Block Structured Preconditioning within an Active-Set Method for Real-Time Optimal Control

Model predictive control (MPC) requires solving a block-structured optimal control problem at each sampling instant. We propose an iterative preconditioned solver with computational cost that scales linearly with the number of intervals and quadratically with the number of state and control variables, and can be efficiently implemented on embedded hardware for real-time optimal control. Block-structured factorizations and low-rank updates are combined with blockdiagonal preconditioning within a primal active-set strategy (PRESAS). Multiple numerical tests using our preliminary C implementation demonstrate competitiveness with the state-of-the-art, as illustrated on an ARM Cortex-A53 processor.