Example #1
0
def comparaison_minkowski(ds, N):
	"""
		makes a comparison for N different values of r factor in Minkowski formula error.
		uses the same generated dataset. 
		In : a Vehicle from Vehicle class in vehicle.py ; N : nb of points for the plot.
	"""

	res_V_sigma = np.zeros(N)
	res_V_mu = np.zeros(N)
	res_omega_sigma = np.zeros(N)
	res_omega_mu = np.zeros(N)

	r_minkowski = np.linspace(1,2,N)

	i=0

	for _r in r_minkowski:              #be careful : _r global variable.

		def minkowski_loss(y_true, y_pred):
			return K.mean(K.pow(K.abs(y_pred-y_true),_r))

		Rr_est, Rl_est, L_est = odo.all_param_AN(ds, minkowski_loss)

		residuals = odo.residuals(ds, Rr_est, Rl_est, L_est, False)
		res_V_sigma[i] = residuals[0][2]
		res_V_mu[i] = residuals[0][1]
		res_omega_sigma[i] = residuals[1][2]
		res_omega_mu[i] = residuals[1][1]

		i+=1


	plt.figure()

	plt.subplot(221)
	plt.plot(r_minkowski, res_V_sigma)
	plt.xlabel("$r_m$")
	plt.ylabel("ecart type")
	plt.title("V residuals")
	plt.subplot(222)

	plt.plot(r_minkowski, res_V_mu)
	plt.xlabel("$r_m$")
	plt.ylabel("moyenne")
	plt.title("V residuals")
	plt.subplot(223)

	plt.plot(r_minkowski, res_omega_sigma)
	plt.xlabel("$r_m$")
	plt.ylabel("ecart type")
	plt.title("omega residuals")
	plt.subplot(224)

	plt.plot(r_minkowski, res_omega_mu)
	plt.xlabel("$r_m$")
	plt.ylabel("moyene")
	plt.title("omega residuals")
Example #2
0
def comparison_noise(sigma_V, sigma_omega):
    """
		makes a comparison of relative error  according 3 methods
		Method 1 : pseudoinv
		Method 2 : 2 AN
		Method 3 : single AN
		In : sigma_V : noise for V
			 sigma_omega : noise for omega
		Out : an array([[error_Rr_method1, error_Rl_method1, error_L_method1],
						[error_Rr_method2, error_Rl_method2, error_L_method2],
						[error_Rr_method3, error_Rl_method3, error_L_method3]])
	"""
    Robot = Vehicle(0.08, 0.081, 0.2, sigma_V, sigma_omega)
    Rr = Robot.right_wheel_radius
    Rl = Robot.left_wheel_radius
    L = Robot.space_wheels

    ds = Robot.generate_data(-10, 10, 5000)

    Rr_est3, Rl_est3, L_est3 = odo.all_param_AN(ds)

    Rr_est2, Rl_est2 = odo.wheels_radius_AN(ds)
    L_est2 = odo.space_wheels_AN(ds, (Rr_est2, Rl_est2))

    Rr_est1, Rl_est1 = odo.wheels_radius_INV(ds)
    L_est1 = odo.space_wheels_INV(ds, (Rr_est1, Rl_est1))

    error = np.array([[
        abs((Rr_est1 - Rr) / Rr),
        abs((Rl_est1 - Rl) / Rl),
        abs((L_est1 - L) / L)
    ],
                      [
                          abs((Rr_est2 - Rr) / Rr),
                          abs((Rl_est2 - Rl) / Rl),
                          abs((L_est2 - L) / L)
                      ],
                      [
                          abs((Rr_est3 - Rr) / Rr),
                          abs((Rl_est3 - Rl) / Rl),
                          abs((L_est3 - L) / L)
                      ]])
    return error
Example #3
0
def comparaison_minkowski(Robot, ds, N):
    """
		makes a comparison for N different values of r factor in Minkowski formula error.
		uses the same generated dataset. 
		In : a Vehicle from Vehicle class in vehicle.py ; N : nb of points for the plot.
	"""

