Explainable AI-Enabled Haptic Feedback Prediction for H2M Applications over Optical Access Networks

For human-to-machine applications such as teleoperation and remote robotics, ultra-low latency and high reliability are essential features to achieve immersive experience. AIenhanced servers located at the optical line terminals of optical access networks can be used to predict feedback signals, thus reducing the round-trip time in control-feedback loops. In this paper, we propose an explainable AI-enabled framework for haptic feedback prediction. Specifically, the framework harnesses a feature selection process based on Shapley additive explanations to reduce the number of features in the machine learning model such that training and inference times can be reduced. Numerical evaluations show that our framework effectively reduces the feature set while maintaining predictive performance. Consequently, the retrained models using the reduced feature set exhibits lower training and inference times, particularly in multi-task prediction scenarios.