当前位置:网站首页>Visual slam Lecture 3 -- Lie groups and Lie Algebras
Visual slam Lecture 3 -- Lie groups and Lie Algebras
2022-07-02 03:39:00 【Reading fitness code】
0. Content

1. What is a group
Why introduce groups ?
Because when calculating the derivative of rotation matrix or transformation matrix , Addition no longer holds , So finding the derivative of a matrix requires introducing other methods , So the group is introduced .
“ Group , Ring , Domain ” They are all things in abstract algebra 

Because addition is a commutative group on integers , therefore 2+3=3+2=5
for example : Operations on addition , Univariate is 0,a+(-a)=0
In multiplication , Univariate is 1, a ∗ 1 a = 1 a* \dfrac{1}{a}=1 a∗a1=1

The rotation matrix is reversible , Is an orthogonal matrix , And matrix multiplication to form a group .
If not in groups , Basically, there is no need to discuss .
General n × n n\times n n×n The invertible matrix of is a general linear group GL(n)

Basic knowledge supplement :《 Abstract algebra 》,《 Differential geometry 》
2. Lie groups and Lie algebras

With continuous ( smooth ) A group of properties is called a lie group (Lie Group).
Basic knowledge supplement :《 Differential geometry 》
Definition of manifold ( I understand it ): A manifold is a space . Manifold is a space formed by mapping low dimensional data to high dimensions . Data in high-dimensional space will produce dimensional redundancy , But in fact, these data can be uniquely represented only by using lower dimensions . And the manifold learning I've heard before is : Suppose the data is a low dimensional manifold uniformly sampled in a high-dimensional Euclidean space , Manifold learning is to recover low dimensional manifold structure from high dimensional sampled data , And find low dimensional manifolds in high dimensional space , And find the corresponding embedding mapping , To achieve dimension reduction or data visualization . He is looking for the essence of things from the observed phenomena , The internal law of data generation has been found .
The following pictures are from Reference blog

SO(3) and SE(3) Only well-defined multiplication , No addition , So it is difficult to reach the limit of retrograde , Derivation and other operations .( It was said before that the group was introduced in order to find the derivative of the rotation matrix ? Why can't we ask here ?)

First of all SO(3)( Rotation matrix ) To introduce Lie algebra .
First, for the rotation matrix , It is orthogonal. , Derivation of time on both sides , After finishing, we get an antisymmetric matrix
R ˙ ( t ) R ( t ) T \dot{R}(t)R(t)^T R˙(t)R(t)T
a^ Represents from vector to antisymmetric matrix , and 
Denotes that the vector is represented by an antisymmetric matrix .

Can be regarded as right R(t) After derivation , One more on the left ϕ ( t ) \phi(t) ϕ(t)^

Yes R(t) First order Taylor expansion , stay t 0 t_0 t0 near ϕ \phi ϕ unchanged , Into differential equations , Bring in the initial condition solution ( There are some details that have not been discussed carefully )
Any given vector ϕ \phi ϕ, Can find a corresponding relationship ,
e ϕ ^ 0 t = R ( t ) e^{\hat\phi_0 t}=R(t) eϕ^0t=R(t)
among ϕ \phi ϕ It's lie algebra , The correspondence is exponential mapping , R R R It's Li Qun .
The following is a strict definition of Lie algebra :
Given the structure, we can check whether it is a lie algebra .
[,] It's a lie bracket operation .
SO(3) It's a matrix ,so(3) It's a vector ( In fact, matrix and vector correspond one by one , Fine , But this is more natural ), Lie brackets are operations that act on vectors .

SE(3) The lie algebra of is 6 Dimension vector , ρ \rho ρ Represents translation , ϕ \phi ϕ For rotation , And the top tip ^ No longer represent a symmetric matrix , But still retain the notation . Just put the rotating part ϕ \phi ϕ Do antisymmetry , The translation part remains unchanged , from (6,1)->(4,4)

3. Exponential mapping and logarithmic mapping
Taylor expansion 

Expand by infinite series ( See high number 18 speak P244)
e x = ∑ n = 0 ∞ x n n ! e^x=\sum_{n=0}^\infty \frac{x^n}{n!} ex=n=0∑∞n!xn
sin x = ∑ n = 0 ∞ ( − 1 ) n x 2 n + 1 ( 2 n + 1 ) ! \sin x=\sum_{n=0}^\infty (-1)^n\frac{x^{2n+1}}{(2n+1)!} sinx=n=0∑∞(−1)n(2n+1)!x2n+1
cos x = ∑ n = 0 ∞ ( − 1 ) n x 2 n ( 2 n ) ! \cos x=\sum_{n=0}^\infty (-1)^n\frac{x^{2n}}{(2n)!} cosx=n=0∑∞(−1)n(2n)!x2n
among 3 The expansions converge to − ∞ < x < + ∞ -\infty<x<+\infty −∞<x<+∞
Add the following properties to deal with higher powers 
You can deduce Rodriguez formula .
Then explain Lie algebra so(3) The physical meaning of is actually rotation vector ( Rodrigues formula is rotation vector to rotation matrix , On the left is the rotation matrix R R R). conversely , Given the rotation matrix, the rotation vector can be obtained , You can use logarithmic mapping or the following transformation relationship :
about SE(3)
Where the translation vector ρ \rho ρ And the translation in the transformation matrix t t t A coefficient matrix is missing , That is Jacobian matrix , It's related to the angle , The rotating part is just a so(3).
This part summarizes