    Rr = Robot.right_wheel_radius  #true Rr
    Rl = Robot.left_wheel_radius  #true Rl
    L = Robot.space_wheels  #true L
    relative_error = []
    r_minkowski = np.linspace(1, 2, N)
    for _r in r_minkowski:  #be careful : _r global variable.

        def minkowski_loss(y_true, y_pred):
            return K.mean(K.pow(K.abs(y_pred - y_true), _r))

        Rr_est, Rl_est, L_est = odo.all_param_AN(ds, minkowski_loss)

        relative_error.append([
            100 * abs(Rr_est - Rr) / Rr, 100 * abs(Rl_est - Rl) / Rl,
            100 * abs(L_est - L) / L
        ])
        #relative error = N*3 array

    plt.figure()
    plt.subplot(311)
    plt.plot(r_minkowski, [relative_error[i][0] for i in range(N)])
    plt.xlabel("r")
    plt.ylabel("erreur relative")
    plt.title("roue droite")
    plt.subplot(312)
    plt.plot(r_minkowski, [relative_error[i][1] for i in range(N)])
    plt.xlabel("r")
    plt.ylabel("erreur relative")
    plt.title("roue gauche")
    plt.subplot(313)
    plt.plot(r_minkowski, [relative_error[i][2] for i in range(N)])
    plt.xlabel("r")
    plt.ylabel("erreur relative")
    plt.title("espace")
    plt.show()
Example #4
0
def comparison_noise(sigma_V_max, sigma_omega_max):
	Npoints = 5 #nb of point for the plot
	tab_sigma_V = np.linspace(0, sigma_V_max, Npoints)
	tab_sigma_omega = np.linspace(0, sigma_omega_max, Npoints)

	res_V_sigma1 = np.zeros(Npoints)
	res_V_mu1 = np.zeros(Npoints)
	res_omega_sigma1 = np.zeros(Npoints)
	res_omega_mu1 = np.zeros(Npoints)
	res_V_sigma2 = np.zeros(Npoints)
	res_V_mu2 = np.zeros(Npoints)
	res_omega_sigma2 = np.zeros(Npoints)
	res_omega_mu2 = np.zeros(Npoints)

	for i in range(Npoints):

		Robot = Vehicle(0.08, 0.081, 0.2, tab_sigma_V[i], tab_sigma_omega[i])

		ds = Robot.generate_data(-10, 10, 5000)

		Rr_est1, Rl_est1, L_est1 = odo.all_param_AN(ds)
		residuals1 = odo.residuals(ds, Rr_est1, Rl_est1, L_est1)
		res_V_sigma1[i] = residuals1[0][2]
		res_V_mu1[i] = residuals1[0][1]
		res_omega_sigma1[i] = residuals1[1][2]
		res_omega_mu1[i] = residuals1[1][1]

		Rr_est2, Rl_est2 = odo.wheels_radius_INV(ds)
		L_est2 = odo.space_wheels_INV(ds, (Rr_est2, Rl_est2))
		residuals2 = odo.residuals(ds, Rr_est2, Rl_est2, L_est2)
		res_V_sigma2[i] = residuals2[0][2]
		res_V_mu2[i] = residuals2[0][1]
		res_omega_sigma2[i] = residuals2[1][2]
		res_omega_mu2[i] = residuals2[1][1]

