Orbit and uncertainty propagation using factor graph techniques
Project ID: 2228cd1275 (You will need this ID for your application)
Research Theme: Artificial Intelligence and Robotics
UCL Lead department: Civil, Environmental and Geomatic Engineering (CEGE)
Lead Supervisor: Santosh Bhattarai
Project Summary:
Human activity in space is growing at an accelerating pace – the expected number of active satellites is predicted to be 50,000 by 2030. However, today’s space surveillance and tracking systems are not equipped to support future space traffic monitoring and management requirements. The key limitation is their ability to accurately predict the trajectories of objects on orbit and the associated uncertainties. Standard approaches, based on variants of the Kalman filter and least squares methods, are inefficient and sub-optimal, specifically when dealing with non-linear systems and complex (non-Gaussian) noise processes.
Challenging estimation problems arises in many other domains. In Simultaneous Localization and Mapping, for example, a platform builds a map of the environment while using this map to localise itself. The system is detectable, high-dimensional and contains very complicated non-linear dependencies. Decades of experience have shown that filter-based methods behave poorly. As a result, almost all mapping systems – from robots to Google Street View – use factor graphs and maximum likelihood estimation. Factor graphs explicitly write out the Markov chain and are completely general. Maximum likelihood estimation uses efficient, mature and scalable implementations of optimisation algorithms to estimate systems with thousands of states at hundreds of Hzs.
Factor graphs have revolutionised the way mapping and tracking is done in robotics. This project will investigate their application in the area of space surveillance and tracking.
The student will develop a factor graph based solution to the orbit determination problem. The technique developed will be compared with established methods to assess its performance (across a range of orbital regimes) on a variety of metrics, e.g., prediction accuracy, uncertainty realism and computational efficiency. We are seeking applicants with a strong academic background in a numerate (STEM) discipline.