Robust Dual Control MPC with Application to Soft-Landing Control

Dual control frameworks for systems subject to uncertainties aim at simultaneously learning the unknown parameters while controlling the system dynamics. We propose a robust dual model predictive control algorithm for systems with bounded uncertainty with application to soft landing control. The algorithm exploits a robust control invariant set to guarantee constraint enforcement in spite of the uncertainty, and a constrained estimation algorithm to guarantee admissible parameter estimates. The impact of the control input on parameter learning is accounted for by including in the cost function a reference input, which is designed online to provide persistent excitation. The reference input design problem is non-convex, and here is solved by a sequence of relaxed convex problems. The results of the proposed method in a soft-landing control application in transportation systems are shown.