示例#1
0
    def follow_wp_path_opt_take_measurements(
            self,
            num_plot_points=250,
            model_type='log',
            b_take_meas=False,
            b_log_data=False,
            tolmm=10,
            tolrad=10e-3,
            start_wp=[1000, 1000, 0.5 * np.pi],
            sample_size=32):
        """
        :param b_take_meas:
        :param start_wp:
        :param sample_size:
        :return:
        """
        if b_take_meas is True:  # dont start SDR if not RSS measurements will be taken
            self.start_RfEar()
            self.__oRf.set_samplesize(sample_size)
            sample_size = self.__oRf.get_samplesize()

        wplist_filename = hc_tools.select_file()
        print(wplist_filename)
        wp_append_list = []
        with open(wplist_filename, 'r') as wpfile:
            for i, line in enumerate(wpfile):
                # print('line = ' + line)
                # print(line.split(','))
                temp_list = line.split(',')
                wp_append_list.append(map(float, temp_list[0:-1]))

#        print(str(np.asarray(wp_append_list)))
        wp_list = np.asarray(wp_append_list)
        if b_log_data:
            measdata_filename = hc_tools.save_as_dialog()
#       print(measdata_filename)

        num_wp = len(wp_list)
        print('Number of way points: ' + str(num_wp))
        start_time = t.time()

        data_list = []
        meas_counter = 0
        time_elapsed = 0.0
        tolx_mm = tolmm  # mm
        toly_mm = tolmm  # mm
        tola_rad = tolrad  # rad

        b_ekf = True
        if b_ekf is True:
            # init EKF
            EKF = estimator.ExtendedKalmanFilter(model_type)
            import estimator_plot_tools
            EKF_plotter = estimator_plot_tools.EKF_Plot(
                EKF.get_tx_pos(), model_type)

        # follow wp sequence
        for wp in wp_list:
            print('wp in list = ' + str(wp))
            # go to wp
            self.set_target_wp(wp)
            self.start_moving_gantry_to_target()
            not_arrived_at_wp = True
            print('Moving to wp = ' + str(wp))

            # following sequence
            while not_arrived_at_wp:
                meas_counter += 1
                time_elapsed = t.time() - start_time
                pos_x_mm, pos_y_mm, pos_a_rad = self.get_gantry_pos_xya_mmrad()

                if b_ekf is True:
                    EKF.ekf_prediction()
                    EKF.ekf_update(-85)
                    # EKF.check_valid_position_estimate()
                    # print(EKF.get_x_est())
                    EKF_plotter.add_x_est_to_plot(EKF.get_x_est())
                    EKF_plotter.update_meas_circles(EKF.get_z_meas(),
                                                    EKF.get_tx_alpha(),
                                                    EKF.get_tx_gamma(), True,
                                                    EKF.get_y_est())
                    EKF_plotter.plot_gantry_pos([pos_x_mm, pos_y_mm])
                    EKF_plotter.plot_ekf_pos_live(True, num_plot_points)
                    # EKF_plotter.add_p_cov_to_plot(EKF.get_p_mat())
                    #EKF_plotter.plot_p_cov(num_plot_points)

                if b_take_meas is True:
                    # taking measurements
                    freq_den_max, pxx_den_max = self.__oRf.get_rss_peaks()
                    data_row = np.append([
                        meas_counter, time_elapsed, pos_x_mm, pos_y_mm,
                        pos_a_rad
                    ], pxx_den_max)

                else:
                    data_row = [
                        meas_counter, time_elapsed, pos_x_mm, pos_y_mm,
                        pos_a_rad
                    ]

