ESA GNC Conference Papers Repository

A novel jump-linear quadratic controller is suggested for discrete-time dynamical systems under the assumptions of full state information and corrupted detection of the hidden mode. The controller is linear in the state and the gains are functions of the mode observations. They stem from a Riccati-like backward propagation involving approximate computations of conditional expectations of the problem parameters. Discarding the state information, closed-form recursive expressions for the mode estimator are developed with increasing and limited memory of the mode observations. For the sake of practicality, the mode estimator with memory limited to the current mode observation is proposed. This results in a controller that maintains a computation burden similar to the full information jump-linear quadratic controller. Comparative results of an extensive Monte-Carlo simulation for a simple system illustrates the ecacy of the proposed algorithm that mitigates the destabilizing eect of corrupted mode observations. The proposed algorithm lends itself to a fault-tolerant spacecraft attitude controller. Extensive Monte-Carlo simulation were performed for a three-mode scenario where magnetic actuation and sensing could fail. The example illustrates the validity of the approach, showing, during failures as well as for the healthy mode, a steady state pointing accuracy of a few degrees. The proposed approach shows exibility from the modeling standpoint and proves to be promising for the development of ecient fault-tolerant attitude controllers.