Towards a modular software package for embedded optimization

Abstract

In this paper we present acados, a new software package for model predictive control. It provides a collection of embedded optimization algorithms written in C, with a strong focus on computational efficiency. Its modular structure makes it useful for rapid prototyping, i.e. designing a control algorithm by putting together different algorithmic components that are readily connected and interchanged. The usefulness of the software is demonstrated with a closed-loop simulation experiment of an inverted pendulum, which shows acados attaining sub-millisecond computation times per iteration. Furthermore, we showcase a new algorithmic idea in the context of embedded nonlinear model predictive control (NMPC), namely sequential convex quadratic programming (SCQP), along with an efficient implementation of it.