Improvement of an LUT-based intelligent motion controller by underestimation of reachable sets

We develop a preprocessing method to improve the performance of an intelligent motion controller. Our method addresses the competitive problems that have been observed in the coupled inverted pendula model, in which a controller outputs impulsive forces to produce the desired final states based on a look-up table (LUT) that stores dynamical correspondences from the initial to the final states. However, the degradation in performance due to misclassifications in the LUT occurs near the boundary points of the reachable sets of the model. The proposed algorithm removes the boundary points from the LUT. The resulting controller successfully reduces misclassification errors and improves overall control performance.