Self-supervised Machine Learning Based Approach to Orbit Modelling Applied to Space Traffic Management

In much the same way that different STM tasks This paper presents a novel methodology for improving rely on our ability to accurately model orbits, different the performance of machine learning based NLP tasks, such as next word prediction or space traffic management tasks through the use of sentiment analysis, rely on an underlying common a pre-trained orbit model. Taking inspiration from understanding of how to model language. The latest BERT-like self-supervised language models in the breakthroughs in this field can be attributed to field of natural language processing, we introduce the use of self-supervised learning (SSL), whereby ORBERT, and demonstrate the ability of such a high performance underlying language models such model to leverage large quantities of readily available as Google's BERT [2] can be pre-trained on extensive orbit data to learn meaningful representations datasets by predicting masked words from text, that can be used to aid in downstream tasks. As a before being fine-tuned to the objectives and data proof of concept of this approach we consider the of specific downstream tasks, thus improving both task of all vs. all conjunction screening, phrased performance and efficiency. Here, we present our here as a machine learning time series classification proposed approach for applying this concept to task. We show that leveraging unlabelled orbit the STM domain, and introduce our pre-trained data leads to improved performance, and that the orbit model ORBERT, as well as its application proposed approach can be particularly beneficial to the downstream task of conjunction screening.