Gaining Insights into a Robot Localization Monitor Using Explainable Artificial Intelligence

Monitoring the state of a localization component in robotic systems has received increasing attention in recent years, as navigation behaviors of robots rely on a reliable pose estimation to a large extend. Nowadays, research focuses on the development of new approaches to monitor the localization state of a robot. Many of those approaches use Machine Learning techniques which do not provide direct insight into the decision making process and are thus often handled as a black box. In this work, we aim to open this black box by making use of an Explainable Artificial Intelligence (XAI) framework that allows us to improve the understanding of a machine learning based localization monitor. To gain insights into the machine learning model, we make use of the open-source framework SHapley Additive exPlanations (SHAP). Results show that investigations in the model structure of a localization monitor using XAI helps to improve the model’s transparency. Overall, XAI proves to be useful in understanding the decision-making process of a localization monitor and can even help to improve the model’s design quality.