Fast hybrid dual mode NMPC for a parallel double inverted pendulum with experimental validation

Abstract

This study presents a novel fast non-linear model predictive control approach for a parallel double inverted pendulum. The approach uses dual-mode closed-loop predictions to obtain numerically robust optimal solutions. Moreover, it uses the real-time iteration (RTI) scheme to reduce the computational burden and achieve real-time performance. Furthermore, two main modifications are proposed which significantly improve the performance of the RTI scheme in the presence of large disturbances, namely; additional energy-based costs, and a hybrid switching scheme. Finally, the approach uses a non-standard discretised model combined with an online system identification scheme to address parameter uncertainty, and with an extended Kalman filter for state-estimation. The resulting performance is validated through both simulations and experimental results.