Parallel implementations of the fast gradient method for high-speed MPC

Abstract

The potential of the fast gradient method for solving linear quadratic model predictive control (MPC) problems in the sub-millisecond range was only recently recognised by Richter et al. This paper aims to push the execution-time limit even further by exploiting the computational power offered by parallel computing architectures. In particular, scalable and adaptive implementations of gradient-based optimisation methods are presented for both multi-core CPUs and field programmable gate arrays. The proposed parallel implementations broaden the applicability of MPC to problems that were considered out-of-reach till recent years.