nanomanifold 0.7.0

SO3/SE3 operations on any backend

nanomanifold

Fast, batched and differentiable SO(3)/SE(3) transforms for any backend (NumPy, PyTorch, JAX, ...)

Works directly on arrays, defined as:

• SO(3): unit quat quaternions with explicit convention="wxyz" or convention="xyzw", shape (..., 4)
• SE(3): concatenated [quat, translation], shape (..., 7)

import numpy as np
from nanomanifold import SO3, SE3

# Rotations stored as quaternion arrays, default convention="wxyz"
q = SO3.from_axis_angle(np.array([0, 0, 1]), np.pi/4)  # 45° around Z
points = np.array([[1, 0, 0], [0, 1, 0]])
rotated = SO3.rotate_points(q, points)

# Rigid transforms stored as 7D arrays [quat, translation]
T = SE3.from_rt(q, np.array([1, 0, 0]))  # rotation + translation
transformed = SE3.transform_points(T, points)

Installation

pip install nanomanifold

Quick Start

Rotations (SO3)

from nanomanifold import SO3

# Create rotations
q1 = SO3.from_axis_angle([1, 0, 0], np.pi/2)    # 90° around X
q2 = SO3.from_euler([0, 0, np.pi/4])            # 45° around Z
q3 = SO3.from_rotmat(rotation_matrix)

# Compose and interpolate
q_combined = SO3.multiply(q1, q2)
q_halfway = SO3.slerp(q1, q2, t=0.5)

# Apply to points
points = np.array([[1, 0, 0], [0, 1, 0]])
rotated = SO3.rotate_points(q_combined, points)

Rigid Transforms (SE3)

from nanomanifold import SE3

# Create transforms
T1 = SE3.from_rt(q1, [1, 2, 3])               # rotation + translation
T2 = SE3.from_matrix(transformation_matrix)

# Compose and interpolate
T_combined = SE3.multiply(T1, T2)
T_inverse = SE3.inverse(T_combined)
T_halfway = SE3.slerp(T1, T2, t=0.5)

# Apply to points
transformed = SE3.transform_points(T_combined, points)

API Reference

All functions are available via nanomanifold.SO3 and nanomanifold.SE3. Shapes follow the Array API convention and accept arbitrarily batched inputs.

SO3 (3D Rotations)

Supported SO3 parametrizations:

Name	Shape	Notes
quat	(...,4)	Unit quaternion with explicit convention="wxyz" or convention="xyzw"
axis_angle	(...,3)	Rotation vector / axis-angle parametrization
euler	(...,3)	Euler angles with an explicit convention such as "ZYX"
rotmat	(...,3,3)	Normalized rotation matrix in SO(3)
matrix	(...,3,3)	Generic 9D matrix, not assumed to be normalized
sixd	(...,6)	6D continuous representation built from the first two rotation-matrix columns

Function	Signature
canonicalize(q, convention="wxyz")	(...,4) -> (...,4)
to_axis_angle(q, convention="wxyz")	(...,4) -> (...,3)
from_axis_angle(axis_angle, convention="wxyz")	(...,3) -> (...,4)
from_hinge(angles, axes, convention="wxyz")	(...,1), (...,3) -> (...,4)
to_hinge(q, axes, convention="wxyz")	(...,4), (...,3) -> (...,1)
to_euler(q, convention="ZYX", quat_convention="wxyz")	(...,4) -> (...,3)
from_euler(euler, convention="ZYX", quat_convention="wxyz")	(...,3) -> (...,4)
convert(x, src=..., dst=...)	dynamic
identity_as(ref, batch_dims=..., rotation_type=..., convention="wxyz")	dynamic
to_rotmat(q, convention="wxyz")	(...,4) -> (...,3,3)
from_rotmat(R, convention="wxyz")	(...,3,3) -> (...,4)
from_matrix(R, convention="wxyz", mode="svd")	(...,3,3) -> (...,4)
from_quat(quat, convention="xyzw")	(...,4) -> (...,4)
to_quat(quat, convention="xyzw")	(...,4) -> (...,4)
to_sixd(q, convention="wxyz")	(...,4) -> (...,6)
from_sixd(sixd, convention="wxyz")	(...,6) -> (...,4)
multiply(q1, q2, convention="wxyz")	(...,4), (...,4) -> (...,4)
inverse(q, convention="wxyz")	(...,4) -> (...,4)
rotate_points(q, points, convention="wxyz")	(...,4), (...,N,3) -> (...,N,3)
slerp(q1, q2, t, convention="wxyz")	(...,4), (...,4), (...,N) -> (...,N,4)
distance(q1, q2, rotation_type="quat", convention="wxyz")	dynamic
log(q, convention="wxyz")	(...,4) -> (...,3)
exp(tangent, convention="wxyz")	(...,3) -> (...,4)
hat(w)	(...,3) -> (...,3,3)
vee(W)	(...,3,3) -> (...,3)
weighted_mean(quats, weights, convention="wxyz")	sequence of (...,4), (...,N) -> (...,4)
mean(quats, convention="wxyz")	sequence of (...,4) -> (...,4)
random(*shape, convention="wxyz")	(...,4)

SE3 (Rigid Transforms)

