More On Transformations

Transformations of a frame (object, or point) which are

relative to the fixed reference frame
Pre-multiply the transformation matrix with the coordinates
described in the moving frame:
e.g. for a rotation about the z axis followed by a translation about
the x axis, followed by a rotation about the y axis:
Transformxyz = Ry · Tx · Rz
More On Transformations (cont’d)
Transformations of a frame (object, or point) which are
relative to the moving reference frame
Post-multiply the transformation matrix with the coordinates
described in the moving (current) frame:
Forward Kinematics vs Inverse Kinematics
Forward Kinematics: the calculation of the position and
orientation of a robot’s end-effector from its joint coordinates
Inverse Kinematics: given a position and orientation of a
robot’s end-effector, calculate the angles θ of the joints.
Forward Kinematics
Forward Kinematics
Representation of Configuration Space of a Robot
The position and orientation of all links.
The pose of the end-effector (i.e. location and orientation) can be
described with basic transformation matrices that can be
multiplied together to get the homogeneous matrix.
HomogeneousMatrix = Transf1 ∗ Transf 2 ∗ Transf 3 ∗ . . . ∗ Transfn
where n is the number of links (assuming that each of these
matrices is the total transformation for each link).
Example of a 1-joint Robot Arm
Rotation by angle q1 and then translation by a1.
The homogeneous transformation describing the overall result can
be calculated using the following:
EndEffector = Rot(q1) · Tx (a1) (4)
Pose of the End-Effector - 1-Joint 2D Robot Arm (cont’d)
In Python Robotics Toolbox:
>>> q1 = Symbol('q1')
>>> trot2(q1)
>>> a1=Symbol('a1')
>>> transl2(a1,0)
>>> E = trot2(q1) @ transl2(a1, 0)
or equivalently as a ETS2 object:
>>> e = ET2.R()*ET2.tx(a1)
>> e.plot(0)
>> e.plot(math.pi/4)
The homogeneous transformation describing the overall result can
be calculated using the following:
EndEffector = Rot(q1) · Tx (a1) · Rot(q2) · Tx (a2) (5)
Pose of the End-Effector - 2-Joint 2D Robot Arm (cont’d)
In Python Robotics Toolbox:
E = simplify(E)
Example of a 3-joint Planar Robot Arm
The homogeneous transformation describing the overall result can
be calculated using the following:
EndEffector = Rot(q1)· Tx (a1)· Rot(q2)· Tx (a2)· Rot(q3)· Tx (a3)
Pose of the End-Effector - 3-Joint 2D Robot Arm (cont’d)
In Python Robotics Toolbox:
The Problem of Forward Kinematics
The calculation of the position and orientation of a robot’s
end-effector from its joint coordinates θi
The Denavit-Hartenberg (DH) Notation
The relationship between two coordinate frames is described by 6
parameters (3 translations and 3 rotations). Can this be improved?
Attach a coordinate frame to the end of each link.
Reduces the relationship between 2 coordinate frames from 6
parameters to 4 parameters.
Each joint in a robot is described by 4 parameters.
How is this achieved?
The coordinate frames have constraints.
x axis of frame j intersects the z axis of frame j − 1
x axis of frame j is perpendicular to the z axis of frame j − 1
−→ 6 parameters - 2 constraints means 4 parameters are needed.
The Denavit-Hartenberg (DH) Notation (cont’d)
4 parameters used associated with each link i and joint i:
θi: joint angle
di: link offset
ri: link length
αi: link twist
Each homogeneous transformation Ai
is represented as the product
of 4 basic transformations:
The DH Table
The DH notation requires a table. The number of rows equals the
number of joints and it has 4 columns each one corresponding to
the 4 parameters for the joint i of that row.
For revolute joints: Only θ changes, all the other 3
parameters are fixed according to the robot mechanism.
For prismatic joints: Only d changes, all the other 3
parameters are fixed according to the robot mechanism.
Example of DH Notation
Finding the Pose of the End-Effector relative to the Base
Frame
Assume that A1, A2, A3 . . . An are the DH matrices of all the robot
joints 1, 2, 3, . . . n.
Then the calculation requires the multiplication of all the matrices:
Poseend effector = A1 · A2 · A3 . . . An (8)
Example: Calculation of the Pose of the End-Effector
The following DH matrices correspond to the joints of a robot,
from robot base to end-effector. Find the pose of the end-effector
relative to the robot base.
Example: Calculation of the Pose of the End-Effector
(cont’d)
Simply calculate A1 ∗ A2 ∗ A3.
The problem of Forward kinematics
Kinematic chain: a series of rigid bodies (e.g. links of a robotic
arm) connected together by joints.
The joint angles of a kinematic chain determine the position and
orientation of the end effector.
