Experimental Results for Adaptive, Optimal Control of a 2-DOF Helicopter

摘要

This paper discusses the practical implementation of an adaptive, optimal controller on an experimental platform of a 2-DOF helicopter. The controller is based on the theory of reinforcement learning; a policy iteration algorithm is used to solve the Linear-Quadratic-Regulator (LQR) problem. More specifically it is used to solve the algebraic Riccati equation online with partial knowledge of the system dynamics, hence the controller gains are changed over time based on the system states and converge to the optimal gains. A simulation of the system is used to test the algorithm and show that the solution obtained online converges to the LQR solution. Afterwards the requirements to implement such a control scheme on an actual system and test its performance are discussed. Finally, experimental results of the controller performance are examined and analyzed.