|Abstract:||Emerging applications in the automotive environment (ADAS, autonomous driving, etc.), are requiring high GNSS accuracy in conjunction with high reliability. For that reason, this kind of applications are demanding for GNSS receivers with enhanced safety, as specified by the automotive functional safety standards, and in particular by the ISO 26262 standard. In order to meet this requirement, an additional security, integrity and safety (SIS) layer has been added to the GNSS receiver to monitor, detect, estimate the risk, and switch to a safe state when integrity appears to be compromised. In this contribution, the evaluation and detection of the risk will be implemented using a two different machine learning approaches, i.e. using two different kind of deep neural network (DNN), the first using supervised learning and the second using unsupervised learning. Machine learning is used in many ADAS systems, and it is an enabling feature for the overall autonomous driving industry. Moreover, the complexity of the network for both approaches will be kept into account, because they have to be deployed on a GNSS receiver hardware, which has memory and computational capacity constraints. At the end, the performances of the two proposed approaches will be illustrated, and the results will be compared to a logistic regression solution used as baseline, to show the large improvement in accuracy provided by both solutions. Finally, the advantages and disadvantages of each approach and the future development of such approaches will be discussed.|
Proceedings of the 32nd International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2019)
September 16 - 20, 2019
Hyatt Regency Miami
|Pages:||1738 - 1752|
|Cite this article:||
Gogliettino, Giovanni, Renna, Michele, Pisoni, Fabio, Di Grazia, Domenico, Pau, Danilo, "A Machine Learning Approach to GNSS Functional Safety," Proceedings of the 32nd International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2019), Miami, Florida, September 2019, pp. 1738-1752.
ION Members/Non-Members: 1 Download Credit