李群李代数（二）

Zhihao Zhan

于 2023-07-13 17:06:53 发布

阅读量83

点赞数

分类专栏： SLAM 文章标签：概率论线性代数

本文链接：https://blog.csdn.net/qq_39683777/article/details/131706939

版权

SLAM 专栏收录该内容

5 篇文章 1 订阅

订阅专栏

文章目录

1 指数和对数映射

1 指数和对数映射

在这里插入图片描述

对李代数拆分， $\ \phi= \theta a$ ，模长 $\ \theta$ ，方向为单位向量 $\ a, \vert\vert{a}\vert\vert=1$ ，有以下性质：
$a^{\land}a^{\land} = aa^T-I \\ a^{\land}a^{\land}a^{\land}=a^{\land}(aa^T-I)=-a^{\land}$
对于 $\ so(3)$ 的任意向量 $\ \phi$ ，可以定义指数映射，以及对应的泰勒展开：
$1-\frac{x^2}{2!}+\frac{x^4}{4!}-\frac{x^6}{6!}+ \cdots \\ sinx = x-\frac{x^3}{3!}+\frac{x^5}{5!}-\frac{x^7}{7!}+ \cdots \\ R = e^{\phi^{\land}}= \sum_{n=0}{\frac{1}{n!}(\phi^{\land})^n} = \sum_{n=0}{\frac{1}{n!}(\theta a^{\land})^n} \\ = I + \theta a^{\land} + \frac{1}{2!}\theta^2a^{\land}a^{\land} + \frac{1}{3!}\theta^3a^{\land}a^{\land}a^{\land} + \frac{1}{4!}\theta^4a^{\land}a^{\land}a^{\land}a^{\land} + \cdots \\ = aa^T - a^{\land}a^{\land} + \theta a^{\land} + \frac{1}{2!}\theta^2a^{\land}a^{\land} -\frac{1}{3!}\theta^3a^{\land} - \frac{1}{4!}\theta^4a^{\land}a^{\land} + \cdots \\ = aa^T + (\theta - \frac{1}{3!}\theta^3 + \frac{1}{5!}\theta^5 - \cdots)a^{\land} - (1-\frac{1}{2!}\theta^2+\frac{1}{4!}\theta^4-\cdots)a^{\land}a^{\land} \\ = a^{\land}a^{\land} + I + sin\theta a^{\land} - cos\theta a^{\land}a^{\land} \\ = (1-cos\theta)a^{\land}a^{\land} +I +sin\theta a^{\land} \\ = cos\theta I + (1-cos\theta)aa^T+sin\theta a^{\land}$
可见 $\ so(3)$ 就是旋转向量组成的空间，而指数映射就是罗德里格斯公式。而对数映射可以将李群转为李代数 $\ \phi=\theta a=ln(R)^{\lor}$ ，但计算较为复杂，可使用以下公式进行计算角度，根据转轴不变求解方向向量：

$cos\theta I + (1-cos\theta)aa^T+sin\theta a^{\land} \\ tr(R) = cos\theta tr(I) + (1-cos\theta)tr(aa^T)+sin\theta tr(a^{\land}) \\ = 3cos\theta+(1-cos\theta) = 1+2cos\theta\\ \theta = arccos \frac{tr(R)-1}{2} \\ Ra = a$
对于 $\ se(3)$ 进行指数映射，推导如下：
$e^{\epsilon^{\land}} = \sum_{n=0}{\frac{1}{n!}(\epsilon^{\land})^n} \\ = I + \sum_{n=1}{\frac{1}{n!}(\begin{bmatrix} \phi^{\land} & \rho \\ 0^T & 0 \end{bmatrix})^n} \\$
对矩阵乘法结果进行归纳后得到：
$\begin{bmatrix} \theta a^{\land} & \rho \\ 0^T & 0 \end{bmatrix} \begin{bmatrix} \theta a^{\land} & \rho \\ 0^T & 0 \end{bmatrix} = \begin{bmatrix} (\theta a^{\land})^2 & \theta a^{\land}\rho \\ 0^T & 0 \end{bmatrix} \\$
最终指数映射结果如下：
$\sum_{n=1}{\frac{1}{n!}\begin{bmatrix} (\theta a^{\land})^n & (\theta a^{\land})^{n-1}\rho \\ 0^T & 0 \end{bmatrix}} \\ = \begin{bmatrix} \sum_{n=0}{\frac{1}{n!}(\phi^{\land})^n} & \sum_{n=1}{\frac{1}{n!}(\phi^{\land})^{n-1}}\rho \\ 0^T & 1 \end{bmatrix} \\ = \begin{bmatrix} \sum{\frac{1}{n!}(\phi^{\land})^n} & \sum{\frac{1}{(n+1)!}(\phi^{\land})^n}\rho \\ 0^T & 1 \end{bmatrix} = \begin{bmatrix} R & J\rho \\ 0^T & 1 \end{bmatrix}$
和 $\ so(3)$ 一样将R转为 $\ \phi=\theta a$ 的推导与so(3)映射一致，
$\sum_{n=0}{\frac{1}{(n+1)!}(\phi^{\land})^n} = \sum_{n=0}{\frac{1}{(n+1)!}(\theta a^{\land})^n} \\ = I + \frac{1}{2!}\theta a^{\land} + \frac{1}{3!}\theta^2a^{\land}a^{\land} + \frac{1}{4!}\theta^3a^{\land}a^{\land}a^{\land} + \frac{1}{5!}\theta^4a^{\land}a^{\land}a^{\land}a^{\land} + \cdots \\ = \frac{1}{\theta}(\frac{1}{2!}\theta^2-\frac{1}{4!}\theta^4+\cdots)a^{\land}+\frac{1}{\theta}(\frac{1}{3!}\theta^3-\frac{1}{5!}\theta^5+\cdots)a^{\land}a^{\land}+I \\ = \frac{1}{\theta}(1-cos\theta)a^{\land}+\frac{1}{\theta}(\theta-sin\theta)a^{\land}a^{\land}+I \\ = \frac{1-cos\theta}{\theta}a^{\land}+\frac{1}{\theta}(\theta-sin\theta)(aa^T - I)+I \\ = \frac{1-cos\theta}{\theta}a^{\land} + (1-\frac{sin\theta}{\theta})(aa^T - I) + I \\ = \frac{1-cos\theta}{\theta}a^{\land} + (1-\frac{sin\theta}{\theta})aa^T + \frac{sin\theta}{\theta}I \\ t = J\rho$
同样对数映射较为复杂，和 $\ SO(3)$ 一样的方式根据R求对应的旋转向量 $\ \phi=\theta a$ ，同时根据 $\ t=J\rho$ 求线性方程得到 $\ \rho$ 。