(Georgia Institute of Technology, 2021-07-30)
Mote, Mark Leo
This thesis investigates the problem of safety-critical control for complex cyber-physical systems, with an emphasis on numerical optimization and autonomy applications in the space domain. First, a set-based approach is introduced for specifying mission constraints, and safety is formalized in the context of a set invariance framework. Next, the research investigates the problem of run time assurance (RTA), which relates to a control system architecture where a performance-oriented controller is augmented with a safety-driven element that filters the control signal in such a way that guarantees safety. The latter part of the thesis consists of application-specific research on various space systems.
Autonomous rendezvous proximity operations and docking (ARPOD) is considered under proximity, collision-avoidance, and speed constraints. Natural motion trajectories are used to identify a set of passively safe parking orbits under the Clohessy-Wiltshire-Hill dynamics, and a mixed integer programming approach is used to generate safety-constrained optimal transfer trajectories to this set. The formulation is encoded into an RTA framework.
The safety problem is considered for a torque-controlled spacecraft in free rotational motion, subject to line-of-sight constraints. A nondeterministic dynamics model is considered, and an RTA filter is constructed that relies on online computation of forward reachable sets around a recovery maneuver. The approach utilizes recent results from reachability theory in addition to optimization-based computation of invariant sets. Safety guarantees exist when a disturbance torque is bounded. The practicality of the approach is demonstrated with an application on a hardware testbed.
Finally, the research studies the topic of harnessing collisional behavior for free-flying spacecraft. A framework is proposed for collision-inclusive trajectory optimization. Experimental comparisons of trajectories with and without collision-avoidance requirements demonstrate the capability of the collision-inclusive strategy to achieve significant performance improvements in realistic scenarios. Additionally, a safety application is considered, and the planner is utilized for the purpose of optimally mitigating damage in the presence of an inevitable collision.