ML-GNSS Research
I was awarded the Lassonde Undergraduate Research Award for both the Summer 2023 term and the current Fall/Winter 2023/2024 terms. These awards allows undergraduate students to conduct research within one of the Science or Engineering labs at York Univieristy
I have been working in the Global Navigation Satellite System (GNSS) Lab York University under the supervision of Professor Sunil Bisnath, Ph.D., P.Eng. Click Here to learn more about the GNSS Lab. My research looks into the applications of machine-learning for the improvement of smartphone navigation. I am currently working to address the drawbacks from summer project and create a more robust solution.
Summer 2023 Project: Application of Machine Learning Algorithms to Smartphone Satellite Navigation Data for Precise Positioning
This project aims to learn about the feasibility of implementing AI tools, specifically machine learning, into the processing of smartphone positioning data to improve its accuracy for driving navigation. Fig. 1 shows the wide range of errors in smartphone positioning. The smartphone solution is compared against a corresponding reference (“true”) trajectory coordinates determined with high-precision equipment. This chart shows that the average smartphone horizontal error is 3.8 meters. During interrupted satellite signals, such as when the car moves under an overpass or bridge, the error can shoot up beyond 10
meters.
Three ML models were trained and tested to see how they performed. The results below highlight the most successful run. The figure shows 4 plots; the actual classification (leftmost plot) and the predictions made by the three different ML models. The actual classification represents the answer. The 3 other plots show each of the models outputs. We compare theses outputs to the answer to identify the model's accuracy.
The KNN algorithm performed the best of the three, correctly predicting 92% of data points. In conclusion, this experiment shows ML models have the potential to be implemented into GNSS solutions to improve position accuracy.
