kalman_experiments/euler_ekf.py

190 lines
5.0 KiB
Python

import numpy as np
from numpy.linalg import inv
import matplotlib.pyplot as plt
from math import cos, sin, tan, asin, pi
import pandas as pd
from scipy.signal import butter, filtfilt
def get_data(i, data):
x = data[0][i] # (41500, 1)
y = data[1][i] # (41500, 1)
z = data[2][i] # (41500, 1)
return x, y, z
def Euler_accel(ax, ay, az):
g = 9.8 # 9.8
theta = asin(ax / g)
phi = asin(-ay / (g * cos(theta)))
return phi, theta
def sec(theta):
return 1/cos(theta)
def Ajacob(xhat, rates, dt):
'''
:param xhat: State Variables(phi, theta, psi)
:param rates: angel speed(p,q,r)
:param dt: variable to make discrete form
'''
A = np.zeros([3,3])
phi = xhat[0]
theta = xhat[1]
p,q,r = rates[0], rates[1], rates[2]
A[0][0] = q * cos(phi)*tan(theta) - r*sin(phi)*tan(theta)
A[0][1] = q * sin(phi)*(sec(theta)**2) + r*cos(phi)*(sec(theta)**2)
A[0][2] = 0
A[1][0] = -q * sin(phi) - r * cos(phi)
A[1][1] = 0
A[1][2] = 0
A[2][0] = q * cos(phi) * sec(theta) - r * sin(phi) * sec(theta)
A[2][1] = q * sin(phi) * sec(theta)*tan(theta) + r*cos(phi)*sec(theta)*tan(theta)
A[2][2] = 0
A = np.eye(3) + A*dt
return A
def fx(xhat, rates, dt):
phi = xhat[0]
theta = xhat[1]
p,q,r = rates[0], rates[1], rates[2]
xdot = np.zeros([3,1])
xdot[0] = p + q * sin(phi) * tan(theta) + r * cos(phi)*tan(theta)
xdot[1] = q * cos(phi) - r * sin(phi)
xdot[2] = q * sin(phi)*sec(theta) + r * cos(phi) * sec(theta)
xp = xhat.reshape(-1,1) + xdot*dt # xhat : (3,) --> (3,1)
return xp
def Euler_EKF(z, rates, dt):
global firstRun
global Q, H, R
global x, P
if firstRun:
H = np.array([[1,0,0],[0,1,0]])
Q = np.array([[0.0001,0,0],[0,0.0001,0],[0,0,0.1]])
R = 10 * np.eye(2)
x = np.array([0, 0, 0]).transpose()
P = 10 * np.eye(3)
firstRun = False
else:
A = Ajacob(x, rates, dt)
Xp = fx(x, rates, dt) # Xp : State Variable Prediction
Pp = A @ P @ A.T + Q # Error Covariance Prediction
K = (Pp @ H.T) @ inv(H@Pp@H.T + R) # K : Kalman Gain
x = Xp + K@(z.reshape(-1,1) - H@Xp) # Update State Variable Estimation
P = Pp - K@H@Pp # Update Error Covariance Estimation
phi = x[0]
theta = x[1]
psi = x[2]
return phi, theta, psi
def butter_lowpass_filter(data, cutoff, fs, order):
nyq = (0.5 * fs) # Nyquist Frequency
normal_cutoff = cutoff / nyq
# Get the filter coefficients
b, a = butter(order, normal_cutoff, btype='low', analog=False)
y = filtfilt(b, a, data)
return y
def normalize_data_vector(data, division):
return [i/division for i in list(data)]
H, Q, R = None, None, None
x, P = None, None
firstRun = True
division_param = 32768
# Filter requirements.
T = 5.0 # Sample Period
fs = 30.0 # sample rate, Hz
cutoff = 2 # desired cutoff frequency of the filter, Hz, slightly higher than actual 1.2 Hz
order = 2 # sin wave can be approx represented as quadratic
df = pd.read_csv('raw_data_6d.xls', sep='\t')
gyroX = df['%GyroX'].values
gyroY = df['GyroY'].values
gyroZ = df['GyroZ'].values
acceX = df['AcceX'].values
acceY = df['AcceY'].values
acceZ = df['AcceZ'].values
gyroX = normalize_data_vector(gyroX, division_param)
gyroY = normalize_data_vector(gyroY, division_param)
gyroZ = normalize_data_vector(gyroZ, division_param)
acceX = normalize_data_vector(acceX, division_param)
acceY = normalize_data_vector(acceY, division_param)
acceZ = normalize_data_vector(acceZ, division_param)
# gyroX = butter_lowpass_filter(gyroX, cutoff, fs, order)
# gyroY = butter_lowpass_filter(gyroY, cutoff, fs, order)
# gyroZ = butter_lowpass_filter(gyroZ, cutoff, fs, order)
# acceX = butter_lowpass_filter(acceX, cutoff, fs, order)
# acceY = butter_lowpass_filter(acceY, cutoff, fs, order)
# acceZ = butter_lowpass_filter(acceZ, cutoff, fs, order)
gyro_data = [gyroX, gyroY, gyroZ]
acce_data = [acceX, acceY, acceZ]
Nsamples = len(df)
EulerSaved = np.zeros([Nsamples,3])
dt = 0.01
for k in range(Nsamples):
p, q, r = get_data(k, gyro_data)
ax, ay, az = get_data(k, acce_data)
phi_a, theta_a = Euler_accel(ax, ay, az)
phi, theta, psi = Euler_EKF(np.array([phi_a, theta_a]).T, [p,q,r], dt)
if type(phi) == type(np.array([])):
EulerSaved[k] = [phi[0], theta[0], psi[0]]
else:
EulerSaved[k] = [phi, theta, psi]
t = np.arange(0, Nsamples * dt ,dt)
PhiSaved = EulerSaved[:,0] * 180/pi
ThetaSaved = EulerSaved[:,1] * 180/pi
PsiSaved = EulerSaved[:,2] * 180/pi
plt.figure()
plt.plot(t, PhiSaved)
plt.xlabel('Time [Sec]')
plt.ylabel('Roll angle [deg]')
plt.savefig('12_EulerEKF_roll.png')
plt.figure()
plt.plot(t, ThetaSaved)
plt.xlabel('Time [Sec]')
plt.ylabel('Pitch angle [deg]')
plt.savefig('12_EulerEKF_pitch.png')
plt.show()
'''
plt.subplot(133)
plt.plot(t, PsiSaved)
plt.xlabel('Time [Sec]')
plt.ylabel('Psi angle [deg]')
'''
n =int(T * fs)
# sin wave
sig = np.sin(1.2*2*np.pi*n)# Lets add some noise
noise = 1.5*np.cos(9*2*np.pi*n) + 0.5*np.sin(12.0*2*np.pi*n)
data = sig + noise
print(data)