GNSSMachine

Machine Learning as a basis for more intelligent navigation algorithms?

Aims

GNSSMachine is an in-depth feasibility study on the benefits of using Machine Learning methods in select navigation problems. The goal is to gain insight into the latest advancements of Machine Learning, to apply them to selected navigation problems, and to decide on a further strategy.

Achievements

Multiple models for multipath and spoofing detection were developed and tested successfully. Since high quality data is always hard to acquire for free, multiple datasets demonstrating different spoofing scenarios were also simulated and labelled in house. The acquired expertise was shared in the final report submitted to FFG, explaining the procedures and results.

Brief Description

Machine Learning has led to great advances and breakthroughs in a variety of different domains. However, until very recently it had hardly been used in applications concerning navigation and GNSS, a domain rich with many challenges that can be tackled effectively using Machine Learning (ML). Today, such challenges are mainly addressed using rule-based approaches, which are hand-coded by developers and engineers. These approaches have the disadvantage of being tedious to implement and adapt and become  increasingly complex when adding new parameters. ML has the potential to significantly increase the performance of GNSS algorithms and realise solutions for future problems that are unthinkable as of today.

In this project, concrete problems in the field of navigation and GNSS were determined and tackled using ML approaches. This project is a first step in development of  future strategies for coping with the current wave of technological advances imposed by ML.
We described three problems that we aimed to solve using machine learning techniques: Multipath classification, spoofing detection, and learning of cost functions for off-road routing. The results will lead to an in-depth feasibility study that will lead to future strategies on incorporation of machine learning methods into current applications.