Classical control methods, such as linear quadratic regulators, often require parameter tuning. Typically, these parameters remain constant throughout the entire window of interest. In this project, our goal is to learn the free parameters through reinforcement learning. The objective is to enhance the optimality of classical controllers by making the weights state-dependent and representing this through a neural network.
We will use simulated spacecraft dynamics, for instance, in proximity operations, as test cases to assess the improved performance of the derived reinforcement learning-based controllers.
Undergraduate
None
Dynamics & Control Lab (405.852, Lab)