示例#1
0
    def gait2dpi_u(self, xs, xsd, xs_meas, vs, u, uo, p, index_open,
                   index_imp):

        half_state = int(self.num_states / 2)
        x = xs[:half_state]
        xd = xs[half_state:]
        xdd = xsd[half_state:]

        QQ, dQQ_dx, dQQ_dxd, dQQ_dxdd, GRF, dGRF_dx, dGRF_dxd, sticks = \
                                    autolev_rhs(x, xd, xdd, vs, self.constants_dict)

        tor_imp, dtor_dx, dtor_dp = self.impedance_control(
            xs, xs_meas, p, index_imp)

        tor_nor, dtor_duo = self.normal_torque(uo, index_open)

        tor = np.zeros(half_state)

        tor[self.index_open_motion] = u
        tor[self.index_control_motion] = tor_nor + tor_imp

        f = np.zeros(self.num_states)
        f[:half_state] = xsd[:half_state] - xs[half_state:]
        f[half_state:] = (tor - QQ) / self.scaling

        dfdx = np.zeros((self.num_states, self.num_states))
        dfdx[:half_state, half_state:] = -np.eye(half_state)
        dfdx[half_state:, :half_state] = -np.reshape(
            dQQ_dx, (half_state, half_state)) / self.scaling
        dfdx[
            half_state:,
            half_state:] = -np.reshape(dQQ_dxd,
                                       (half_state, half_state)) / self.scaling
        dfdx[half_state +
             self.index_control_motion, :] += dtor_dx / self.scaling

        dfdxd = np.zeros((self.num_states, self.num_states))
        dfdxd[:half_state, :half_state] = np.eye(half_state)
        dfdxd[
            half_state:,
            half_state:] = -np.reshape(dQQ_dxdd,
                                       (half_state, half_state)) / self.scaling

        dfdu = np.zeros((self.num_states, self.num_open))
        dfdu[half_state +
             self.index_open_motion, :] = np.eye(self.num_open) / self.scaling

        dfduo = np.zeros((self.num_states, self.num_normal * self.num_control))
        dfduo[half_state +
              self.index_control_motion, :] = dtor_duo / self.scaling

        dfdp = np.zeros((self.num_states, self.num_par))
        dfdp[half_state +
             self.index_control_motion, :] = dtor_dp / self.scaling

        return f, dfdx, dfdxd, dfdu, dfduo, dfdp
    ax8.bar(xaxis, impedance[14:16], align='center', alpha=0.3)
    ax8.plot(xaxis, imped_sim[14:16], 'ro')

fig4.savefig(store_path + 'ForwardSimulation_ImpedanceGains.png')

constants_dict = map_values_to_autolev_symbols(constants_dict)

sticks_pred = np.zeros((num_nodes, 20))
qd = np.zeros(9)
qdd = np.zeros(9)
belts = np.zeros(num_nodes)
GRF = np.zeros((num_nodes, 6))

for o in range(num_nodes):
    _, _, _, _, GRF[o, :], _, _, sticks_pred[o, :] = \
                autolev_rhs(Result[o, :9], qd, qdd,
                            vs_meas[o, :], constants_dict)
    if o == 0:
        belts[o] = 0
    else:
        belts[o] = belts[o - 1] + np.sum(vs_meas[o, :] -
                                         (0.8 + 0.4 * T)) / 2 * 0.02

time_plot = np.linspace(0, num_nodes * interval, num_nodes)
anni_names = store_path + 'annimation.mp4'
fps_def = 12.5
animate_pendulum(time_plot, sticks_pred, belts, fps_def, filename=anni_names)

fig8 = plt.figure(figsize=(16, 8))
ax11 = fig8.add_subplot(3, 2, 1)
plt.ylabel('left Fx (N)', fontsize=14)
ax12 = fig8.add_subplot(3, 2, 2)
    def gait2dpi_u(self, xs, xsd, vs, u_close, u_stanceL, u_stanceR, u_ankle, iniL, iniR):
        
        # separate angle, anglar velocity, and anglar acceleration
        
        x = xs[:9]
        xd = xs[9:18]
        xdd = xsd[9:18]
        
