The first three contributions focus on embedded on-device solutions and our contributions pave the way from 1) testing the efficacy of small and efficient machine learning models (e.g., support-vector machine (SVM)), 2) testing the consistency of different machine learning models (SVM and XGBoost trees) in different environments, and 3) enabling NLOS classification, error correction, and anchor selection on resource-constrained UWB devices using (SVM, XGBoost, and convolutional neural networks).