orb_slam3_details/Thirdparty/Sophus/py/sophus/se2.py

225 lines
7.4 KiB
Python

import sympy
import sys
import unittest
import sophus
import functools
class Se2:
""" 2 dimensional group of rigid body transformations """
def __init__(self, so2, t):
""" internally represented by a unit complex number z and a translation
2-vector """
self.so2 = so2
self.t = t
@staticmethod
def exp(v):
""" exponential map """
theta = v[2]
so2 = sophus.So2.exp(theta)
a = so2.z.imag / theta
b = (1 - so2.z.real) / theta
t = sophus.Vector2(a * v[0] - b * v[1],
b * v[0] + a * v[1])
return Se2(so2, t)
def log(self):
theta = self.so2.log()
halftheta = 0.5 * theta
a = -(halftheta * self.so2.z.imag) / (self.so2.z.real - 1)
V_inv = sympy.Matrix([[a, halftheta],
[-halftheta, a]])
upsilon = V_inv * self.t
return sophus.Vector3(upsilon[0], upsilon[1], theta)
def __repr__(self):
return "Se2: [" + repr(self.so2) + " " + repr(self.t)
@staticmethod
def hat(v):
upsilon = sophus.Vector2(v[0], v[1])
theta = v[2]
return sophus.So2.hat(theta).\
row_join(upsilon).\
col_join(sympy.Matrix.zeros(1, 3))
def matrix(self):
""" returns matrix representation """
R = self.so2.matrix()
return (R.row_join(self.t)).col_join(sympy.Matrix(1, 3, [0, 0, 1]))
def __mul__(self, right):
""" left-multiplication
either rotation concatenation or point-transform """
if isinstance(right, sympy.Matrix):
assert right.shape == (2, 1), right.shape
return self.so2 * right + self.t
elif isinstance(right, Se2):
return Se2(self.so2 * right.so2,
self.t + self.so2 * right.t)
assert False, "unsupported type: {0}".format(type(right))
def __getitem__(self, key):
""" We use the following convention [q0, q1, q2, q3, t0, t1, t2] """
assert (key >= 0 and key < 4)
if key < 2:
return self.so2[key]
else:
return self.t[key - 2]
@staticmethod
def calc_Dx_exp_x(x):
return sympy.Matrix(4, 3, lambda r, c:
sympy.diff(Se2.exp(x)[r], x[c]))
@staticmethod
def Dx_exp_x_at_0():
return sympy.Matrix([[0, 0, 0],
[0, 0, 1],
[1, 0, 0],
[0, 1, 0]])
def calc_Dx_this_mul_exp_x_at_0(self, x):
v = Se2.exp(x)
return sympy.Matrix(4, 3, lambda r, c:
sympy.diff((self * Se2.exp(x))[r], x[c])). \
subs(x[0], 0).subs(x[1], 0).limit(x[2], 0)
@staticmethod
def calc_Dx_exp_x_at_0(x):
return Se2.calc_Dx_exp_x(x).subs(x[0], 0).subs(x[1], 0).limit(x[2], 0)
@staticmethod
def Dxi_x_matrix(x, i):
if i < 2:
return sophus.So2.Dxi_x_matrix(x, i).\
row_join(sympy.Matrix.zeros(2, 1)).\
col_join(sympy.Matrix.zeros(1, 3))
M = sympy.Matrix.zeros(3, 3)
M[i - 2, 2] = 1
return M
@staticmethod
def calc_Dxi_x_matrix(x, i):
return sympy.Matrix(3, 3, lambda r, c:
sympy.diff(x.matrix()[r, c], x[i]))
@staticmethod
def Dxi_exp_x_matrix(x, i):
T = Se2.exp(x)
Dx_exp_x = Se2.calc_Dx_exp_x(x)
l = [Dx_exp_x[j, i] * Se2.Dxi_x_matrix(T, j) for j in range(0, 4)]
return functools.reduce((lambda a, b: a + b), l)
@staticmethod