                # add new data to list
                data_list.append(data_row)

                # arrived at wp? -> go to next
                if self.confirm_arrived_at_target_wp(tolx_mm, toly_mm,
                                                     tola_rad):
                    not_arrived_at_wp = False

            meas_freq = meas_counter / time_elapsed
            print('Logging with avg. ' + str(meas_freq) + ' Hz')

        if b_log_data:
            with open(measdata_filename, 'w') as measfile:
                measfile.write('Measurement file for trajectory following\n')
                measfile.write('Measurement was taken on ' + t.ctime() + '\n')
                measfile.write('### begin grid settings\n')
                if b_take_meas is True:
                    measfile.write('sample size = ' + str(sample_size) +
                                   ' [*1024]\n')
                    measfile.write('avg. meas frequency = ' + str(meas_freq) +
                                   ' Hz\n')
                measfile.write('start_point =' + str(start_wp) + '\n')
                measfile.write('wp_list =' + str(wp_list) + '\n')
                if b_take_meas is True:
                    measfile.write(
                        'data format = [meas_counter, time_elapsed, pos_x_mm, pos_y_mm, pos_a_rad], pxx_den_max\n'
                    )
                else:
                    measfile.write(
                        'data format = [meas_counter, time_elapsed, pos_x_mm, pos_y_mm, pos_a_rad]\n'
                    )

                measfile.write('### begin data log\n')

                data_mat = np.asarray(data_list)
                for row in data_mat:
                    row_string = ''
                    for i in range(len(row)):
                        row_string += str(row[i]) + ','
                    row_string += '\n'
                    measfile.write(row_string)

                measfile.close()

        return True
示例#2
0
def main(simulate_meas,
         x_start=None,
         measfile_rel_path=None,
         cal_param_file=None,
         make_plot=False):
    """
    executive programm

    :param x_start:
    :param measfile_rel_path:
    :param cal_param_file:
    :param make_plot:
    :param simulate_meas:
    :return:
    """
    '''load measurement data'''
    meas_data, alpha = est_to.get_meas_values(
        simulate_meas,
        measfile_rel_path)  # possibly write this data into class
    # print('meas_data:\n' + str(meas_data))
    '''initialize values for EKF'''
    EKF = Extended_Kalman_Filter(
        alpha=alpha, x_start=x_start
    )  # initialize object ->check initial values in __init__ function
    est_to.read_measfile_header(
        object=EKF, analyze_tx=[1, 2],
        measfile_path=measfile_rel_path)  # write params from header in object
    # EKF.set_num_meas(10)
    EKF.set_cal_params(cal_param_file=cal_param_file)
    EKF.set_tx_param()

    EKF.set_alpha(alpha)
    EKF.set_initial_values()
    '''EKF loop'''
    num_meas = EKF.get_num_meas()
    x_est_list = []
    x_est_list.append(x_start)
    y_est_list = []
    rss_meas = []
    p_mat_list = []
    psi_list = []
    theta_cap_list = []

    for i in range(num_meas):
        print('Passage number:' + str(i + 1) + '\n')
        # print('meas_data num :' + str(i) + str(meas_data[i][:]) + '\n')

        # setting depth value from wp position to simulate depth sensor
        EKF.set_pos_real(meas_data[i][0:3])
        EKF.ekf_prediction()

        EKF.ekf_update(meas_data[i][:])

        # print('x_est = ')
        # print(str(EKF.get_x_est()) + '\n')
        '''make list for plotting'''
        x = EKF.get_x_est()[0][0]
        y = EKF.get_x_est()[1][0]
        z = EKF.get_x_est()[2][0]

        x_est_list.append([x, y, z])  # for plotting purposes
        # x_est_list.append(EKF.get_x_est())
        '''make list for plotting'''
        sigma_x = EKF.get_p_mat()[0][0]
        sigma_y = EKF.get_p_mat()[1][1]
        sigma_z = EKF.get_p_mat()[2][2]

        p_mat_list.append([sigma_x, sigma_y, sigma_z])  # for plotting purposes
        # x_est_list.append(EKF.get_x_est())

        y_tild_list = EKF.get_y_tild()

        y_est_list.append(EKF.get_y_est().tolist())

        theta_cap_list.append(EKF.get_theta_cap())

        meas = meas_data[i][:]
        rss_meas.append(meas[3:3 + EKF.get_tx_num()].tolist())

