Abstract:

A least absolute shrinkage and selection operator (LASSO) based node adaptive refinement approach for the direct method in trajectory optimization is proposed. Firstly, a higher resolution grid and its associated radial basis function set are created. Sequentially, the control variables are approximated using the resulting radial basis function set, and its sampling sequence is generated by interpolation. Finally, the coefficients of the formulated approximation function are estimated based on the statistical variable selection method-LASSO. The higher multi-resolution nodes associated with radial basis functions with non-zero coefficient are selected as new nodes. The proposed method refines the mesh without estimation of states and/or errors controls, and few extra parameters are involved. Therefore, the formulated trajectory optimization algorithm behaves strong adaptability and generality. The validity of this method is demonstrated by several typical examples.