To Supervise or not - ML based UWB Obstacle Detection

Abstract:
In a previous paper [1] we presented methods for integrating ranging functionality into radio communication systems and using machine learning (ML) algorithms to identify obstacles in a train pairing scenario. Further investigations have been made to a test a broader spectrum of ML methods in an effort to identify the best suited solution for a scenario where automated approaches of two vehicles potentially pose a safety problem to nearby people. In order for two vehicles to safely pair or couple together automatically we have to consider the possibility of people accidentally stepping in the way. A method of detecting an obstacle in between is needed to mitigate the threat. A wide variety of methods have been tested and can be classified as either supervised or unsupervised. In this paper we show how the evaluation of the different methods are done, present the metrics by which they are measured against each other and how each method scored. The paper concludes with the supervised learning methods as the best fit for the posed problem.