Evolutionary Neurocontrol for Spacecraft Proximity Operations

Authors

Cole George
Joshua A. Hess
Richard G. Cobb, Air Force Institute of Technology

Document Type

Conference Proceeding

Publication Date

1-13-2019

Abstract

Recent research has demonstrated the potential of evolutionary strategies to the solution of deep reinforcement learning problems. In this paper, such techniques are applied to the control of relative spacecraft motion. Open-loop and closed-loop feedback controllers, represented by multi-layer artificial neural networks, are developed using evolutionary optimization. The implementation and performance of these so-called evolutionary neurocontrollers are compared to classical optimal and nonlinear control techniques. For this problem space, recommendations are made on appropriate network hyperparameters and input-output representations. The resulting analysis is expected to provide an alternative control approach for spacecraft proximity operations.

Comments

Conference Session 27: Proximity Missions & Formation Flying III

This conference paper was later published in Advances in the Astronautical Sciences, Volume 168, Pages 3453-3469, 2019, with the paper title "Open-and closed-loop neural network control for proximal spacecraft maneuvers"

Source Publication

29th AAS/AIAA Space Flight Mechanics Meeting

Recommended Citation

George, C., Hess, J. A., & Cobb, R. A. (2019). Evolutionary Neurocontrol for Spacecraft Proximity Operations. 29th AAS/AIAA Space Flight Mechanics Meeting, Session 27: Proximity Missions & Formation Flying III, AAS 19-325.