旋转矩阵求导推导

李群李代数常用结论

1. $exp({\lfloor \delta \mathbf{\theta }\rfloor }_{\times })=\mathbf{I}+{\lfloor \delta \mathbf{\theta }\rfloor }_{\times }$
2. $exp({\lfloor \delta \mathbf{\theta }\rfloor }_{\times })\mathbf{R}=\mathbf{R}exp({\lfloor {\mathbf{R}}^T\delta \mathbf{\theta }\rfloor }_{\times })$
3. $\mathbf{R}exp({\lfloor \delta \mathbf{\theta }\rfloor }_{\times })=exp({\lfloor \mathbf{R}\delta \mathbf{\theta }\rfloor }_{\times })\mathbf{R}$
4. $log(exp({\lfloor \mathbf{\theta }\rfloor }_{\times })exp({\lfloor \delta \mathbf{\theta }\rfloor }_{\times }))={\mathbf{J}}_r^{-1}(\mathbf{\theta })\delta \mathbf{\theta }+\mathbf{\theta }$
5. $log(exp({\lfloor \delta \mathbf{\theta }\rfloor }_{\times })exp({\lfloor \mathbf{\theta }\rfloor }_{\times }))={\mathbf{J}}_l^{-1}(\mathbf{\theta })\delta \mathbf{\theta }+\mathbf{\theta }$
6. ${\mathbf{R}}_1\ominus {\mathbf{R}}_2=log({\mathbf{R}}_2^{-1}{\mathbf{R}}_1)$
7. $\mathbf{R}exp({\lfloor \delta \mathbf{\theta }\rfloor }_{\times }){\mathbf{R}}^{-1}=exp(\mathbf{R}{\lfloor \delta \mathbf{\theta }\rfloor }_{\times }{\mathbf{R}}^{-1})$
8. $\mathbf{R}{\lfloor \delta \mathbf{\theta }\rfloor }_{\times }{\mathbf{R}}^{-1}={\lfloor \mathbf{R}\delta \mathbf{\theta }\rfloor }_{\times }$

$R^{-1}$求导(右乘)

$$\begin{array}{rl}\frac{\partial {\mathbf{R}}^{-1}}{\partial \delta \mathbf{\theta }}& =\frac{(\mathbf{R}\exp ({\lfloor \delta \mathbf{\theta }\rfloor }_{\times }){)}^{-1}\ominus {\mathbf{R}}^{-1}}{\delta \mathbf{\theta }}\stackrel{(6)}{=}\frac{log(\mathbf{R}exp({\lfloor -\delta \mathbf{\theta }\rfloor }_{\times }){\mathbf{R}}^{-1})}{\delta \mathbf{\theta }}\\ & \stackrel{(7)}{=}\frac{log(exp(\mathbf{R}{\lfloor -\delta \mathbf{\theta }\rfloor }_{\times }{\mathbf{R}}^{-1}))}{\delta \mathbf{\theta }}=\frac{log(exp({\lfloor -\mathbf{R}\delta \mathbf{\theta }\rfloor }_{\times }))}{\delta \mathbf{\theta }}\\ & =-\mathbf{R}\end{array}$$

$\mathbf{R}\mathbf{p}$求导(右乘)

$$\begin{array}{rl}\frac{\partial (\mathbf{R}\mathbf{p})}{\partial \delta \mathbf{\theta }}& =\frac{\mathbf{R}exp({\lfloor \delta \mathbf{\theta }\rfloor }_{\times })\mathbf{p}-\mathbf{R}\mathbf{p}}{\delta \mathbf{\theta }}\\ & \stackrel{(1)}{=}\frac{\mathbf{R}{\lfloor \delta \mathbf{\theta }\rfloor }_{\times }\mathbf{p}}{\delta \mathbf{\theta }}=-\mathbf{R}{\lfloor \mathbf{p}\rfloor }_{\times }\end{array}$$

$\mathbf{R}\mathbf{Q}$求导(右乘)

$$\begin{array}{rl}\frac{\partial (\mathbf{R}\mathbf{Q})}{\partial \delta \mathbf{\theta }}& =\frac{\mathbf{R}exp({\lfloor \delta \mathbf{\theta }\rfloor }_{\times })\mathbf{Q}\ominus \mathbf{R}\mathbf{Q}}{\delta \mathbf{\theta }}\\ & \stackrel{(6)}{=}\frac{log({\mathbf{Q}}^{T}exp({\lfloor \delta \mathbf{\theta }\rfloor }_{\times })\mathbf{Q})}{\delta \mathbf{\theta }}\\ & \stackrel{(7)}{=}\frac{log(exp({\mathbf{Q}}^T{\lfloor \delta \mathbf{\theta }\rfloor }_{\times }\mathbf{Q}))}{\delta \mathbf{\theta }}\\ & =\frac{log(exp({\lfloor {\mathbf{Q}}^T\delta \mathbf{\theta }\rfloor }_{\times }))}{\delta \mathbf{\theta }}={\mathbf{Q}}^T\end{array}$$

$log(\mathbf{R})$求导(右乘)

$$\begin{array}{rl}\frac{\partial log(\mathbf{R})}{\partial \delta \mathbf{\theta }}& =\frac{log(\mathbf{R}exp({\lfloor \delta \mathbf{\theta }\rfloor }_{\times }))-log(\mathbf{R})}{\delta \mathbf{\theta }}\\ & \stackrel{(4)}{=}\frac{{\mathbf{J}}_r^{-1}(log\mathbf{R})\delta \mathbf{\theta }+log\mathbf{R}-log\mathbf{R}}{\delta \mathbf{\theta }}\\ & ={\mathbf{J}}_r^{-1}(log\mathbf{R})\end{array}$$

$log(\mathbf{R}\mathbf{Q})$求导(右乘)

$$\begin{array}{rl}\frac{\partial log(\mathbf{R}\mathbf{Q})}{\partial \delta \mathbf{\theta }}& =\frac{log(\mathbf{R}exp({\lfloor \delta \mathbf{\theta }\rfloor }_{\times })\mathbf{Q})-log(\mathbf{R}\mathbf{Q})}{\delta \mathbf{\theta }}\\ & \stackrel{(2)}{=}\frac{log(\mathbf{R}\mathbf{Q}exp({\lfloor {\mathbf{Q}}^T\delta \mathbf{\theta }\rfloor }_{\times }))-log(\mathbf{R}\mathbf{Q})}{\delta \mathbf{\theta }}\\ & \stackrel{(4)}{=}\frac{{\mathbf{J}}_r^{-1}(log(\mathbf{R}\mathbf{Q})){\mathbf{Q}}^T\delta \mathbf{\theta }+log(\mathbf{R}\mathbf{Q})-log(\mathbf{R}\mathbf{Q})}{\delta \mathbf{\theta }}\\ & ={\mathbf{J}}_r^{-1}(log(\mathbf{R}\mathbf{Q})){\mathbf{Q}}^T\end{array}$$

左右雅克比矩阵

关于左乘

对于左乘，只用把扰动量 $exp({\lfloor \delta \mathbf{\theta }\rfloor }_{\times })$ 放在旋转矩阵 $\mathbf{R}$ 的左边进行推导即可

参考

A micro Lie theory for state estimation in robotics