def update_function(command, objects, noise):
return [Angle3D.fromRotationAxis(pi/4+objects[0].getRotation()+noise,objects[0].getAxis())]
covars = [pi/6*matrix(identity(3))]
def update_function(command, objects, noise):
return [Angle3D.fromRotationAxis(pi/4+objects[0].getRotation()+noise,objects[0].getAxis())]
def measurement_predict_function(objects, noise):
pdb.set_trace()
if magnitude(noise[1:]) == 0:
axis = noise[1:]
else:
axis = noise[1:]/magnitude(noise[1:])
theta = noise[0,0]
return [Angle3D(Quaternion(sin(theta/2)*axis[0,0],sin(theta/2)*axis[1,0],sin(theta/2)*axis[2,0],cos(theta/2))*objects[0].r)]
covar_process_noise = matrix([[.001]])
covar_measurement_noise = .005*matrix(identity(4))
k = unscentedKalmanFilter(objects, covars, update_function, measurement_predict_function, [None], covar_process_noise, covar_measurement_noise)
realPos = [Angle3D.fromRotationAxis(pi/2,matrix([[1/sqrt(3)],[1/sqrt(3)],[1/sqrt(3)]]))]
for i in xrange(100):
realPos = update_function(0,realPos,random.normal(0,covar_process_noise[0,0]))
measurement = measurement_predict_function(realPos,matrix(random.multivariate_normal([0,0,0,0],covar_measurement_noise)).getT())
print measurement[0].r
k.step(measurement, [0])
ret = k.mean
print "filter, real"
print ret[0].getRotation(), realPos[0].getRotation()
print ret[0].getAxis()
print realPos[0].getAxis()
pdb.set_trace()