ParNMPC – a parallel optimisation toolkit for real-time nonlinear model predictive control

Abstract

Real-time optimisation for nonlinear model predictive control (NMPC) has always been challenging, especially for fast-sampling and large-scale applications. This paper presents an efficient implementation of a highly parallelisable method for NMPC, called ParNMPC. The implementation details of ParNMPC are introduced, including a dedicated discretisation method suitable for parallelisation, a framework that unifies search direction calculation done using Newton's method and the parallel method, line search methods for guaranteeing convergence, and a warm start strategy for the interior-point method. To assess the performance of ParNMPC under different configurations, three experiments including a closed-loop simulation of a quadrotor, a real-world control example of a laboratory helicopter and a closed-loop simulation of a robot manipulator are shown. These experiments show the effectiveness and efficiency of ParNMPC both in serial and parallel.

Keywords:

• Nonlinear model predictive control
• parallel computing
• real-time optimisation

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by JSPS KAKENHI [15H02257].