A dual Newton strategy with fixed iteration complexity for multi-stage MPC

Rapport, 2017

This paper considers the problem of solving Quadratic Programs (QPs) in the context of multi-stage Model Predictive Control (MPC). A Newton strategy is considered on the dual of the problem to achieve a parallelizable method. In this context, it has been observed that the globalization strategy can be expensive. In this paper, we propose to dualize both the non-anticipativity constraints and the dynamics in order to obtain a computationally cheap globalization. The dual Newton system is then reformulated to small highly structured linear systems that to a large extent can be solved in parallel.