Machine Learning for Trajectory Design in Multi-Body Dynamics
Background
Exploiting natural structures of multi-body dynamical systems for trajectory design can lead to substantial savings in propellant and reduced mission complexity. In Circular Restricted Three Body Problems, stable and unstable invariant manifolds give rise to low-energy interplanetary pathways that could transform deep-space mission analysis.[1] Past and current missions like Hiten, Genesis and BepiColombo make use of weak stability boundaries and low-energy interplanetary transport networks to significantly reduce the cost of orbital maneuvers and achieve what would have been impossible under traditional Keplerian assumptions.[2]
As of today, this low-energy trajectory design framework is however unsuitable for real-time optimization and fast transfer cost estimation due to:
- The computational cost of finding intersections between stable and unstable structures emerging from orbits;
- The cost and intuition required to perform orbit cartography given a multi-body system of interest.
Project goals
References
[1] Ross, S.D et al. “Dynamical Systems, the Three-Body Problem and Space Mission Design”, (2011)
[2] Koon, W.S. et al. “The Genesis Trajectory and Heteroclinic Connections” (2000)
[3] S. De Smet, D.J. Scheeres, “Identifying heteroclinic connections using artificial neural networks”, Acta Astronaut. 161 (2019) 192–199