Dual MPC for Active Learning of Nonparametric Uncertainties

Abstract: This manuscript presents a dual model predictive controller (DMPC) that balances the two objectives of dual control, namely, system identification and control. In particular, we propose a Gaussian process (GP)-based MPC that uses the posterior GP covariance for active learning. The DMPC can steer the system towards states that have high covariance, or to the setpoint, thereby balancing system identification and control performance (exploration vs. exploitation). We establish robust constraint satisfaction of the novel DMPC through the use of a contingency plan. We demonstrate the DMPC in a numerical study of a nonlinear system with nonparametric uncertainties.