    print('\n* * * * * *\n' 'estimator.py stopped!\n' '* * * * * *\n')
    '''
    Plott functions from Jonas
    '''
    print('Start plotting results.\n')
    '''Erstellung der X und Y Koordinatenlisten zum einfachen und effizienteren Plotten'''
    x_n_x = [None] * num_meas
    x_n_y = [None] * num_meas
    x_n_z = [None] * num_meas

    x_est_x = [None] * len(x_est_list)
    x_est_y = [None] * len(x_est_list)
    x_est_z = [None] * len(x_est_list)

    tx_pos_x = [None] * len(EKF.get_tx_pos())
    tx_pos_y = [None] * len(EKF.get_tx_pos())
    tx_pos_z = [None] * len(EKF.get_tx_pos())

    for i in range(0, num_meas):
        x_n_x[i] = meas_data[i, 0]  # position of waypoints
        x_n_y[i] = meas_data[i, 1]  # position of waypoints
        x_n_z[i] = meas_data[i, 2]  # position of waypoints

    for i in range(0, len(x_est_list)):
        x_est_x[i] = x_est_list[i][0]
        x_est_y[i] = x_est_list[i][1]
        x_est_z[i] = x_est_list[i][2]

    for i in range(0, len(EKF.get_tx_pos())):
        tx_pos_x[i] = EKF.get_tx_pos()[i][0]
        tx_pos_y[i] = EKF.get_tx_pos()[i][1]
        tx_pos_z[i] = EKF.get_tx_pos()[i][2]

    fig = plt.figure(2, figsize=(10, 2.5))
    # fig = plt.figure(1, figsize=(25, 12))
    '''Fehlerplot ueber Koordinaten'''

    plot_fehlerplotueberkoordinaten = False

    if plot_fehlerplotueberkoordinaten:
        plt.subplot(144)

    x_est_fehler = [None] * len(x_est_x)

    for i in range(3, len(x_n_x)):
        x_est_fehler[i] = est_to.get_distance_1d(x_est_x[i], x_n_x[i - 1])
    # plt.plot(x_est_fehler)

    xmax = max(x_est_fehler)

    for i in range(3, len(x_n_y)):
        x_est_fehler[i] = est_to.get_distance_1d(x_est_y[i], x_n_y[i - 1])

    # plt.plot(x_est_fehler)

    ymax = max([max(x_est_fehler), xmax])

    a = np.asarray(x_est_list[3])
    b = meas_data[3 - 1, 0:3]

    for i in range(3, len(x_est_list)):
        x_est_fehler[i] = est_to.get_distance_2d(np.asarray(x_est_list[i]),
                                                 meas_data[i - 1, 0:3])

    # plt.plot(x_est_fehler)

    ymax = max([max(x_est_fehler), ymax])  # maximum error

    x_est_fehler_ges_mean = [np.mean(x_est_fehler[3:])] * len(x_est_x)

    x_est_fehler_ges_sdt = np.std(x_est_fehler[3:])  # just for 2D error??

    if plot_fehlerplotueberkoordinaten:
        plt.plot(x_est_fehler_ges_mean, '--')
        plt.xlabel('Messungsnummer')
        plt.ylabel('Fehler')
        plt.legend([
            'Fehler X-Koordinate', 'Fehler Y-Koordinate',
            '(Gesamt-) Abstandsfehler',
            ('Mittlerer Gesamtfehler = ' +
             str(np.round(x_est_fehler_ges_mean[0], 1)))
        ],
                   loc=0)
        plt.ylim(0, ymax + 300)
    '''Analyse der Einzelmessungen fuer einfacheres Tuning'''

    analyze_individual_meas = False

    if analyze_individual_meas:
        ekf_plotter = ept.EKF_Plot(EKF.get_tx_pos(), bplot_circles=True)
        # Einzelanalyse der Punkte mit Kreisen
        direct_terms = [None] * EKF.get_tx_num()

        for i in range(EKF.get_tx_num()):
            direct_terms[i] = np.log10(EKF.get_D_0_rx() * EKF.get_D_0_tx())

