示例#1
0
    def test_racecar_for_docs(self):
        import numpy as np
        import openmdao.api as om
        from openmdao.utils.assert_utils import assert_near_equal
        import dymos as dm
        import matplotlib.pyplot as plt
        import matplotlib as mpl

        from dymos.examples.racecar.combinedODE import CombinedODE
        from dymos.examples.racecar.spline import (get_spline, get_track_points, get_gate_normals,
                                                   reverse_transform_gates, set_gate_displacements,
                                                   transform_gates, )
        from dymos.examples.racecar.linewidthhelper import linewidth_from_data_units
        from dymos.examples.racecar.tracks import ovaltrack  # track curvature imports

        # change track here and in curvature.py. Tracks are defined in tracks.py
        track = ovaltrack

        # generate nodes along the centerline for curvature calculation (different
        # than collocation nodes)
        points = get_track_points(track)

        # fit the centerline spline.
        finespline, gates, gatesd, curv, slope = get_spline(points, s=0.0)
        # by default 10000 points
        s_final = track.get_total_length()

        # Define the OpenMDAO problem
        p = om.Problem(model=om.Group())

        # Define a Trajectory object
        traj = dm.Trajectory()
        p.model.add_subsystem('traj', subsys=traj)

        # Define a Dymos Phase object with GaussLobatto Transcription
        phase = dm.Phase(ode_class=CombinedODE,
                         transcription=dm.GaussLobatto(num_segments=50, order=3, compressed=True))

        traj.add_phase(name='phase0', phase=phase)

        # Set the time options, in this problem we perform a change of variables. So 'time' is
        # actually 's' (distance along the track centerline)
        # This is done to fix the collocation nodes in space, which saves us the calculation of
        # the rate of change of curvature.
        # The state equations are written with respect to time, the variable change occurs in
        # timeODE.py
        phase.set_time_options(fix_initial=True, fix_duration=True, duration_val=s_final,
                               targets=['curv.s'], units='m', duration_ref=s_final,
                               duration_ref0=10)

        # Define states
        phase.add_state('t', fix_initial=True, fix_final=False, units='s', lower=0,
                        rate_source='dt_ds', ref=100)  # time
        phase.add_state('n', fix_initial=False, fix_final=False, units='m', upper=4.0, lower=-4.0,
                        rate_source='dn_ds', targets=['n'],
                        ref=4.0)  # normal distance to centerline. The bounds on n define the
        # width of the track
        phase.add_state('V', fix_initial=False, fix_final=False, units='m/s', ref=40, ref0=5,
                        rate_source='dV_ds', targets=['V'])  # velocity
        phase.add_state('alpha', fix_initial=False, fix_final=False, units='rad',
                        rate_source='dalpha_ds', targets=['alpha'],
                        ref=0.15)  # vehicle heading angle with respect to centerline
        phase.add_state('lambda', fix_initial=False, fix_final=False, units='rad',
                        rate_source='dlambda_ds', targets=['lambda'],
                        ref=0.01)  # vehicle slip angle, or angle between the axis of the vehicle
        # and velocity vector (all cars drift a little)
        phase.add_state('omega', fix_initial=False, fix_final=False, units='rad/s',
                        rate_source='domega_ds', targets=['omega'], ref=0.3)  # yaw rate
        phase.add_state('ax', fix_initial=False, fix_final=False, units='m/s**2',
                        rate_source='dax_ds', targets=['ax'], ref=8)  # longitudinal acceleration
        phase.add_state('ay', fix_initial=False, fix_final=False, units='m/s**2',
                        rate_source='day_ds', targets=['ay'], ref=8)  # lateral acceleration

        # Define Controls
        phase.add_control(name='delta', units='rad', lower=None, upper=None, fix_initial=False,
                          fix_final=False, targets=['delta'], ref=0.04)  # steering angle
        phase.add_control(name='thrust', units=None, fix_initial=False, fix_final=False, targets=[
            'thrust'])  # the thrust controls the longitudinal force of the rear tires and is
        # positive while accelerating, negative while braking

        # Performance Constraints
        pmax = 960000  # W
        phase.add_path_constraint('power', upper=pmax, ref=100000)  # engine power limit

