Hi 
I want to use a lot of interpolants in my model. However, using acados, compiling the MPC takes forever for a lot of interpolants. Using dompc for setting up the MPC only takes a fraction of the time.
Below you find an example code where you can see the difference. In my actual example I want to use 215 interpolants and 120 repititions. Repitition in my example means how often I want to evaluate the same interpolant at a different point.
Any idea how I could speed up the compilation or reduce the generated code? My first idea would be that, for the repititions, similar code could be reused during code generation. I don’t know if this is done.
import uuid
from time import perf_counter
import casadi as cd
import numpy as np
from acados_template import AcadosModel, AcadosOcp, AcadosOcpSolver
from do_mpc.controller import MPC
from do_mpc.model import Model
rng = np.random.default_rng(seed=42)
def create_interpolant() -> cd.Function:
xgrid = np.linspace(-5, 5, 12)
ygrid = np.linspace(-4, 4, 10)
X, Y = np.meshgrid(xgrid, ygrid, indexing='ij')
R = np.sqrt(5 * X**2 + Y**2) + rng.random(1)
data = np.sin(R) / R
data_flat = data.ravel(order='F')
name = 'interpolant_' + str(uuid.uuid4()).replace('-', '_')
lut = cd.interpolant(name, 'bspline', [xgrid, ygrid], data_flat)
return lut
def export_simple_acados_model(n_interpolants: int = 2,
n_repititions: int = 1) -> AcadosModel:
model_name = 'minimal_example' + str(uuid.uuid4()).replace('-', '_')
# set up states & controls
x = cd.MX.sym('x')
u = cd.MX.sym('u')
# xdot
x_dot = cd.MX.sym('x_dot')
# dynamics
f_expl = 0
for _ in range(n_interpolants):
interpolant = create_interpolant()
for index in range(n_repititions):
f_expl = f_expl + interpolant(cd.vertcat(u, index))
f_impl = x_dot - f_expl
model = AcadosModel()
model.f_impl_expr = f_impl
model.f_expl_expr = f_expl
model.x = x
model.xdot = x_dot
model.u = u
# set model_name
model.name = model_name
return model
def export_simple_dompc_model(n_interpolants: int = 2,
n_repititions: int = 1) -> AcadosModel:
# set up states & controls
model = Model('continuous', 'MX')
u = model.set_variable('_u', 'u')
x = model.set_variable('_x', 'x')
# dynamics
f_expl = 0
for _ in range(n_interpolants):
interpolant = create_interpolant()
for index in range(n_repititions):
f_expl = f_expl + interpolant(cd.vertcat(u, index / n_repititions))
model.set_rhs('x', f_expl)
model.setup()
return model
def test_with_interpolants(n_interpolants: int = 2,
n_repititions: int = 1) -> None:
# create ocp object to formulate the OCP
dompc_time = create_dompc_mpc(n_interpolants, n_repititions)
acados_time = create_acados_mpc(n_interpolants, n_repititions)
print(f'{dompc_time=}')
print(f'{acados_time=}')
def create_dompc_mpc(n_interpolants: int = 2, n_repititions: int = 1) -> float:
model = export_simple_dompc_model(n_interpolants, n_repititions)
start = perf_counter()
mpc = MPC(model)
lterm = model.x['x']**2 + model.u['u']**2
mterm = model.x['x']**2
mpc.set_objective(mterm, lterm)
dompc_options = {
'n_horizon': 2,
't_step': 1,
'n_robust': 0,
}
mpc.set_param(**dompc_options)
mpc.setup()
end = perf_counter()
elapsed_time = end - start
return elapsed_time
def create_acados_mpc(n_interpolants: int = 2, n_repititions: int = 1) -> float:
ocp = AcadosOcp()
model = export_simple_acados_model(n_interpolants, n_repititions)
ocp.model = model
ocp.dims.N = 20
ocp.cost.cost_type = 'EXTERNAL'
ocp.cost.cost_type_e = 'EXTERNAL'
ocp.model.cost_expr_ext_cost = model.x**2 + model.u**2
ocp.model.cost_expr_ext_cost_e = model.x**2
ocp.solver_options.tf = 1.0
start = perf_counter()
ocp_solver = AcadosOcpSolver(ocp, json_file='acados_ocp.json')
end = perf_counter()
elapsed_time = end - start
return elapsed_time
if __name__ == '__main__':
test_with_interpolants(n_interpolants=200, n_repititions=5)
Thanks,
David
