XentricAI: A Gesture Sensing Calibration Approach Through Explainable and User-Centric AI

Gesture recognition systems offering contactless human-machine interaction have diverse applications, from smart homes to healthcare. However, they often face challenges from unexpected changes in user behavior and a lack of explainability, especially concerning fields like medical diagnosis or security systems. To address these issues, we introduce a novel approach that exploits advances in Explainable Artificial Intelligence (AI) and Experience Replay techniques for human-centric AI in radar-based gesture sensing. Our contributions include model calibration via Transfer Learning using Experience Replay and feedback on anomalous gestures through feature analysis with Explainable AI. Experimental results show improved accuracy, low forgetting rate, and enhanced user engagement, suggesting the potential for fostering trust in AI technology. The model calibration leads to an average accuracy improvement of 5.4% with respect to the uncalibrated model. Furthermore, leveraging the Explainable AI feedback to enhance gesture execution yields a 38.1% average accuracy improvement compared to unguided user behavior.