        # The following four constraints are the tire friction limits, with 'rr' designating the
        # rear right wheel etc. This limit is computed in tireConstraintODE.py
        phase.add_path_constraint('c_rr', upper=1)
        phase.add_path_constraint('c_rl', upper=1)
        phase.add_path_constraint('c_fr', upper=1)
        phase.add_path_constraint('c_fl', upper=1)

        # Some of the vehicle design parameters are available to set here. Other parameters can
        # be found in their respective ODE files.
        phase.add_parameter('M', val=800.0, units='kg', opt=False,
                            targets=['car.M', 'tire.M', 'tireconstraint.M', 'normal.M'],
                            static_target=True)  # vehicle mass
        phase.add_parameter('beta', val=0.62, units=None, opt=False, targets=['tire.beta'],
                            static_target=True)  # brake bias
        phase.add_parameter('CoP', val=1.6, units='m', opt=False, targets=['normal.CoP'],
                            static_target=True)  # center of pressure location
        phase.add_parameter('h', val=0.3, units='m', opt=False, targets=['normal.h'],
                            static_target=True)  # center of gravity height
        phase.add_parameter('chi', val=0.5, units=None, opt=False, targets=['normal.chi'],
                            static_target=True)  # roll stiffness
        phase.add_parameter('ClA', val=4.0, units='m**2', opt=False, targets=['normal.ClA'],
                            static_target=True)  # downforce coefficient*area
        phase.add_parameter('CdA', val=2.0, units='m**2', opt=False, targets=['car.CdA'],
                            static_target=True)  # drag coefficient*area

        # Minimize final time.
        # note that we use the 'state' time instead of Dymos 'time'
        phase.add_objective('t', loc='final')

        # Add output timeseries
        phase.add_timeseries_output('*')

        # Link the states at the start and end of the phase in order to ensure a continous lap
        traj.link_phases(phases=['phase0', 'phase0'],
                         vars=['V', 'n', 'alpha', 'omega', 'lambda', 'ax', 'ay'],
                         locs=('final', 'initial'))

        # Set the driver. IPOPT or SNOPT are recommended but SLSQP might work.
        p.driver = om.pyOptSparseDriver(optimizer='IPOPT')

        p.driver.opt_settings['mu_init'] = 1e-3
        p.driver.opt_settings['max_iter'] = 500
        p.driver.opt_settings['acceptable_tol'] = 1e-3
        p.driver.opt_settings['constr_viol_tol'] = 1e-3
        p.driver.opt_settings['compl_inf_tol'] = 1e-3
        p.driver.opt_settings['acceptable_iter'] = 0
        p.driver.opt_settings['tol'] = 1e-3
        p.driver.opt_settings['nlp_scaling_method'] = 'none'
        p.driver.opt_settings['print_level'] = 5
        p.driver.opt_settings['nlp_scaling_method'] = 'gradient-based'  # for faster convergence
        p.driver.options['print_results'] = False

        # Allow OpenMDAO to automatically determine our sparsity pattern.
        # Doing so can significant speed up the execution of Dymos.
        p.driver.declare_coloring()

        # Setup the problem
        p.setup(check=True)  # force_alloc_complex=True
        # Now that the OpenMDAO problem is setup, we can set the values of the states.

        # States
        # Nonzero velocity to avoid division by zero errors
        p.set_val('traj.phase0.states:V', phase.interp('V', [20, 20]), units='m/s')
        # All other states start at 0
        p.set_val('traj.phase0.states:lambda', phase.interp('lambda', [0.0, 0.0]), units='rad')
        p.set_val('traj.phase0.states:omega', phase.interp('omega', [0.0, 0.0]), units='rad/s')
        p.set_val('traj.phase0.states:alpha', phase.interp('alpha', [0.0, 0.0]), units='rad')
        p.set_val('traj.phase0.states:ax', phase.interp('ax', [0.0, 0.0]), units='m/s**2')
        p.set_val('traj.phase0.states:ay', phase.interp('ay', [0.0, 0.0]), units='m/s**2')
        p.set_val('traj.phase0.states:n', phase.interp('n', [0.0, 0.0]), units='m')
        # initial guess for what the final time should be
        p.set_val('traj.phase0.states:t', phase.interp('t', [0.0, 100.0]), units='s')

