Project Ref: NGCM-0087
Supervisor: Hodei Urrutxua
Academic Unit: AACE
Research Group: Astronautics
Co-supervisor: Jonathon Hare
Academic Unit: Vision, Learning and Control
Research Area: Computational Engineering
Project Description: As space missions become increasingly complex and robotic missions exceedingly ambitious, there is a growing need for autonomous capabilities on-board spacecraft. One such example is the close proximity operations around non-cooperative bodies, which is a key technology required to enable robotic active space debris removal or satellite refuelling missions. Autonomous navigation and automatic decision-making are crucial capabilities for critical parts of the mission. These concepts not only involve challenges in close formation flying, but eventually docking or capture operations as well. For such operations, extremely accurate sensing techniques will be needed to provide on-board, precise, real-time estimates of the spacecraft dynamical state relative to the target body.
Though the working principles of navigation sensors are relatively mature after decades of in-orbit rendez-vous and docking operations, recent advances in artificial intelligence (AI) and machine learning technologies enable previously unimaginable tasks to be effectively, efficiently and reliably accomplished by computers. In fact, machine vision algorithms for navigation purposes have been successfully developed and deployed for many terrestrial applications, e.g. visual odometry for cars, state estimation for unmanned aerial vehicles, etc. The latest breakthrough in the field of machine learning is coming from "deep learning" technologies, which attempt to model high-level abstractions in data and can naturally be applied to computer vision. The applications and ramifications of deep learning technologies are immense and novel advances in computer vision could play a significant role in the field of autonomous guidance, control & navigation systems. This PhD project will study deep learning technologies and explore how they can be incorporated into spacecraft guidance and navigation systems to improve the currently available capabilities and algorithms.
In scenarios of active space debris removal, these technologies could not only improve the automatic detection of spacecraft shapes, but even enable structural integrity checks and risk assessments before attempting a proximity manoeuvre, or get accurate estimates of their exact centre of mass, pose, rotational state, superficial features, and a long range of information, which can then be fed to the on-board control, decision-making and manoeuvre scheduling algorithms.
Thus, this research project pursues to make major contributions developing innovative algorithms that exploit the potential of modern machine learning and deep learning technologies applied to computer vision for proximity operations in space.
Keywords: Computational Modelling, Computational Engineering, Aeronautical Engineering, Computer Science, Software Engineering
Support: All studentships provide access to our unique facilities and training and research support .