A fast Nonlinear Model Predictive Control strategy for real-time motion control of mechanical systems

Abstract

Widespread application of real-time, Nonlinear Model Predictive Control (NMPC) algorithms to systems of large scale or with fast dynamics is challenged by the high associated computational cost, in particular in presence of long prediction horizons. In this paper, a fast NMPC strategy to reduce the on-line computational cost is proposed. A Curvature-based Measure of Nonlinearity (CMoN) of the system is exploited to reduce the required number of sensitivity computations, which largely contribute to the overall computational cost. The proposed scheme is validated by a simulation study on the chain of masses motion control problem, a toy example that can be easily extended to an arbitrary dimension. Simulations have been run with long prediction horizons and large state dimensions. Results show that sensitivity computations are significantly reduced with respect to other sensitivity updating schemes, while preserving control performance.