Abstract

Global navigation satellite systems (GNSS) are a very well established technology, which provides a rich and flexible framework for many different positioning applications. The most familiar uses are standard positioning applications which usually provides accuracy of a few meters, but can sometimes give very wrong answers. With some care and extra external resources, GNSS can also provide very high accuracy (centimetre scale) positioning by exploiting carrier phase measurements. In other cases (e.g. aircraft landing systems) we need safety critical GNSS systems; in practice this is done by ignoring carrier phase observations, and using highly conservative inference frameworks which sacrifice accuracy in return for safety. For self-driving vehicle applications, none of these existing systems are really suitable: we want both accuracy and safety, but conventional approaches can not deliver both at the same time.

In this talk, I argue that non-Gaussian error models are the key ingredient needed to build future GNSS systems which can be both safe and accurate. Although considerable care and insight is needed to use these models, the pay-off is a system which can exploit the high accuracy of carrier phase observations, without introducing false assumptions which can lead to unsafe performance. I also show the practical engineering steps needed to overcome the formidable computational challenge this approach brings when applied to a real-time system.

Link to join virtually: https://cl-cam-ac-uk.zoom.us/j/97767639783?pwd=T09GcVJxZUNEUFEvRnZnbWwxeEwzQT09

A recording of this talk is available at the following link: https://www.cl.cam.ac.uk/seminars/wednesday/video/