        # Controls
        # a small amount of thrust can speed up convergence
        p.set_val('traj.phase0.controls:delta', phase.interp('delta', [0.0, 0.0]), units='rad')
        p.set_val('traj.phase0.controls:thrust', phase.interp('thrust', [0.1, 0.1]), units=None)

        dm.run_problem(p, run_driver=True)
        print('Optimization finished')

        # Test this example in Dymos' continuous integration process
        assert_near_equal(p.get_val('traj.phase0.timeseries.states:t')[-1], 22.2657, tolerance=0.01)

        # Get optimized time series
        n = p.get_val('traj.phase0.timeseries.states:n')
        s = p.get_val('traj.phase0.timeseries.time')
        V = p.get_val('traj.phase0.timeseries.states:V')
        thrust = p.get_val('traj.phase0.timeseries.controls:thrust')
        delta = p.get_val('traj.phase0.timeseries.controls:delta')
        power = p.get_val('traj.phase0.timeseries.power', units='W')

        print("Plotting")

        # We know the optimal distance from the centerline (n). To transform this into the racing
        # line we fit a spline to the displaced points. This will let us plot the racing line in
        # x/y coordinates
        normals = get_gate_normals(finespline, slope)
        newgates = []
        newnormals = []
        newn = []
        for i in range(len(n)):
            index = ((s[i] / s_final) * np.array(finespline).shape[1]).astype(
                int)  # interpolation to find the appropriate index
            if index[0] == np.array(finespline).shape[1]:
                index[0] = np.array(finespline).shape[1] - 1
            if i > 0 and s[i] == s[i - 1]:
                continue
            else:
                newgates.append([finespline[0][index[0]], finespline[1][index[0]]])
                newnormals.append(normals[index[0]])
                newn.append(n[i][0])

        newgates = reverse_transform_gates(newgates)
        displaced_gates = set_gate_displacements(newn, newgates, newnormals)
        displaced_gates = np.array((transform_gates(displaced_gates)))

        npoints = 1000
        # fit the racing line spline to npoints
        displaced_spline, gates, gatesd, curv, slope = get_spline(displaced_gates, 1 / npoints, 0)

        plt.rcParams.update({'font.size': 12})

        def plot_track_with_data(state, s):
            # this function plots the track
            state = np.array(state)[:, 0]
            s = np.array(s)[:, 0]
            s_new = np.linspace(0, s_final, npoints)

            # Colormap and norm of the track plot
            cmap = mpl.cm.get_cmap('viridis')
            norm = mpl.colors.Normalize(vmin=np.amin(state), vmax=np.amax(state))

            fig, ax = plt.subplots(figsize=(15, 6))
            # establishes the figure axis limits needed for plotting the track below
            plt.plot(displaced_spline[0], displaced_spline[1], linewidth=0.1, solid_capstyle="butt")

            plt.axis('equal')
            # the linewidth is set in order to match the width of the track
            plt.plot(finespline[0], finespline[1], 'k',
                     linewidth=linewidth_from_data_units(8.5, ax),
                     solid_capstyle="butt")
            plt.plot(finespline[0], finespline[1], 'w', linewidth=linewidth_from_data_units(8, ax),
                     solid_capstyle="butt")  # 8 is the width, and the 8.5 wide line draws 'kerbs'
            plt.xlabel('x (m)')
            plt.ylabel('y (m)')

            # plot spline with color
            for i in range(1, len(displaced_spline[0])):
                s_spline = s_new[i]
                index_greater = np.argwhere(s >= s_spline)[0][0]
                index_less = np.argwhere(s < s_spline)[-1][0]

                x = s_spline
                xp = np.array([s[index_less], s[index_greater]])
                fp = np.array([state[index_less], state[index_greater]])
                interp_state = np.interp(x, xp,
                                         fp)  # interpolate the given state to calculate the color

                # calculate the appropriate color
                state_color = norm(interp_state)
                color = cmap(state_color)
                color = mpl.colors.to_hex(color)