        # gait 2d dynamics functions' residule, derivatives
        QQ, dQQ_dx, dQQ_dxd, dQQ_dxdd, GRF, dGRF_dx, dGRF_dxd, sticks = \
                                    autolev_rhs(x, xd, xdd, vs, self.constants_dict)
                                    

        if iniL[0] and iniR[0]:
            left_torque = u_stanceL
            right_torque = u_stanceR
            
            torque = np.zeros(9)
            torque[3:5] = left_torque
            torque[5] = u_ankle[0]
            torque[6:8] = right_torque
            torque[8] = u_ankle[1]
            
            dTorque_dConLOpen = np.zeros((9, 2))
            dTorque_dConROpen = np.zeros((9, 2))
            dTorque_dConAnkle = np.zeros((9, 2))
            dTorque_dConLOpen[3, 0] = 1
            dTorque_dConLOpen[4, 1] = 1
            dTorque_dConAnkle[5, 0] = 1
            dTorque_dConROpen[6, 0] = 1
            dTorque_dConROpen[7, 1] = 1
            dTorque_dConAnkle[8, 1] = 1
    
            f = np.zeros(self.num_states)
            
            f[:9] = xsd[:9] - xs[9:18]
            f[9:18] = (torque - QQ)/self.scaling
             
            df_dXs = np.zeros((self.num_states, self.num_states))
            df_dXsd = np.zeros((self.num_states, self.num_states))
            df_dConLOpen = np.zeros((self.num_states, 2))
            df_dConROpen = np.zeros((self.num_states, 2))
            df_dConAnkle = np.zeros((self.num_states, 2))
            df_dConClose = np.zeros((self.num_states, len(u_close)))
            
            df_dXs[:9,9:18] = -np.eye(9)
            df_dXs[9:18,:9] = (-np.reshape(dQQ_dx, (9, 9)))/self.scaling
            df_dXs[9:18,9:18] = (-np.reshape(dQQ_dxd, (9, 9)))/self.scaling
            
            df_dXs[10, 9] = 0.0
    		# force row 10 col 9 element in dfdx be zero, since it 
            # sometimes is zero, but sometimes is very small number (10^-10).
            # This caused jacobian structure unstable. Thereforce its element
            # is force to be zero here.
            
            df_dXsd[:9,:9] = np.eye(9)
            df_dXsd[9:18,9:18] = -np.reshape(dQQ_dxdd, (9, 9))/self.scaling
                        
            df_dConLOpen[9:18, :] = dTorque_dConLOpen/self.scaling
            df_dConROpen[9:18, :] = dTorque_dConROpen/self.scaling
            df_dConAnkle[9:18, :] = dTorque_dConAnkle/self.scaling
            
        elif iniL[0]:
            
            left_torque = u_stanceL
            
            (right_torque, dRightTorque_dXs, dRightTorque_dCon,
                fRight, dfRight_dXs, dfRight_dCon) =\
             self.swing_control(xs, u_close, iniR[1:3], iniR[3:5], sta_left=0, sta_right=1)
            
            torque = np.zeros(9)
            torque[3:5] = left_torque
            torque[5] = u_ankle[0]
            torque[6:8] = right_torque
            torque[8] = u_ankle[1]
            
            dTorque_dConClose = np.zeros((9, len(u_close)))
            dTorque_dConLOpen = np.zeros((9, 2))
            dTorque_dConAnkle = np.zeros((9, 2))
            
            dTorque_dConClose[6:8, :] = dRightTorque_dCon
            dTorque_dConLOpen[3, 0] = 1
            dTorque_dConLOpen[4, 1] = 1
            dTorque_dConAnkle[5, 0] = 1
            dTorque_dConAnkle[8, 1] = 1
            
            dTorque_dXs = np.zeros((9, self.num_states))
            dTorque_dXs[6:8, :] = dRightTorque_dXs
    
            f = np.zeros(self.num_states)
            
            f[:9] = xsd[:9] - xs[9:18]
            f[9:18] = (torque - QQ)/self.scaling
            f[18:22] = fRight
             
            df_dXs = np.zeros((self.num_states, self.num_states))
            df_dXsd = np.zeros((self.num_states, self.num_states))
            df_dConLOpen = np.zeros((self.num_states, 2))
            df_dConROpen = np.zeros((self.num_states, 2))
            df_dConAnkle = np.zeros((self.num_states, 2))
            df_dConClose = np.zeros((self.num_states, len(u_close)))
            
            df_dXs[:9,9:18] = -np.eye(9)
            df_dXs[9:18,:9] = (dTorque_dXs[:, :9]-np.reshape(dQQ_dx, (9, 9)))/self.scaling
            df_dXs[9:18,9:18] = (dTorque_dXs[:, 9:18]-np.reshape(dQQ_dxd, (9, 9)))/self.scaling
            df_dXs[9:18, 18:22] = dTorque_dXs[:, 18:22]/self.scaling
            df_dXs[18:22,:] = dfRight_dXs
            
            df_dXs[10, 9] = 0.0
    		# force row 10 col 9 element in dfdx be zero, since it 
            # sometimes is zero, but sometimes is very small number (10^-10).
            # This caused jacobian structure unstable. Thereforce its element
            # is force to be zero here.
            
