A Highly Parallelizable Newton-type Method for Nonlinear Model Predictive Control

We propose a highly parallelizable Newton-type method for nonlinear model predictive control by exploiting the particular structure of the associated Karush-Kuhn-Tucker conditions. These equations are approximately decoupled into single step subproblems along the prediction horizon for parallelization. The coupling variable of each subproblem is approximated toward its optimal value by a simple but effective method in every iteration. The proposed algorithm is applied to control a quadrotor. The numerical simulation results show that the proposed algorithm is highly parallelizable and converges with only a few iterations even to a high accuracy. The proposed method is also shown to be faster compared with several state-of-the-art algorithms.