Li Qun –> Exponential mapping –> Lie algebra
Lie algebra –> Logarithmic mapping –> Li Qun
4. Derivation and perturbation model

Or derivation , Because lie groups have no addition , Therefore, the derivation is required to start from the perspective of Lie algebra , Add a small quantity to lie algebra δ ϕ \delta \phi δϕ Then it is transformed into Lie group by exponential operation , But the question is whether addition in Lie algebra is equivalent to multiplication of Lie groups . It is obviously true in scalar ,a,b It is obviously true when it is scalar e a ∗ e b = e a + b e^a*e^b=e^{a+b} ea∗eb=ea+b, But it doesn't hold when the power is a matrix .
introduce BCH The formula :
Matrix two exp And then take ln Expansion is a formula about lie brackets , Quite complicated , After some approximation :
When one of the two matrices is a small quantity , Gu slightly makes the following approximation in higher order :
When multiplying left by a small quantity , The coefficient is the left Jacobian matrix ; When multiplying left by right by a small amount , The coefficient is the right Jacobian matrix , It is often compared with Zuo Ya .
So when adding small quantities on Lie algebra , Equivalent to the upper left of the Group ( Right ) Multiply by a quantity with Jacobi .
SE(3) Than SO(3) complex , In the top right corner of the Q r Q_r Qr Very complicated .

There are two ways to find derivatives :
- Add a small quantity to lie algebra , Find the rate of change
- Liqun is superior to a small quantity , Find the rate of change relative to lie algebra
Derivative model :
Reverse use BCH The formula , Then Taylor expands , The quadratic term is 0 了 , So Taylor expansion only retains the first two .
Disturbance model , Left ( Right ) Multiply the group corresponding to a disturbance 
SE(3) Perturbation model : Left multiply perturbation 
5. Sophus practice
边栏推荐
- Oracle viewing locked tables and unlocking
- [mv-3d] - multi view 3D target detection network
- Object oriented thinking
- Sentry experience and architecture, a fledgling monitoring product with a market value of $100million
- Introduction to Robotics II. Forward kinematics, MDH method
- [untitled] basic operation of raspberry pie (2)
- MySQL advanced (Advanced) SQL statement (II)
- 蓝桥杯单片机省赛第十二届第二场
- Global and Chinese markets for electronic laryngoscope systems 2022-2028: Research Report on technology, participants, trends, market size and share
- Global and Chinese markets for ultrasonic probe disinfection systems 2022-2028: Research Report on technology, participants, trends, market size and share
猜你喜欢

蓝桥杯单片机省赛第十一届第二场

蓝桥杯单片机省赛第八届

Kubernetes cluster storageclass persistent storage resource core concept and use

Eight steps of agile development process

Review materials of project management PMP high frequency examination sites (8-1)

Basic syntax of unity script (6) - specific folder
![[yolo3d]: real time detection of end-to-end 3D point cloud input](/img/5e/f17960d302f663db75ad82ae0fd70f.jpg)
[yolo3d]: real time detection of end-to-end 3D point cloud input

The 10th Blue Bridge Cup single chip microcomputer provincial competition

The first game of the 12th Blue Bridge Cup single chip microcomputer provincial competition

Pointer array & array pointer
随机推荐
Analyse de 43 cas de réseaux neuronaux MATLAB: Chapitre 42 opérations parallèles et réseaux neuronaux - - opérations parallèles de réseaux neuronaux basées sur CPU / GPU
In depth analysis of C language - variable error prone knowledge points # dry goods inventory #
Generate random numbers that obey normal distribution
Pycharm2021 delete the package warehouse list you added
Review materials of project management PMP high frequency examination sites (8-1)
MySQL advanced (Advanced) SQL statement (II)
JIT deep analysis
Basic syntax of unity script (6) - specific folder
Global and Chinese market of autotransfusion bags 2022-2028: Research Report on technology, participants, trends, market size and share
蓝桥杯单片机第六届温度记录器
Unity脚本的基础语法(6)-特定文件夹
In depth interpretation of pytest official documents (26) customized pytest assertion error information
Kotlin basic learning 16
Introduction to Robotics II. Forward kinematics, MDH method
[database]jdbc
What do you know about stock selling skills and principles
跳出舒适区,5年点工转型自动化测试工程师,我只用了3个月时间
Discrimination between sap Hana, s/4hana and SAP BTP
[数据库]JDBC
[mv-3d] - multi view 3D target detection network