A coordinate frame i relative to coordinate frame i − 1 is
denoted by i−1Ti
The Problem of Forward Kinematics (cont’d)
The problem of forward kinematics is expressed as the calculation
of the transformation between a coordinate frame fixed in the
end-effector and another coordinate frame fixed in the base.
For example, for 6-joint manipulator:
The Problem of Inverse Kinematics
Given a position and orientation of a robot’s end-effector, calculate
the angles θ of the joints.
Solving the kinematics equations of a manipulator robot is a
nonlinear problem.
Given the homogeneous matrix of the end-effector with
respect to the base frame, solve for all the joint angles
θ1, θ2, . . . , θn.
Challenging mathematical problem due to:
1. nature of the nonlinear equations. Often no analytic solutions
(closed form) can be calculated and numerical methods are
required.
2. often, there are multiple solutions (i.e. multiple sets of joint
angles) that can place the end effector at the desired position.
−→ The algorithm must choose the solution that results in
the most natural and efficient motion of the robot.
3. It is possible that no solutions exist
Applications of Inverse Kinematics
Manufacturing and assembly
Surgery
Search and rescue
Methods for Solving Inverse Kinematics
Closed-form methods
Iterative methods
Closed-form solutions are desirable because they are faster than
numerical solutions and identify all possible solutions.
They are not general, but robot dependent.
To calculate, they take advantage of particular geometric
features of specific robot mechanisms.
As the number of joints increases, this becomes increasingly
difficult.
For some serial-link robot manipulators, no analytical (closed
form) solution exists!
Closed-form Methods
Algebraic methods
Geometric methods
Algebraic Methods for Solving Inverse Kinematics
1. Identify the significant equations containing the joint variables.
2. Manipulate them into a soluble form.
A common strategy is reduction to a equation in a single
variable, e.g.
C1 cos θi + C2 sin θi + C3 = 0
where C1, C2, C3 are constants.
Algebraic Methods for Solving Inverse Kinematics (cont’d)
Another useful strategy is the reduction to a pair of equations
having the form:
C1 cos θi + C2 sin θi + C3 = 0
C1 sin θi − C2 cos θi + C4 = 0
only one solution:
θi = atan2(−C1C4 − C2C3, C2C4 − C1C3)
Geometric Methods for Solving Inverse Kinematics
Such methods involve identifying points on the manipulator
relative to which position and/or orientation can be expressed as a
function of a reduced set of the joint variables using trigonometric
relationships.
−→ often results to the decomposition of the spatial problem into
separate planar problems.
Decomposition of the full problem into inverse position
kinematics and inverse orientation kinematics.
The solution is derived by rewriting equation (2) as:
Calculating Analytical Solutions in the Python Robotics
Toolbox
Consider a 2-joint Planar (2D) Robot
Given the position of the end-effector (xE , yE ) calculate the
required joint angles to achieve this position.
Calculating Analytical Solutions in Python (con’d)
>>> a1 = Symbol('a1')
>>> a2 = Symbol('a2')
>>> q1, q2 = symbols("q1:3")
>>> e = ET2.R()*ET2.tx(a1)*ET2.R()*ET2.tx(a2)
>>> TE = e.fkine([q1, q2])
>>> x_fk, y_fk = TE.t
>>> print(x_fk)
>>> print(y_fk)
>> x, y = symbols("x, y")
>>> eq1 = (x_fk**2 + y_fk**2 - x**2 - y**2).trigsimp()
>>> print(eq1)
>>> q2_sol = sympy.solve(eq1, q2)
>>> print(q2_sol)
>>> eq2 = tuple(map(sympy.expand_trig, [x_fk - x, y_fk - y]))
>>> print(eq2)
>>> q1_sol = sympy.solve(eq2, [sympy.sin(q1), sympy.cos(q1)])
>>> print(q1_sol)
Calculating Analytical Solutions in Python (con’d)
>>> print(q1_sol[sin(q1)]/q1_sol[cos(q1)])
>>> sympy.atan2(q1_sol[sin(q1)], q1_sol[cos(q1)]).simplify()
Iterative (Numerical) Methods for Solving Inverse
Kinematics
Can be applied to any kinematic robot structure (not robot
dependent).
Slower
In some cases they do not compute all possible solutions
Refining the solution through iterations
Initial starting point affects the solution time
How? Minimise the error between the forward kinematics solution
and the desired end-effector pose ξE :
(FK(q) − ξE )
Numerical Methods for Solving Inverse Kinematics (cont’d)
Various classical numerical methods can be applied, including
among others:
Newton-Raphson: first order approximation of original
equations
Levenberg–Marquardt optimisation: using the second order
derivative for the approximation of the original system.