                # the track plot consists of thousands of tiny lines:
                point = [displaced_spline[0][i], displaced_spline[1][i]]
                prevpoint = [displaced_spline[0][i - 1], displaced_spline[1][i - 1]]
                if i <= 5 or i == len(displaced_spline[0]) - 1:
                    plt.plot([point[0], prevpoint[0]], [point[1], prevpoint[1]], color,
                             linewidth=linewidth_from_data_units(1.5, ax), solid_capstyle="butt",
                             antialiased=True)
                else:
                    plt.plot([point[0], prevpoint[0]], [point[1], prevpoint[1]], color,
                             linewidth=linewidth_from_data_units(1.5, ax),
                             solid_capstyle="projecting", antialiased=True)

            clb = plt.colorbar(mpl.cm.ScalarMappable(norm=norm, cmap=cmap), fraction=0.02,
                               pad=0.04)  # add colorbar

            if np.array_equal(state, V[:, 0]):
                clb.set_label('Velocity (m/s)')
            elif np.array_equal(state, thrust[:, 0]):
                clb.set_label('Thrust')
            elif np.array_equal(state, delta[:, 0]):
                clb.set_label('Delta')

            plt.tight_layout()
            plt.grid()

        # Create the plots
        plot_track_with_data(V, s)
        plot_track_with_data(thrust, s)
        plot_track_with_data(delta, s)

        # Plot the main vehicle telemetry
        fig, axes = plt.subplots(nrows=4, ncols=1, figsize=(15, 8))

        # Velocity vs s
        axes[0].plot(s,
                     p.get_val('traj.phase0.timeseries.states:V'), label='solution')

        axes[0].set_xlabel('s (m)')
        axes[0].set_ylabel('V (m/s)')
        axes[0].grid()
        axes[0].set_xlim(0, s_final)

        # n vs s
        axes[1].plot(s,
                     p.get_val('traj.phase0.timeseries.states:n', units='m'), label='solution')

        axes[1].set_xlabel('s (m)')
        axes[1].set_ylabel('n (m)')
        axes[1].grid()
        axes[1].set_xlim(0, s_final)

        # throttle vs s
        axes[2].plot(s, thrust)

        axes[2].set_xlabel('s (m)')
        axes[2].set_ylabel('thrust')
        axes[2].grid()
        axes[2].set_xlim(0, s_final)

        # delta vs s
        axes[3].plot(s,
                     p.get_val('traj.phase0.timeseries.controls:delta', units=None),
                     label='solution')

        axes[3].set_xlabel('s (m)')
        axes[3].set_ylabel('delta')
        axes[3].grid()
        axes[3].set_xlim(0, s_final)

        plt.tight_layout()

        # Performance constraint plot. Tire friction and power constraints
        fig, axes = plt.subplots(nrows=1, ncols=1, figsize=(15, 4))
        plt.subplots_adjust(right=0.82, bottom=0.14, top=0.97, left=0.07)

        axes.plot(s,
                  p.get_val('traj.phase0.timeseries.c_fl', units=None), label='c_fl')
        axes.plot(s,
                  p.get_val('traj.phase0.timeseries.c_fr', units=None), label='c_fr')
        axes.plot(s,
                  p.get_val('traj.phase0.timeseries.c_rl', units=None), label='c_rl')
        axes.plot(s,
                  p.get_val('traj.phase0.timeseries.c_rr', units=None), label='c_rr')

        axes.plot(s, power / pmax, label='Power')

        axes.legend(bbox_to_anchor=(1.04, 0.5), loc="center left")
        axes.set_xlabel('s (m)')
        axes.set_ylabel('Performance constraints')
        axes.grid()
        axes.set_xlim(0, s_final)

        plt.show()
示例#2
0
def _run_racecar_problem(transcription, timeseries=False):
    # change track here and in curvature.py. Tracks are defined in tracks.py
    track = ovaltrack

    # generate nodes along the centerline for curvature calculation (different
    # than collocation nodes)
    points = get_track_points(track)

    # fit the centerline spline.
    finespline, gates, gatesd, curv, slope = get_spline(points, s=0.0)
    # by default 10000 points
    s_final = track.get_total_length()