        for itx in range(len(x_n_x)):
            msg_x_est_temp = x_est_list[itx]
            # print('x= ' + str(msg_x_est))
            msg_yaw_rad = 0
            msg_z_meas = meas_data[itx, 3:(3 + EKF.get_tx_num())]
            msg_y_est = meas_data[itx, 3:(3 + EKF.get_tx_num())]
            msg_next_wp = meas_data[itx, 0:3]
            # print('wp=' + str(msg_next_wp))

            ekf_plotter.add_x_est_to_plot(msg_x_est_temp, msg_yaw_rad)
            ekf_plotter.update_next_wp(msg_next_wp)
            ekf_plotter.update_meas_circles(msg_z_meas,
                                            EKF.get_tx_lambda(),
                                            EKF.get_tx_gamma(),
                                            direct_terms,
                                            msg_y_est,
                                            b_plot_yest=True)
            ekf_plotter.plot_ekf_pos_live(b_yaw=False, b_next_wp=True)
            plt.show()  # Hier Breakpoint hinsetzen fuer Analyse der punkte
    '''Strecke im Scatterplot'''

    plot_3d = False

    if plot_3d:
        # ax = fig.add_subplot(121, projection='3d')
        ax = fig.add_subplot(111, projection='3d')
        ax.scatter(tx_pos_x,
                   tx_pos_y,
                   tx_pos_z,
                   marker="*",
                   c='k',
                   s=100,
                   depthshade=True,
                   zorder=0)
        ax.scatter(x_n_x,
                   x_n_y,
                   x_est_z,
                   marker="^",
                   c='c',
                   s=25,
                   depthshade=True,
                   zorder=1)
        ax.scatter(x_est_x,
                   x_est_y,
                   x_est_z,
                   marker="o",
                   c='r',
                   s=100,
                   depthshade=True,
                   zorder=2)
        xmin = float(
            min([min(x_n_x),
                 min(tx_pos_x),
                 min(x_n_y),
                 min(tx_pos_y)])) - 100.0
        xmax = float(
            max([max(x_n_x),
                 max(tx_pos_x),
                 max(x_n_y),
                 max(tx_pos_y)])) + 100.0
        ymin = float(min([min(x_n_y), min(tx_pos_y)])) - 100.0
        ymax = float(max([max(x_n_y), max(tx_pos_y)])) + 100.0

        if ymax < xmax:
            ymax += 0.5 * xmax
            ymin = ymax - xmax

        zmin = float(min([min(tx_pos_z), x_est_z]))
        zmax = float(max([max(tx_pos_z), x_est_z]))

        if zmax < xmax:
            zmax += 0.5 * xmax
            zmin = zmax - xmax

        elif zmax < ymax:
            zmax += 0.5 * ymax
            zmin = zmax - ymax

        ax.set_xlim(xmin, xmax)
        ax.set_ylim(ymin, ymax)
        ax.set_zlim(zmin, zmax)
        ax.set_xlabel('X - Achse', fontsize=20)
        ax.set_ylabel('Y - Achse', fontsize=20)
        ax.set_zlabel('Z - Achse', fontsize=20)
        ax.grid()
        ax.legend(
            ['Transmitter Antennen', 'Wahre Position', 'Geschaetzte Position'],
            loc=3)  # best:0, or=1, ol=2, ul=3, ur=4
        ax.set_title('Plot der wahren und geschaetzten Punkte', fontsize=25)
        # plt.show()

    else:
        '''
        here begins the tracking plot
        '''
        plt.subplot(221)
        # plt.subplot(111)

        plt.plot(x_n_x, x_n_y, marker=".", c='c')
        plt.plot(x_est_x, x_est_y, marker=".", c='r')
        plt.scatter(tx_pos_x, tx_pos_y, marker="*", c='k', s=100)

        xmin = float(min([min(x_n_x), min(tx_pos_x), min(x_est_x)])) - 100.0
        xmax = float(max([max(x_n_x), max(tx_pos_x), max(x_est_x)])) + 100.0
        ymin = float(min([min(x_n_y), min(tx_pos_y),
                          min(x_est_y)])) - 150.0  # - 350
        ymax = float(max([max(x_n_y), max(tx_pos_y), max(x_est_y)])) + 100.0
        ymax2 = ymin + ((ymax - ymin) / 2) + ((xmax - xmin) / 2)
        ymin2 = ymin + ((ymax - ymin) / 2) - ((xmax - xmin) / 2)

        plt.axis([xmin, xmax, ymin, ymax])
        plt.xlabel('X - Achse [mm]')
        plt.ylabel('Y - Achse [mm]')
        plt.grid()