	plt.figure()
	plt.subplot(221)
	plt.plot(tab_sigma_V, res_V_sigma1, label="AN")
	plt.plot(tab_sigma_V, res_V_sigma2, label="INV")
	plt.xlabel("$\sigma_V$")
	plt.ylabel("ecart type")
	plt.title("V residuals")
	plt.legend()
	plt.subplot(222)
	plt.plot(tab_sigma_V, res_V_mu1)
	plt.plot(tab_sigma_V, res_V_mu2)
	plt.xlabel("$\sigma_V$")
	plt.ylabel("moyenne")
	plt.title("V residuals")
	plt.subplot(223)
	plt.plot(tab_sigma_omega, res_omega_sigma1)
	plt.plot(tab_sigma_omega, res_omega_sigma2)
	plt.xlabel("$\sigma_{\omega}$")
	plt.ylabel("ecart type")
	plt.title("omega residuals")
	plt.subplot(224)
	plt.plot(tab_sigma_omega, res_omega_mu1)
	plt.plot(tab_sigma_omega, res_omega_mu2)
	plt.xlabel("$\sigma_{\omega}$")
	plt.ylabel("moyene")
	plt.title("omega residuals")
Example #5
0
def comparison_ANs(sigma_V_max, sigma_omega_max):
	"""
		draws a comparison of two-AN algo and all_param_An algo according noise. No outliers.
		sigma(noise)_V = [0, ..., sigma_V_max]
		sigma(noise)_omega = [0, ..., sigma_omega_max]
		Comparison based on mean and std of residuals. (fct odometry.residuals())
		Results : no difference !
	"""
	Npoints = 5 #nb of point for the plot
	tab_sigma_V = np.linspace(0, sigma_V_max, Npoints)
	tab_sigma_omega = np.linspace(0, sigma_omega_max, Npoints)

	res_V_sigma1 = np.zeros(Npoints)
	res_V_mu1 = np.zeros(Npoints)
	res_omega_sigma1 = np.zeros(Npoints)
	res_omega_mu1 = np.zeros(Npoints)
	res_V_sigma2 = np.zeros(Npoints)
	res_V_mu2 = np.zeros(Npoints)
	res_omega_sigma2 = np.zeros(Npoints)
	res_omega_mu2 = np.zeros(Npoints)

	for i in range(Npoints):

		Robot = Vehicle(0.08, 0.081, 0.2, tab_sigma_V[i], tab_sigma_omega[i])

		ds = Robot.generate_data(-10, 10, 5000)

		Rr_est1, Rl_est1 = odo.wheels_radius_AN(ds)
		L_est1 = odo.space_wheels_AN(ds, (Rr_est1, Rl_est1))
		residuals1 = odo.residuals(ds, Rr_est1, Rl_est1, L_est1)
		res_V_sigma1[i] = residuals1[0][2]
		res_V_mu1[i] = residuals1[0][1]
		res_omega_sigma1[i] = residuals1[1][2]
		res_omega_mu1[i] = residuals1[1][1]

		Rr_est2, Rl_est2, L_est2 = odo.all_param_AN(ds)
		residuals2 = odo.residuals(ds, Rr_est2, Rl_est2, L_est2)
		res_V_sigma2[i] = residuals2[0][2]
		res_V_mu2[i] = residuals2[0][1]
		res_omega_sigma2[i] = residuals2[1][2]
		res_omega_mu2[i] = residuals2[1][1]

	plt.figure()
	plt.subplot(221)
	plt.plot(tab_sigma_V, res_V_sigma1, label="2ANs")
	plt.plot(tab_sigma_V, res_V_sigma2, label="1ANs")
	plt.xlabel("$\sigma_V$")
	plt.ylabel("ecart type")
	plt.title("V residuals")
	plt.legend()
	plt.subplot(222)
	plt.plot(tab_sigma_V, res_V_mu1)
	plt.plot(tab_sigma_V, res_V_mu2)
	plt.xlabel("$\sigma_V$")
	plt.ylabel("moyenne")
	plt.title("V residuals")
	plt.subplot(223)
	plt.plot(tab_sigma_omega, res_omega_sigma1)
	plt.plot(tab_sigma_omega, res_omega_sigma2)
	plt.xlabel("$\sigma_{\omega}$")
	plt.ylabel("ecart type")
	plt.title("omega residuals")
	plt.subplot(224)
	plt.plot(tab_sigma_omega, res_omega_mu1)
	plt.plot(tab_sigma_omega, res_omega_mu2)
	plt.xlabel("$\sigma_{\omega}$")
	plt.ylabel("moyene")
	plt.title("omega residuals")