    # Define the OpenMDAO problem
    p = om.Problem(model=om.Group())

    # Define a Trajectory object
    traj = dm.Trajectory()
    p.model.add_subsystem('traj', subsys=traj)

    # Define a Dymos Phase object with radau Transcription
    phase = dm.Phase(ode_class=CombinedODE,
                     transcription=transcription(num_segments=50,
                                                 order=3,
                                                 compressed=True))

    traj.add_phase(name='phase0', phase=phase)

    # Set the time options, in this problem we perform a change of variables. So 'time' is
    # actually 's' (distance along the track centerline)
    # This is done to fix the collocation nodes in space, which saves us the calculation of
    # the rate of change of curvature.
    # The state equations are written with respect to time, the variable change occurs in
    # timeODE.py
    phase.set_time_options(fix_initial=True,
                           fix_duration=True,
                           duration_val=s_final,
                           targets=['curv.s'],
                           units='m',
                           duration_ref=s_final,
                           duration_ref0=10)

    # Define states
    phase.add_state('t',
                    fix_initial=True,
                    fix_final=False,
                    units='s',
                    lower=0,
                    rate_source='dt_ds',
                    ref=100)  # time
    phase.add_state(
        'n',
        fix_initial=False,
        fix_final=False,
        units='m',
        upper=4.0,
        lower=-4.0,
        rate_source='dn_ds',
        targets=['n'],
        ref=4.0)  # normal distance to centerline. The bounds on n define the
    # width of the track
    phase.add_state('V',
                    fix_initial=False,
                    fix_final=False,
                    units='m/s',
                    ref=40,
                    ref0=5,
                    rate_source='dV_ds',
                    targets=['V'])  # velocity
    phase.add_state(
        'alpha',
        fix_initial=False,
        fix_final=False,
        units='rad',
        rate_source='dalpha_ds',
        targets=['alpha'],
        ref=0.15)  # vehicle heading angle with respect to centerline
    phase.add_state(
        'lambda',
        fix_initial=False,
        fix_final=False,
        units='rad',
        rate_source='dlambda_ds',
        targets=['lambda'],
        ref=0.01
    )  # vehicle slip angle, or angle between the axis of the vehicle
    # and velocity vector (all cars drift a little)
    phase.add_state('omega',
                    fix_initial=False,
                    fix_final=False,
                    units='rad/s',
                    rate_source='domega_ds',
                    targets=['omega'],
                    ref=0.3)  # yaw rate
    phase.add_state('ax',
                    fix_initial=False,
                    fix_final=False,
                    units='m/s**2',
                    rate_source='dax_ds',
                    targets=['ax'],
                    ref=8)  # longitudinal acceleration
    phase.add_state('ay',
                    fix_initial=False,
                    fix_final=False,
                    units='m/s**2',
                    rate_source='day_ds',
                    targets=['ay'],
                    ref=8)  # lateral acceleration

    # Define Controls
    phase.add_control(name='delta',
                      units='rad',
                      lower=None,
                      upper=None,
                      fix_initial=False,
                      fix_final=False,
                      targets=['delta'],
                      ref=0.04)  # steering angle
    phase.add_control(
        name='thrust',
        units=None,
        fix_initial=False,
        fix_final=False,
        targets=['thrust']
    )  # the thrust controls the longitudinal force of the rear tires and is
    # positive while accelerating, negative while braking

    # Performance Constraints
    pmax = 960000  # W
    phase.add_path_constraint('power', upper=pmax,
                              ref=100000)  # engine power limit

    # The following four constraints are the tire friction limits, with 'rr' designating the
    # rear right wheel etc. This limit is computed in tireConstraintODE.py
    phase.add_path_constraint('c_rr', upper=1)
    phase.add_path_constraint('c_rl', upper=1)
    phase.add_path_constraint('c_fr', upper=1)
    phase.add_path_constraint('c_fl', upper=1)