        # add annotations

        # plt.legend(['$x_text{wahr}', '$x_text{R,est}', '$x_text{Ti}'], loc=3, ncol=3)  # best:0, or=1, ol=2, ul=3, ur=4
        # plt.title('Plot der wahren und geschaetzten Punkte', fontsize=25)
        # plt.annotate('Z-Koordinate Sendeantennen: ' + str(tx_h[0]) + 'mm', xy=(xmin+50, ymax-50), xytext=(xmin+50, ymax-50))
        # plt.annotate('Z-Koordinate Empfaengerantenne: ' + str(h_mauv) + 'mm', xy=(xmin+50, ymax-160), xytext=(xmin+50, ymax-160))
        plt.annotate('Durchschnittlicher Fehler: ' +
                     str(np.round(x_est_fehler_ges_mean[0], 0)) + ' +- ' +
                     str(np.round(x_est_fehler_ges_sdt)) + 'mm',
                     xy=(xmin + 50, ymax - 350),
                     xytext=(xmin + 50, ymax - 350))
        plt.annotate('$x_text{0,est,wahr}',
                     xy=(x_est_x[0], x_est_y[0]),
                     xytext=(x_est_x[0] - 50, x_est_y[0] - 100))
        # plt.annotate('$x_text{0,wahr}', xy=(x_n_x[0], x_n_y[0]), xytext=(x_n_x[0]-50, x_n_y[0]-100))

        # plt.show()
    '''Strecke im Linienplot'''

    plot_streckeimlinienplot = True

    if plot_streckeimlinienplot:
        # x_est_fehler = [None] * len((x_est_x))

        for i in range(len(x_est_x)):
            x_est_fehler[i] = est_to.get_distance_1d(x_est_x[i], x_est_y[i])

        # plt.figure(figsize=(12, 12))
        plt.subplot(222)

        plt.plot(range(0, num_meas), x_n_x, c='c')
        plt.plot(x_est_x, c='r')

        plt.plot(range(0, num_meas), x_n_y, c='m')
        plt.plot(x_est_y, c='y')

        p_est_x = [None] * len(p_mat_list)
        p_est_y = [None] * len(p_mat_list)
        p_est_z = [None] * len(p_mat_list)

        #plot variance next to coordinate
        # for i in range(0, len(p_mat_list)):
        #     p_est_x[i] = p_mat_list[i][0]
        #     p_est_y[i] = p_mat_list[i][1]
        #     p_est_z[i] = p_mat_list[i][2]
        #
        # x_plot = np.asarray(range(0, num_meas))
        # plt.fill_between(x_plot, np.asarray(x_est_x[:-1]) + np.asarray(p_est_x) * 0.1,
        #                  np.asarray(x_est_x[:-1]) - np.asarray(p_est_x) * 0.1, alpha=0.5)
        # plt.fill_between(x_plot, np.asarray(x_est_y[:-1]) + np.asarray(p_est_y) * 0.1,
        #                  np.asarray(x_est_y[:-1]) - np.asarray(p_est_y) * 0.1, alpha=0.5)

        plt.xlabel('Messungsnummer')
        plt.ylabel('Koordinate [mm]')
        plt.legend(['x_true', 'x_est', 'y_true', 'y_est'], loc=1)

        ymin = min([min(x_n_x), min(x_n_y), min(x_est_x), min(x_est_y)])
        ymax = max([max(x_n_x), max(x_n_y), max(x_est_x), max(x_est_y)])

        plt.ylim(ymin - 100, ymax + 100)
        plt.xticks(np.arange(0, len(x_n_x), step=5))
        plt.grid()

