A Machine Learning-Based GNSS Performance Prediction For Urban Air Mobility Using Environment Recognition

Abstract

Urban Air Mobility (UAM) has gotten a lot of attention because it could change the way people get around in cities with lots of people. But getting UAM cars to navigate reliably is still hard, especially since Global Navigation Satellite Systems (GNSS) aren't perfect in these situations. GNSS performance is often hampered by infrastructure in cities, signal blockages, and interference. This paper suggests a new way to predict GNSS success in UAM systems using machine learning and environment recognition methods together. The suggested model takes into account a lot of external factors that affect the quality of GNSS signals, like the number of buildings, plants, temperature, and other barriers. Based on these external factors, machine learning methods like regression models and support vector machines are used to look at and guess how well GNSS will work. GNSS systems built into UAM cars and weather sensors are used to collect data for the method. Both GNSS and weather data are preprocessed with feature extraction methods. This data is then put into machine learning models. Key measures like GNSS accuracy and forecast accuracy in different urban settings are used to judge the performance of the model. A case study shows how the model can be used in a real-life UAM situation, showing how well it improves the accuracy of GNSS performance predictions. The results show that adding environmental detection makes GNSS performance much more reliable in complex urban settings. This study helps UAM systems get better by giving a strong answer to one of the main problems these cars have: making sure they can navigate consistently and correctly. The study also talks about areas that could be studied in the future, such as adding more outdoor factors and better machine learning methods for predicting GNSS in real time.