    # Some of the vehicle design parameters are available to set here. Other parameters can
    # be found in their respective ODE files.
    phase.add_parameter(
        'M',
        val=800.0,
        units='kg',
        opt=False,
        targets=['car.M', 'tire.M', 'tireconstraint.M', 'normal.M'],
        static_target=True)  # vehicle mass
    phase.add_parameter('beta',
                        val=0.62,
                        units=None,
                        opt=False,
                        targets=['tire.beta'],
                        static_target=True)  # brake bias
    phase.add_parameter('CoP',
                        val=1.6,
                        units='m',
                        opt=False,
                        targets=['normal.CoP'],
                        static_target=True)  # center of pressure location
    phase.add_parameter('h',
                        val=0.3,
                        units='m',
                        opt=False,
                        targets=['normal.h'],
                        static_target=True)  # center of gravity height
    phase.add_parameter('chi',
                        val=0.5,
                        units=None,
                        opt=False,
                        targets=['normal.chi'],
                        static_target=True)  # roll stiffness
    phase.add_parameter('ClA',
                        val=4.0,
                        units='m**2',
                        opt=False,
                        targets=['normal.ClA'],
                        static_target=True)  # downforce coefficient*area
    phase.add_parameter('CdA',
                        val=2.0,
                        units='m**2',
                        opt=False,
                        targets=['car.CdA'],
                        static_target=True)  # drag coefficient*area

    # Minimize final time.
    # note that we use the 'state' time instead of Dymos 'time'
    phase.add_objective('t', loc='final')

    # Add output timeseries
    if timeseries:
        phase.add_timeseries_output('*')

    # Link the states at the start and end of the phase in order to ensure a continous lap
    traj.link_phases(phases=['phase0', 'phase0'],
                     vars=['V', 'n', 'alpha', 'omega', 'lambda', 'ax', 'ay'],
                     locs=('final', 'initial'))

    # Set the driver. IPOPT or SNOPT are recommended but SLSQP might work.
    p.driver = om.pyOptSparseDriver(optimizer='IPOPT')

    p.driver.opt_settings['mu_init'] = 1e-3
    p.driver.opt_settings['max_iter'] = 500
    p.driver.opt_settings['acceptable_tol'] = 1e-3
    p.driver.opt_settings['constr_viol_tol'] = 1e-3
    p.driver.opt_settings['compl_inf_tol'] = 1e-3
    p.driver.opt_settings['acceptable_iter'] = 0
    p.driver.opt_settings['tol'] = 1e-3
    p.driver.opt_settings['nlp_scaling_method'] = 'none'
    p.driver.opt_settings['print_level'] = 5
    p.driver.opt_settings[
        'nlp_scaling_method'] = 'gradient-based'  # for faster convergence

    # Allow OpenMDAO to automatically determine our sparsity pattern.
    # Doing so can significant speed up the execution of Dymos.
    p.driver.declare_coloring()

    # Setup the problem
    p.setup(check=True)  # force_alloc_complex=True
    # Now that the OpenMDAO problem is setup, we can set the values of the states.

    # States
    # non-zero velocity in order to protect against 1/0 errors.
    p.set_val('traj.phase0.states:V', phase.interp('V', [20, 20]), units='m/s')
    p.set_val('traj.phase0.states:lambda',
              phase.interp('lambda', [0.0, 0.0]),
              units='rad')
    # all other states start at 0
    p.set_val('traj.phase0.states:omega',
              phase.interp('omega', [0.0, 0.0]),
              units='rad/s')
    p.set_val('traj.phase0.states:alpha',
              phase.interp('alpha', [0.0, 0.0]),
              units='rad')
    p.set_val('traj.phase0.states:ax',
              phase.interp('ax', [0.0, 0.0]),
              units='m/s**2')
    p.set_val('traj.phase0.states:ay',
              phase.interp('ay', [0.0, 0.0]),
              units='m/s**2')
    p.set_val('traj.phase0.states:n', phase.interp('n', [0.0, 0.0]), units='m')
    # initial guess for what the final time should be
    p.set_val('traj.phase0.states:t',
              phase.interp('t', [0.0, 100.0]),
              units='s')

    # Controls
    p.set_val('traj.phase0.controls:delta',
              phase.interp('delta', [0.0, 0.0]),
              units='rad')
    p.set_val('traj.phase0.controls:thrust',
              phase.interp('thrust', [0.1, 0.1]),
              units=None)
    # a small amount of thrust can speed up convergence

    dm.run_problem(p, run_driver=True, simulate=False, make_plots=False)
    print('Optimization finished')

    t = p.get_val('traj.phase0.timeseries.states:t')
    assert_near_equal(t[-1], 22.2657, tolerance=0.01)