            df_dXsd[:9,:9] = np.eye(9)
            df_dXsd[9:18,9:18] = -np.reshape(dQQ_dxdd, (9, 9))/self.scaling
                       
            df_dConLOpen[9:18, :] = dTorque_dConLOpen/self.scaling
            df_dConAnkle[9:18, :] = dTorque_dConAnkle/self.scaling
            
            df_dConClose[9:18, :] = dTorque_dConClose/self.scaling
            df_dConClose[18:22, :] = dfRight_dCon
            
        elif iniR[0]:
            
            (left_torque, dLeftTorque_dXs, dLeftTorque_dCon,
                fLeft, dfLeft_dXs, dfLeft_dCon) =\
             self.swing_control(xs, u_close, iniL[1:3], iniL[3:5], sta_left=1, sta_right=0)
            
            right_torque = u_stanceR
        
            torque = np.zeros(9)
            torque[3:5] = left_torque
            torque[5] = u_ankle[0]
            torque[6:8] = right_torque
            torque[8] = u_ankle[1]
            
            dTorque_dConClose = np.zeros((9, len(u_close)))
            dTorque_dConROpen = np.zeros((9, 2))
            dTorque_dConAnkle = np.zeros((9, 2))
            
            dTorque_dConClose[3:5, :] = dLeftTorque_dCon
            dTorque_dConAnkle[5, 0] = 1
            dTorque_dConROpen[6, 0] = 1
            dTorque_dConROpen[7, 1] = 1
            dTorque_dConAnkle[8, 1] = 1
            
            dTorque_dXs = np.zeros((9, self.num_states))
            dTorque_dXs[3:5, :] = dLeftTorque_dXs
    
            f = np.zeros(self.num_states)
            
            f[:9] = xsd[:9] - xs[9:18]
            f[9:18] = (torque - QQ)/self.scaling
            f[18:22] = fLeft
             
            df_dXs = np.zeros((self.num_states, self.num_states))
            df_dXsd = np.zeros((self.num_states, self.num_states))
            df_dConLOpen = np.zeros((self.num_states, 2))
            df_dConROpen = np.zeros((self.num_states, 2))
            df_dConAnkle = np.zeros((self.num_states, 2))
            df_dConClose = np.zeros((self.num_states, len(u_close)))
            
            df_dXs[:9,9:18] = -np.eye(9)
            df_dXs[9:18,:9] = (dTorque_dXs[:, :9]-np.reshape(dQQ_dx, (9, 9)))/self.scaling
            df_dXs[9:18,9:18] = (dTorque_dXs[:, 9:18]-np.reshape(dQQ_dxd, (9, 9)))/self.scaling
            df_dXs[9:18, 18:22] = dTorque_dXs[:, 18:22]/self.scaling
            df_dXs[18:22,:] = dfLeft_dXs
            
            df_dXs[10, 9] = 0.0
    		# force row 10 col 9 element in dfdx be zero, since it 
            # sometimes is zero, but sometimes is very small number (10^-10).
            # This caused jacobian structure unstable. Thereforce its element
            # is force to be zero here.
            
            df_dXsd[:9,:9] = np.eye(9)
            df_dXsd[9:18,9:18] = -np.reshape(dQQ_dxdd, (9, 9))/self.scaling
                       
            df_dConROpen[9:18, :] = dTorque_dConROpen/self.scaling
            df_dConAnkle[9:18, :] = dTorque_dConAnkle/self.scaling
            
            df_dConClose[9:18, :] = dTorque_dConClose/self.scaling
            df_dConClose[18:22, :] = dfLeft_dCon
        
        return f, df_dXs, df_dXsd, df_dConClose, df_dConLOpen, df_dConROpen, df_dConAnkle
                    StringP += " "
                StringP += "\n"
            outfile.write(StringP)

        show_nodes_plot = 1001

        with open(para_name, 'r') as f:
            subj_info = yaml.load(f)

        constants_dict = map_values_to_autolev_symbols(subj_info)

        sticks = np.zeros((show_nodes_plot, 20))
        belt_disp = np.zeros(show_nodes_plot)

        for qq in range(show_nodes_plot):

            x = show_motion_2d[qq, :9]
            xd = np.zeros(9)
            xdd = np.zeros(9)
            vs = show_belt[qq, :]

            if qq > 0:
                belt_disp[qq] = belt_disp[qq - 1] - (show_belt[
                    qq, 0] + show_belt[qq - 1, 0]) / 2 * 0.01 - 0.8 * 0.01

            _, _, _, _, _, _, _, sticks[qq, :] = \
                                autolev_rhs(x, xd, xdd, vs, constants_dict)

        t = show_motion[:show_nodes_plot, 0] - show_motion[0, 0]
        animate_pendulum(t, sticks, belt_disp, filename=video_save_name)