Function	Signature
canonicalize(se3, convention="wxyz")	(...,7) -> (...,7)
from_rt(quat, translation, convention="wxyz")	(...,4), (...,3) -> (...,7)
to_rt(se3, convention="wxyz")	(...,7) -> (quat, translation)
from_matrix(T)	(...,4,4) -> (...,7)
to_matrix(se3, convention="wxyz")	(...,7) -> (...,4,4)
multiply(se3_1, se3_2, convention="wxyz")	(...,7), (...,7) -> (...,7)
inverse(se3)	(...,7) -> (...,7)
transform_points(se3, points)	(...,7), (...,N,3) -> (...,N,3)
slerp(se3_1, se3_2, t)	(...,7), (...,7), (...,N) -> (...,N,7)
log(se3)	(...,7) -> (...,6)
exp(tangent)	(...,6) -> (...,7)
hat(v)	(...,6) -> (...,4,4)
vee(M)	(...,4,4) -> (...,6)
weighted_mean(transforms, weights)	sequence of (...,7), (...,N) -> (...,7)
mean(transforms)	sequence of (...,7) -> (...,7)
random(*shape)	(...,7)

Pairwise Conversions (SO3.conversions)

Convert directly between any two rotation representations without going through quaternions manually. All 30 pairwise functions follow the naming pattern from_{source}_to_{target}.

Representations: axis_angle, euler, matrix, rotmat, quat, sixd. Quaternion conventions use lowercase convention="wxyz" or convention="xyzw".

Function	Signature
SO3.conversions.from_axis_angle_to_rotmat(aa)	(...,3) -> (...,3,3)
SO3.conversions.from_axis_angle_to_euler(aa, convention)	(...,3) -> (...,3)
SO3.conversions.from_axis_angle_to_quat(aa, convention="wxyz")	(...,3) -> (...,4)
SO3.conversions.from_axis_angle_to_sixd(aa)	(...,3) -> (...,6)
SO3.conversions.from_euler_to_axis_angle(e, convention)	(...,3) -> (...,3)
SO3.conversions.from_euler_to_rotmat(e, convention)	(...,3) -> (...,3,3)
SO3.conversions.from_euler_to_quat(e, src_convention="ZYX", dst_convention="wxyz")	(...,3) -> (...,4)
SO3.conversions.from_euler_to_sixd(e, convention)	(...,3) -> (...,6)
SO3.conversions.from_rotmat_to_axis_angle(R)	(...,3,3) -> (...,3)
SO3.conversions.from_rotmat_to_euler(R, convention)	(...,3,3) -> (...,3)
SO3.conversions.from_rotmat_to_quat(R, convention="wxyz")	(...,3,3) -> (...,4)
SO3.conversions.from_rotmat_to_sixd(R)	(...,3,3) -> (...,6)
SO3.conversions.from_matrix_to_rotmat(M, mode="svd")	(...,3,3) -> (...,3,3)
SO3.conversions.from_matrix_to_axis_angle(M, mode="svd")	(...,3,3) -> (...,3)
SO3.conversions.from_matrix_to_euler(R, convention, mode="svd")	(...,3,3) -> (...,3)
SO3.conversions.from_matrix_to_quat(R, convention="wxyz", mode="svd")	(...,3,3) -> (...,4)
SO3.conversions.from_matrix_to_sixd(R, mode="svd")	(...,3,3) -> (...,6)
SO3.conversions.from_quat_to_axis_angle(q, convention="wxyz")	(...,4) -> (...,3)
SO3.conversions.from_quat_to_euler(q, src_convention="wxyz", dst_convention="ZYX")	(...,4) -> (...,3)
SO3.conversions.from_quat_to_rotmat(q, convention="wxyz")	(...,4) -> (...,3,3)
SO3.conversions.from_quat_to_quat(q, src_convention="wxyz", dst_convention="xyzw")	(...,4) -> (...,4)
SO3.conversions.from_quat_to_sixd(q, convention="wxyz")	(...,4) -> (...,6)
SO3.conversions.from_sixd_to_axis_angle(sixd)	(...,6) -> (...,3)
SO3.conversions.from_sixd_to_euler(sixd, convention)	(...,6) -> (...,3)
SO3.conversions.from_sixd_to_rotmat(sixd)	(...,6) -> (...,3,3)
SO3.conversions.from_sixd_to_quat(sixd, convention="wxyz")	(...,6) -> (...,4)

For runtime-selected conversions, use SO3.convert. src="matrix" treats the input as a generic 3x3 matrix and projects it to rotmat before converting. Euler uses the usual axis-order convention strings:

matrix = SO3.convert(axis_angle, src="axis_angle", dst="matrix")
rotmat = SO3.convert(matrix, src="matrix", dst="rotmat")
quat_alt = SO3.convert(euler, src="euler", dst="quat", src_convention="XYZ", dst_convention="xyzw")
quat = SO3.convert(quat_alt, src="quat", dst="quat", src_convention="xyzw", dst_convention="wxyz")
euler = SO3.convert(rotmat, src="rotmat", dst="euler", dst_convention="ZYX")

For constrained one-axis rotations, use SO3.from_hinge and SO3.to_hinge. The hinge helpers map (...,1) scalar coefficients to axis-angle vectors with angles * axes. to_hinge projects the principal axis-angle vector back onto axes; for unit axes this is the signed principal rotation angle.

Backend-Explicit Mode

By default, nanomanifold auto-detects the array backend from the input array type. Every function also accepts an optional xp keyword argument to specify the backend explicitly. This is required for torch.compile(fullgraph=True), since Dynamo cannot trace the dynamic dispatch:

import torch
from nanomanifold import SO3, SE3

@torch.compile(fullgraph=True)
def forward(q1, q2, T1, T2):
    q_mid = SO3.slerp(q1, q2, torch.tensor([0.5]), xp=torch)
    T_mid = SE3.slerp(T1, T2, torch.tensor([0.5]), xp=torch)
    return q_mid, T_mid