carrot/third_party/acados/acados_template/gnsf/check_reformulation.py
Vehicle Researcher b2f2dabe71 openpilot v0.9.5 release
date: 2023-11-17T23:53:40
master commit: d3aad9ca4601ae0a448ed971c1cd151c7c1eb690
2023-11-17 23:53:40 +00:00

217 lines
6.5 KiB
Python

#
# Copyright (c) The acados authors.
#
# This file is part of acados.
#
# The 2-Clause BSD License
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are met:
#
# 1. Redistributions of source code must retain the above copyright notice,
# this list of conditions and the following disclaimer.
#
# 2. Redistributions in binary form must reproduce the above copyright notice,
# this list of conditions and the following disclaimer in the documentation
# and/or other materials provided with the distribution.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
# ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
# LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
# CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
# SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
# INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
# CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
# ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
# POSSIBILITY OF SUCH DAMAGE.;
#
from acados_template.utils import casadi_length
from casadi import *
import numpy as np
def check_reformulation(model, gnsf, print_info):
## Description:
# this function takes the implicit ODE/ index-1 DAE and a gnsf structure
# to evaluate both models at num_eval random points x0, x0dot, z0, u0
# if for all points the relative error is <= TOL, the function will return::
# 1, otherwise it will give an error.
TOL = 1e-14
num_eval = 10
# get dimensions
nx = gnsf["nx"]
nu = gnsf["nu"]
nz = gnsf["nz"]
nx1 = gnsf["nx1"]
nx2 = gnsf["nx2"]
nz1 = gnsf["nz1"]
nz2 = gnsf["nz2"]
n_out = gnsf["n_out"]
# get model matrices
A = gnsf["A"]
B = gnsf["B"]
C = gnsf["C"]
E = gnsf["E"]
c = gnsf["c"]
L_x = gnsf["L_x"]
L_xdot = gnsf["L_xdot"]
L_z = gnsf["L_z"]
L_u = gnsf["L_u"]
A_LO = gnsf["A_LO"]
E_LO = gnsf["E_LO"]
B_LO = gnsf["B_LO"]
c_LO = gnsf["c_LO"]
I_x1 = range(nx1)
I_x2 = range(nx1, nx)
I_z1 = range(nz1)
I_z2 = range(nz1, nz)
idx_perm_f = gnsf["idx_perm_f"]
# get casadi variables
x = gnsf["x"]
xdot = gnsf["xdot"]
z = gnsf["z"]
u = gnsf["u"]
y = gnsf["y"]
uhat = gnsf["uhat"]
p = gnsf["p"]
# create functions
impl_dae_fun = Function("impl_dae_fun", [x, xdot, u, z, p], [model.f_impl_expr])
phi_fun = Function("phi_fun", [y, uhat, p], [gnsf["phi_expr"]])
f_lo_fun = Function(
"f_lo_fun", [x[range(nx1)], xdot[range(nx1)], z, u, p], [gnsf["f_lo_expr"]]
)
# print(gnsf)
# print(gnsf["n_out"])
for i_check in range(num_eval):
# generate random values
x0 = np.random.rand(nx, 1)
x0dot = np.random.rand(nx, 1)
z0 = np.random.rand(nz, 1)
u0 = np.random.rand(nu, 1)
if gnsf["ny"] > 0:
y0 = L_x @ x0[I_x1] + L_xdot @ x0dot[I_x1] + L_z @ z0[I_z1]
else:
y0 = []
if gnsf["nuhat"] > 0:
uhat0 = L_u @ u0
else:
uhat0 = []
# eval functions
p0 = np.random.rand(gnsf["np"], 1)
f_impl_val = impl_dae_fun(x0, x0dot, u0, z0, p0).full()
phi_val = phi_fun(y0, uhat0, p0)
f_lo_val = f_lo_fun(x0[I_x1], x0dot[I_x1], z0[I_z1], u0, p0)
f_impl_val = f_impl_val[idx_perm_f]
# eval gnsf
if n_out > 0:
C_phi = C @ phi_val
else:
C_phi = np.zeros((nx1 + nz1, 1))
try:
gnsf_val1 = (
A @ x0[I_x1] + B @ u0 + C_phi + c - E @ vertcat(x0dot[I_x1], z0[I_z1])
)
# gnsf_1 = (A @ x[I_x1] + B @ u + C_phi + c - E @ vertcat(xdot[I_x1], z[I_z1]))
except:
import pdb
pdb.set_trace()
if nx2 > 0: # eval LOS:
gnsf_val2 = (
A_LO @ x0[I_x2]
+ B_LO @ u0
+ c_LO
+ f_lo_val
- E_LO @ vertcat(x0dot[I_x2], z0[I_z2])
)
gnsf_val = vertcat(gnsf_val1, gnsf_val2).full()
else:
gnsf_val = gnsf_val1.full()
# compute error and check
rel_error = np.linalg.norm(f_impl_val - gnsf_val) / np.linalg.norm(f_impl_val)
if rel_error > TOL:
print("transcription failed rel_error > TOL")
print("you are in debug mode now: import pdb; pdb.set_trace()")
abs_error = gnsf_val - f_impl_val
# T = table(f_impl_val, gnsf_val, abs_error)
# print(T)
print("abs_error:", abs_error)
# error('transcription failed rel_error > TOL')
# check = 0
import pdb
pdb.set_trace()
if print_info:
print(" ")
print("model reformulation checked: relative error <= TOL = ", str(TOL))
print(" ")
check = 1
## helpful for debugging:
# # use in calling function and compare
# # compare f_impl(i) with gnsf_val1(i)
#
# nx = gnsf['nx']
# nu = gnsf['nu']
# nz = gnsf['nz']
# nx1 = gnsf['nx1']
# nx2 = gnsf['nx2']
#
# A = gnsf['A']
# B = gnsf['B']
# C = gnsf['C']
# E = gnsf['E']
# c = gnsf['c']
#
# L_x = gnsf['L_x']
# L_z = gnsf['L_z']
# L_xdot = gnsf['L_xdot']
# L_u = gnsf['L_u']
#
# A_LO = gnsf['A_LO']
#
# x0 = rand(nx, 1)
# x0dot = rand(nx, 1)
# z0 = rand(nz, 1)
# u0 = rand(nu, 1)
# I_x1 = range(nx1)
# I_x2 = nx1+range(nx)
#
# y0 = L_x @ x0[I_x1] + L_xdot @ x0dot[I_x1] + L_z @ z0
# uhat0 = L_u @ u0
#
# gnsf_val1 = (A @ x[I_x1] + B @ u + # C @ phi_current + c) - E @ [xdot[I_x1] z]
# gnsf_val1 = gnsf_val1.simplify()
#
# # gnsf_val2 = A_LO @ x[I_x2] + gnsf['f_lo_fun'](x[I_x1], xdot[I_x1], z, u) - xdot[I_x2]
# gnsf_val2 = A_LO @ x[I_x2] + gnsf['f_lo_fun'](x[I_x1], xdot[I_x1], z, u) - xdot[I_x2]
#
#
# gnsf_val = [gnsf_val1 gnsf_val2]
# gnsf_val = gnsf_val.simplify()
# dyn_expr_f = dyn_expr_f.simplify()
# import pdb; pdb.set_trace()
return check