A feasible sequential quadratic programming strategy with iterated second-order corrections

This package provides a solver for parametric nonconvex programs of the form

installation

• clone the repo:

git clone https://github.com/zanellia/feasible_sqp.git

• init submodules (qpOASES, CasADi):

git submodule update --init

• install the Python package:

cd feasible_sqp
pip install -e .

• build and install the dependencies:

import feasible_sqp
feasible_sqp.install_dependencies(matlab_root_path=<...>)

usage

from feasible_sqp import *

matlab_root_path =     <...> # e.g., '/usr/local/MATLAB/R2017b'

install_dependencies(matlab_root_path=matlab_root_path)

# number of primal variables
nv = 2

# create solver
solver = feasible_sqp(nv)

# get primal variables
y = solver.y

# define cost
f = 1.0/2.0*ca.dot(y-10,y-10)

# define constraints
g = ca.vertcat(ca.sin(y[1]) - y[0])

# define bounds
lby = -np.ones((nv,1))
uby = np.ones((nv,1))

# generate solver
solver.generate_solver(f, f, g, lby = lby, uby = uby)

# solve NLP
solver.solve()

notice: a call to solver.generate_solver() generates a script called set_LD_LIBRARY_PATH.sh which can be sourced in order to update the LD_LIBRARY_PATH environment variable such that the shared libraries needed by the solver can be located.