Uncertainty management in solar sail attitude control

Solar sails are emerging as a viable alternative to conventional forms of propulsion. Still at their infancy and relatively untested, many sources of uncertainty remain that are unique to solar sails and which will continue to affect their design as solar sails increase in performance and size. Controlling their attitude in the context of these uncertainties therefore becomes critical to spaceflight missions that will explore our solar system and beyond from new, previously-unattainable perspectives. Two distinct frameworks are developed to manage these uncertainties to control the attitude of solar sails. The first utilizes a provided system model and utilizes an observation of the control history to operate the sail about an equilibrium position that is passively stable. The approach utilizes past information about controller input for the purposes of rejecting disturbances that arise from several sources of uncertainty. The second approach is forward-looking and is inspired by trajectory-based reachability analysis. This approach was developed in the context of six degree-of-freedom supervised control of an unmanned aerial vehicle whose faster dynamics in a disturbance-rich environment provide a computational challenge and thus require machine-learned approximations to a reachable set of safe inputs. These methodologies are then applied to the original solar sail attitude control problem. Predictions are made about the future state of the sail after performing a minimum-time large angle maneuver. Uncertainty distributions are assumed a priori and are then used to create a buffer angle for the maneuver such that no overshoot occurs within a tunable statistical measure of safety. Uncertainties handled in this way include the sail effective reflectivity, flexural rigidity, and moment of inertia. However, the framework is designed to be very adaptable and so is able to accommodate arbitrary sources of uncertainties and flexible modeling techniques. Utilization of machine learning allows for arbitrary complexity in the simulation and modeling framework without impacting the on-board computational requirements of the solar sail hardware.