MPC and MHE for a laboratory scale overhead crane

This was the first application of an auto-generated Nonlinear Model Predictive Control (MPC) from the ACADO toolkit. The experiments are explained in detail in this publication. Although the tracking was slow, we showed that we can run the so-called real-time iteration (RTI) scheme in hard real-time. Furthermore, experiments proved a fast convergence rate of the RTI scheme. This implementation used a simple MPC formulation and a simple estimator.

The results of the improved experiments are published in this paper. We used a more detailed model and a more complex formulation for the MPC. Moreover, this time we used nonlinear Moving Horizon Estimation (MHE) to more accurately estimate the system' states. As a result, faster motions are achieved. Below is the video of the improved experiments. Besides faster motions, we also achieved faster execution times of the controller and the estimator: we used a newer version of the code generator.

A detailed comparison of the two aforementioned approaches can be found in Chapter 5 of my PhD thesis.