def calc_Dxi_exp_x_matrix(x, i):
return sympy.Matrix(3, 3, lambda r, c:
sympy.diff(Se2.exp(x).matrix()[r, c], x[i]))
@staticmethod
def Dxi_exp_x_matrix_at_0(i):
v = sophus.ZeroVector3()
v[i] = 1
return Se2.hat(v)
@staticmethod
def calc_Dxi_exp_x_matrix_at_0(x, i):
return sympy.Matrix(3, 3, lambda r, c:
sympy.diff(Se2.exp(x).matrix()[r, c], x[i])
).subs(x[0], 0).subs(x[1], 0).limit(x[2], 0)
class TestSe2(unittest.TestCase):
def setUp(self):
upsilon0, upsilon1, theta = sympy.symbols(
'upsilon[0], upsilon[1], theta',
real=True)
x, y = sympy.symbols('c[0] c[1]', real=True)
p0, p1 = sympy.symbols('p0 p1', real=True)
t0, t1 = sympy.symbols('t[0] t[1]', real=True)
self.upsilon_theta = sophus.Vector3(
upsilon0, upsilon1, theta)
self.t = sophus.Vector2(t0, t1)
self.a = Se2(sophus.So2(sophus.Complex(x, y)), self.t)
self.p = sophus.Vector2(p0, p1)
def test_exp_log(self):
for v in [sophus.Vector3(0., 1, 0.5),
sophus.Vector3(0.1, 0.1, 0.1),
sophus.Vector3(0.01, 0.2, 0.03)]:
w = Se2.exp(v).log()
for i in range(0, 3):
self.assertAlmostEqual(v[i], w[i])
def test_matrix(self):
T_foo_bar = Se2.exp(self.upsilon_theta)
Tmat_foo_bar = T_foo_bar.matrix()
point_bar = self.p
p1_foo = T_foo_bar * point_bar
p2_foo = sophus.proj(Tmat_foo_bar * sophus.unproj(point_bar))
self.assertEqual(sympy.simplify(p1_foo - p2_foo),
sophus.ZeroVector2())
def test_derivatives(self):
self.assertEqual(sympy.simplify(
Se2.calc_Dx_exp_x_at_0(self.upsilon_theta) -
Se2.Dx_exp_x_at_0()),
sympy.Matrix.zeros(4, 3))
for i in range(0, 4):
self.assertEqual(sympy.simplify(Se2.calc_Dxi_x_matrix(self.a, i) -
Se2.Dxi_x_matrix(self.a, i)),
sympy.Matrix.zeros(3, 3))
for i in range(0, 3):
self.assertEqual(sympy.simplify(
Se2.Dxi_exp_x_matrix(self.upsilon_theta, i) -
Se2.calc_Dxi_exp_x_matrix(self.upsilon_theta, i)),
sympy.Matrix.zeros(3, 3))
self.assertEqual(sympy.simplify(
Se2.Dxi_exp_x_matrix_at_0(i) -
Se2.calc_Dxi_exp_x_matrix_at_0(self.upsilon_theta, i)),
sympy.Matrix.zeros(3, 3))
def test_codegen(self):
stream = sophus.cse_codegen(Se2.calc_Dx_exp_x(self.upsilon_theta))
filename = "cpp_gencode/Se2_Dx_exp_x.cpp"
# set to true to generate codegen files
if False:
file = open(filename, "w")
for line in stream:
file.write(line)
file.close()
else:
file = open(filename, "r")
file_lines = file.readlines()
for i, line in enumerate(stream):
self.assertEqual(line, file_lines[i])
file.close()
stream.close
stream = sophus.cse_codegen(self.a.calc_Dx_this_mul_exp_x_at_0(
self.upsilon_theta))
filename = "cpp_gencode/Se2_Dx_this_mul_exp_x_at_0.cpp"
# set to true to generate codegen files
if False:
file = open(filename, "w")
for line in stream:
file.write(line)
file.close()
else:
file = open(filename, "r")
file_lines = file.readlines()
for i, line in enumerate(stream):
self.assertEqual(line, file_lines[i])
file.close()
stream.close
if __name__ == '__main__':
unittest.main()