Several missions require transfers to multiple targets in space; examples include asteroid tours and spacecraft servicing. Designing low-thrust trajectories to accomplish these transfers can be complex, as it typically results in a dynamic traveling salesman problem. A key aspect to reduce the complexity of the problem is the ability to compute low-thrust transfers in a very efficient way, enabling the rapid evaluation of many solutions.
This project aims to achieve this objective through supervised learning. A large database of low-thrust transfers will be produced using convex optimization techniques. Subsequently, students will need to develop a framework to learn key transfer properties via neural networks. This neural network can then be used as a means to estimate the feasibility and cost of object-to-object transfers, acting as a building block for multi-target transfer design
Undergraduate
None
No lab has been assigned to this project