Hierachical Quadratic Problem solver
This software implements a MATLAB-based hierarchical least-square quadratic solver (HQP) based on the paper [1]
The software is organized as follow:
- in the main directory, the main functions of the solver:
-
hcod computes the HCOD from the problem specification A,b and an initial guess of the active set. The function returns h and Y that are manipulated by the other function during the active-search loop.
-
eHQP_primal computes the primal optimum of the eHQP corresponding to the current active set. The function returns the optimum x and the corresponding expression in the Y basis y. This corresponds to Alg. 1 of the paper
-
step_length corresponds to Alg. 5 of the paper. The function returns the step length, along with a Boolean specifying if one constraint needs to be activated and the constraint reference.
-
eHQP_dual compute the multiplier of one level of the hierarchy, as proposed in Alg. 2 of the paper.
-
check_mult checks if all the multipliers are positive for lexicographic order. The function returns the Boolean ``one multiplier is negative'' along with the lowest multiplier reference.
-
active_search is the main function: it computes the HCOD and then run the search loop. It returns the primal and dual optima and the HCOD.
-
in utils, the secondary functions used by the solver.
-
in unittests, some unitary tests that validates the main functionnalities of the solver.
-
in unittests/utils, a set of secondary functions used by the tests, with no real arrangement nor documentation.
All the code should be run from the main directory. Add unittests in your path and launch the unitary tests:
addpath('unittests')
test_search
Nicolas Mansard, LAAS/CNRS, Toulouse, France
[1] "Hiearchical Quadratic Programming", by Adrien Escande (JRL-Japan/CNRS, Tsukuba, Japan), Nicolas Mansard (LAAS/CNRS, Toulouse, France) and Pierre-Brice Wieber (INRIA, Grenoble, France), on-going submission to IJRR.