    plot_z = True
    if plot_z:
        # plt.figure(figsize=(12, 12))
        plt.subplot(223)

        plt.plot(range(0, num_meas), x_n_z, c='c')
        plt.plot(x_est_z, c='r')

        plt.xlabel('Messungsnummer')
        plt.ylabel('Koordinate [mm]')
        plt.legend(['z_true', 'z_est'], loc=1)

        ymin = min([min(x_n_z), min(x_est_z)])
        ymax = max([max(x_n_z), max(x_est_z)])

        plt.ylim(ymin - 100, ymax + 100)
        plt.xticks(np.arange(0, len(x_n_x), step=5))
        plt.grid()
        plt.annotate('Durchschnittlicher Fehler: ' +
                     str(np.round(x_est_fehler_ges_mean[0], 0)) + ' +- ' +
                     str(np.round(x_est_fehler_ges_sdt)) + 'mm',
                     xy=(xmin + 50, ymax - 350),
                     xytext=(xmin + 50, ymax - 350))

    plot_meas_sym = False
    if plot_meas_sym:
        # plt.figure(figsize=(12, 12))
        plt.subplot(224)

        # make list for every Tx
        y_est_0 = []
        y_est_1 = []
        y_est_2 = []
        y_est_3 = []
        y_est_4 = []

        rss_0 = []
        rss_1 = []
        rss_2 = []
        rss_3 = []
        rss_4 = []

        for i in range(EKF.get_num_meas()):
            y_est_0.append(y_est_list[i][0])
            y_est_1.append(y_est_list[i][1])
            # y_est_2.append(y_est_list[i][2])
            # y_est_3.append(y_est_list[i][3])
            # y_est_4.append(y_est_list[i][4])

            rss_0.append(rss_meas[i][0])
            rss_1.append(rss_meas[i][1])
            # rss_2.append(rss_meas[i][2])
            # rss_3.append(rss_meas[i][3])
            # rss_4.append(rss_meas[i][4])

        plt.plot(range(1, (num_meas + 1)), y_est_0, '.--', c='r')
        plt.plot(range(1, (num_meas + 1)), y_est_1, '.--', c='y')
        # plt.plot(range(1, (num_meas + 1)), y_est_2, '.--', c='m')

        plt.plot(range(1, (num_meas + 1)), rss_0, '.-', c='c')
        plt.plot(range(1, (num_meas + 1)), rss_1, '.-', c='g')
        # plt.plot(range(1, (num_meas + 1)), rss_2, '.-', c='b')

        ymin = min([min(y_est_0), min(rss_0), min(y_est_1), min(rss_1)])
        ymax = max([max(y_est_0), max(rss_0), max(y_est_1), max(rss_1)])

        # plt.axis([xmin, xmax, ymin, ymax])
        plt.xlabel('Messungsnummer')
        plt.ylabel('RSS')

        # plt.legend(['RSS_est_T1', 'RSS_est_T2', 'RSS_est_T3', 'RSS_true_T1', 'RSS_true_T2', 'RSS_true_T3'], loc=1)
        plt.legend(['RSS_est_T1', 'RSS_est_T2', 'RSS_true_T1', 'RSS_true_T2'],
                   loc=1)

        plt.ylim(ymin - 2, ymax + 2)
        plt.xticks(np.arange(0, len(y_est_0) + 1, step=5))
        plt.grid()

        plt.subplot(223)

        ymin = min([min(y_est_0), min(rss_0), min(y_est_1), min(rss_1)])
        ymax = max([max(y_est_0), max(rss_0), max(y_est_1), max(rss_1)])

