so3（苏 3）

Pywayne VIO SO3

Overview

Complete SO(3) rotation matrix toolkit for 3D rotations with Lie group/ Lie algebra operations, rotation representation conversions, skew-symmetric matrix operations, and rotation averaging.

Quick Start

from pywayne.vio.SO3 import SO3_skew, SO3_Exp, SO3_Log, SO3_to_quat
import numpy as np

# Skew-symmetric matrix
vec = np.array([1, 2, 3])
skew = SO3_skew(vec)  # Returns 3x3 skew-symmetric matrix

# Log/Exp mapping
R = np.eye(3)
rotvec = SO3_Log(R)  # Rotation vector (Lie algebra)
R_recon = SO3_Exp(rotvec)  # Back to rotation matrix

# Quaternion conversion
quat = SO3_to_quat(R)  # Returns [w, x, y, z]

Core Functions

Basic Operations

check_SO3(R)

Check if matrix is a valid SO(3) rotation matrix.

• Validates shape (3, 3)
• Checks R.T @ R = I (orthogonality)

SO3_mul(R1, R2)

Multiply two rotation matrices: R1 @ R2.

SO3_diff(R1, R2, from_1_to_2=True)

Compute relative rotation between two matrices.

• from_1_to_2=True: Returns R1.T @ R2
• from_1_to_2=False: Returns R2.T @ R1

SO3_inv(R)

Compute inverse of rotation matrix (transpose).

• Supports single (3, 3) or batch (N, 3, 3) inputs

Skew-Symmetric Matrices

SO3_skew(vec)

Convert 3D vector to skew-symmetric matrix.

vec = [x, y, z] -> [[ 0, -z,  y],
                    [ z,  0, -x],
                    [-y,  x,  0]]

• Supports single vector (3,) or batch (N, 3)

SO3_unskew(skew)

Extract vector from skew-symmetric matrix.

• Single matrix (3, 3) -> vector (3,)
• Batch (N, 3, 3) -> vectors (N, 3)

Rotation Representation Conversions

Quaternion

• SO3_from_quat(q) - Quaternion [w, x, y, z] to rotation matrix
• SO3_to_quat(R) - Rotation matrix to quaternion [w, x, y, z]
• Uses Hamilton convention (wxyz)

Axis-Angle

• SO3_from_axis_angle(axis, angle) - Axis-angle to rotation matrix
• SO3_to_axis_angle(R) - Returns (axis, angle) tuple

Euler Angles

• SO3_from_euler(euler_angles, axes='zyx', intrinsic=True) - Euler to matrix
• SO3_to_euler(R, axes='zyx', intrinsic=True) - Matrix to Euler
• Supports all rotation sequences

Lie Group/ Lie Algebra Mapping

SO3_Log(R)

SO(3) to so(3) log map, returns rotation vector (3D).

• Input: (3, 3) or (N, 3, 3)
• Output: (3,) or (N, 3)

SO3_log(R)

SO(3) to so(3) log map, returns skew-symmetric matrix (3x3).

• Equivalent to SO3_skew(SO3_Log(R))

SO3_Exp(rotvec)

so(3) to SO(3) exp map from rotation vector.

• Handles zero vectors gracefully
• Input: (3,) or (N, 3)
• Output: (3, 3) or (N, 3, 3)

SO3_exp(omega_hat)

so(3) to SO(3) exp map from skew-symmetric matrix.

• Equivalent to SO3_Exp(SO3_unskew(omega_hat))

Averaging

SO3_mean(R)

Compute mean rotation matrix from multiple rotations.

• Uses scipy Rotation.mean()
• Input: (N, 3, 3)
• Output: (3, 3)

Data Formats

Single vs Batch

• Single matrix: shape (3, 3)
• Batch: shape (N, 3, 3)

Most functions handle both automatically.

SO(3) Matrix Properties

R @ R.T = I  (orthogonal)
det(R) = 1   (special)

Lie Algebra Vector

Rotation vector where direction is axis, magnitude is angle.

Dependencies

Required packages:

• numpy - Array operations
• qmt - Quaternion utilities
• scipy - Rotation averaging

Install with:

pip install numpy qmt scipy

Example Usage

# Create rotation from axis-angle
axis = np.array([0, 0, 1])  # Z-axis
angle = np.pi / 4  # 45 degrees
R = SO3_from_axis_angle(axis, angle)

# Verify it's valid
print(check_SO3(R))  # True

# Get Lie algebra representation
rotvec = SO3_Log(R)
print(f"Rotation vector: {rotvec}")

# Convert back
R_recon = SO3_Exp(rotvec)
print(f"Reconstruction error: {np.linalg.norm(R - R_recon):.2e}")

# Batch averaging
R_batch = np.array([R, SO3_inv(R), SO3_mul(R, R)])
R_mean = SO3_mean(R_batch)