological assessment, recruitment, employee monitoring, law enforcement, or other high-impact decisions. Licence This project is released under the Apache License 2.0. External models, code, images, and supporting assets remain subject to their original licences and attribution requirements.