        # plt.axis([xmin, xmax, ymin, ymax])
        plt.xlabel('Messungsnummer')
        plt.ylabel('RSS')
        # plt.plot(range(1, (num_meas + 1)), y_est_3, '.--', c='y')
        # plt.plot(range(1, (num_meas + 1)), y_est_4, '.--', c='b')
        #
        # plt.plot(range(1, (num_meas + 1)), rss_3, '.-', c='g')
        # plt.plot(range(1, (num_meas + 1)), rss_4, '.-', c='c')
        plt.legend(['RSS_est_T4', 'RSS_est_T5', 'RSS_true_T4', 'RSS_true_T5'],
                   loc=1)

        plt.ylim(ymin - 2, ymax + 2)
        plt.xticks(np.arange(0, len(y_est_0) + 1, step=5))
        plt.grid()
    '''Fehlerhistogramm'''

    plot_fehlerhistogramm = False

    if plot_fehlerhistogramm:
        plt.subplot(248)
        plt.hist(x_est_fehler[3:], 30, (0, 800))  # 800
        plt.xlabel('Fehler')
        plt.ylabel('Anzahl der Vorkommnisse')

    plt.subplots_adjust(left=None,
                        bottom=None,
                        right=None,
                        top=None,
                        wspace=0.4,
                        hspace=None)
    plt.show()

    fig_1 = plt.figure(1, figsize=(25, 12))
    '''Theta'''
    fig_1.add_subplot(131)

    return True
示例#3
0
import socket_server
import estimator
import estimator_plot_tools
import numpy as np
import time as t


""" TX position StillWasserBecker """
#tx_pos = [[520.0, 430.0], [1540.0, 430.0], [2570.0, 430.0], [2570.0, 1230.0], [1540.0, 1230.0], [530.0, 1230.0]]
""" TX position - origin at Beacon #1 """
tx_pos = tx_6pos = [[0, 0], [1000, 0], [2000, 0], [2000, 900], [1000, 900], [0, 900]]

ekf_plotter = estimator_plot_tools.EKF_Plot(tx_pos, bplot_circles=False)

"""
WARNING!!!!!!
IF YOU CHANGE THE ALPHA + GAMMA VALUES IN EKF YOU HAVE!!!! to change them here MANUALLY!!!
"""
tx_alpha = [0.01149025464796399, 0.016245419273983631, 0.011352095690562954, 0.012125937076390217, 0.0092717529591962722, 0.01295918160582895]
tx_gamma = [-0.49471304043015696, -1.2482393190627841, -0.17291318936462172, -0.61587988305564456, 0.99831151034040444, 0.85711994311461936]
"""
WARNING!!!!!!
IF YOU CHANGE THE ALPHA + GAMMA VALUES IN EKF YOU HAVE!!!! to change them here MANUALLY!!!
"""
server_ip = '192.168.0.100'  # thinkpad ethernet
#server_ip = '192.168.0.101'  # thinkpad wifi-intern
server_port = 50008

soc_server = socket_server.SocServer(server_ip, server_port)

示例#4
0
phi_cap_t = [0.0]*tx_num
theta_cap_t = [0.0]*tx_num
psi_low_t = [0.0]*tx_num
theta_low_t = [0.0]*tx_num

'''Einstellungen fuer Messwerterzeugung'''
messung_benutzen = True
if messung_benutzen:
    '''Laden der Messdatei'''
    plotdata_mat = analyze_measdata_from_file(measfilename='wp_list_2018_08_10_grid_meas_d505025_measurement_h500.txt')

extra_plotting = False
direct_terms = [0.0] * tx_num
if extra_plotting:
    '''Setup fuer Kreisplots'''
    ekf_plotter = ept.EKF_Plot(tx_pos, bplot_circles=True)

    for i in range(tx_num):
        direct_terms[i] = np.log10(directivity_r * directivities_t[i])

    '''Setup fuer ytild Plot'''
    fig_ytild_p = plt.figure(42)
    sub1_ytild = fig_ytild_p.add_subplot(121)
    linspace_ploty_txnum = np.linspace(1, tx_num, tx_num)
    ydata_ploty = np.linspace(-20, 20, tx_num)
    line1sub1, = sub1_ytild.plot(linspace_ploty_txnum, ydata_ploty, 'r-')  # Returns a tuple of line objects, thus the comma
    plt.grid()

    '''Setup fuer P Plot'''
    sub2_pmat = fig_ytild_p.add_subplot(122)
    linspace_plotp = [1, 2]