Simulation step fails with shape error when parameter is an N-dimensional array

mcgarnett commented 3 weeks ago

Description

Sometimes a parameter input to a Dymos ODE needs to be something other than a scalar or 1-D array. One such situation might be when the ODE describes 6 degree of freedom motion, where the Moment of Inertia matrix of the body whose motion you want to simulate is a parameter for the phase and trajectory.

It turns out that the parameters with shapes greater than 1-D work when finding the solution -- that is calling dymos.run_problem(p, run_driver=True, simulate=False)

However, when setting the "simulate" flag to True, the simulation fails with this error

   File dymos\utils\introspection.py:471 in configure_parameters_introspection
    raise RuntimeError(f'Shape provided to parameter `{name}` differs from its targets.\n'

A minimal example is provided that reproduces the behavior

Example

import numpy as np
import openmdao.api as om
import dymos as dm

class ODE(om.Group):

    def initialize(self):
        self.options.declare("num_nodes", types=int)

    def setup(self):

        nn = self.options["num_nodes"]
        lin_sys = om.LinearSystemComp(
            size=3,
            vec_size=nn,
            vectorize_A=True,
        )

        self.add_subsystem(
            "rates",
            lin_sys,
            promotes_inputs=["*"],
            promotes_outputs=["*"],
        )

p = om.Problem()
p.driver = om.pyOptSparseDriver()
p.driver.options["optimizer"] = "SNOPT"
p.driver.opt_settings["Major iterations limit"] = 100
p.driver.options["print_results"] = False

t = dm.Radau(num_segments=20, order=3)
phase = dm.Phase(ode_class=ODE, transcription=t)

phase.set_time_options(fix_initial=True, duration_bounds=(5, 60), duration_ref=1)
phase.add_state("w", fix_initial=True, rate_source="x")
phase.add_control("M", opt=False, targets=["b"], lower=-50, upper=50)
phase.set_control_options("M", val=np.array([30, -50, 50]))

A_mat = np.array(
    [
        [10, 0, 0],
        [0, 10, 0],
        [0, 0, 10],
    ]
)

# argument "dynamic" doesn't seem to help
phase.add_parameter("A", val=A_mat, targets=["A"])
phase.add_objective("time", loc="final", scaler=1)

traj = dm.Trajectory()
traj.add_phase("phase0", phase)

p.model.add_subsystem("traj", traj)

p.driver.declare_coloring()
p.setup(force_alloc_complex=True)

# This works
# dm.run_problem(p, run_driver=True, simulate=False, make_plots=True)
# This doesn't
dm.run_problem(p, run_driver=True, simulate=True, make_plots=True)

Dymos Version

1.10.0

Relevant environment information

OpenMDAO version 3.32.0 Dymos version 1.10.0

mcgarnett commented 3 weeks ago

Will delete the issue if it's already related to the following issue https://github.com/OpenMDAO/dymos/issues/1076

kaushikponnapalli commented 2 weeks ago

@mcgarnett The recently merged PR #1127 should resolve your issue. Included in it is a simple example showing matrix parameters. The example you provided is somewhat more complicated from a sizing/connections perspective but it should work if you make the two following changes

Set vectorize_A=False when declaring the linear system comp
Set static_target=True when adding A as a parameter

OpenMDAO / dymos