The Newton-Raphson Algorithm
The slope (tangent) of a function f (x) for x = xn is defined
(calculated) by the derivative of the function at that point:
Calculating Numerical Solutions in Python
>>> a1 = 1; a2 = 1
>>> q1, q2 = symbols("q1:3")
>>> e = ET2.R()*ET2.tx(a1)*ET2.R()*ET2.tx(a2)
>>> des_pos = np.array([0.5, 0.4])
>>> def E(q):
>>> return np.linalg.norm(e.fkine(q).t - des_pos)
>>> sol = optimize.minimize(E, [0, 0])
>>> print(sol.x)
[1.91964289 3.7933814 ]
>>> e.fkine(sol.x).printline()
Other topics/issues in Robotics
1. Forward Instantaneous Kinematics
→ Given all members of the kinematic chain and the rates of
motion about all joints, find the total velocity of the
end-effector.
→ Usage of the Jacobian matrix J(q) k vN = J(q)q˙ (6)
where k vN is the velocity of the end-effector expressed in any
frame k
2. Inverse Instantaneous Kinematics
→ Given the positions of all the members of the kinematic chain
and the total velocity of the end-effector, find the rates of the
motion of all joints.
→ Usage of the inverse of the Jacobian matrix
q˙ = J −1 (q)vn (7)
What is Robot Control?
A robot needs to move its joints to achieve tasks
A mobile robot moves to different locations
The movement of a robot (joints) is done using actuators.
In general, everything can be considered as control:
Decisions we make affect (control) our future
Decision while driving affect (control) the next position and
the final location
Control theory is a big area used not only in engineering and
robotics, but in computer science
Can be seen as what is the best next action to take (given a
specific state) so as to achieve (optimise) specific objectives!
Actuators (Motors)
An actuator is a device that causes motion.
Linear motion Rotary
The output of an actuator can be:
Speed Force
Types of actuators:
Hydraulic (using compressed oil)
Pneumatic (using compressed air)
Electric (using current)
Components of a Robot Joint Control System
The dynamic system that is to be controlled is called the plant.
Open-loop Control
In open-loop (feedforward) control the control action from the
controller is independent of the process (dynamic system or plant)
output.
Example:
Control of a boiler using a timer
The controller (timer) switches the boiler on or off based on
specific times and independent of the temperatures that the
boiler has reached
There is no feedback from the plant output to the controller on
what actual temperature the boiler reached.
Closed-loop (Feedback) control
Block diagram of closed-loop control system:
Error Response
How to measure the current error?
The desired behaviour (reference input) is compared with the
current actual output value of the plant:
Example:
the desired joint position of a multi-joint robot is θd (t) and the
actual joint position is θ(t), then the joint error is:
θe = θd (t) − θ(t) (1)
Error Response - Characteristics of a Good Controller
An ideal controller would drive the error to 0 instantly and keep it
0 forever.
In practice, it takes time to reduce the error and a value of 0
might never be achieved.
Error Response - Characteristics of a good Controller
(cont’d)
An error response θe (t) can be described by:
A steady-state response: the error as time goes to infinity
A transient response
→ overshoot
→ settling time
A good controller achieves an error response θe (t) with:
small or no steady-state error
little or no overshoot
a short settling time
stability: a steady state error is achieved (no oscillations)
A Surgical Robot - Impedance Control
(Mass-Spring-Damper Example)
Simulation of a robot used as a haptic surgical simulator,
mimicking the mass, stiffness and damping properties of a virtual
surgical instrument in contact with virtual tissue.
surgical instrument −→ mass
tissue −→ spring
The Surgical Robot (cont’d)
The dynamics of a 1-degree of freedom robot rendering an
impedance is described by:
mx¨ + bx˙ + kx = f (2) where:
x: is the position
m: is the mass (surgical instrument)
b: is the damping
k: is the stiffness (of the tissue)
f : is the force
The Surgical Robot (cont’d)
To simplify, in the case that the robot force f = 0, the above
second order differential equation can be written in the equivalent
form of an algebraic equation:
ms2 + bs + k = 0 (3)
One can solve for s, called the poles of the system
The poles define the response (position) of the system as a
function of time
Simulink Model of the Surgical Robot
Experiment with different values... Start with:
b = 10, k = 400, m = 2, x0 = 1, v0 = 0.
Running the Simulink Model
Multi-joint Manipulator Robot Control - Independent Joint Control
A common robot control strategy for manipulators:
Each joint has its own controller
Only a few parameters of the software controller need to be
changed
The parameters changed are done in a way so that they can
control the different size motors attached to each joint.
Discrete vs Continuous Dynamic Systems
Discrete dynamic systems are described by difference
equations.
x(n) = 5 ∗ x(n − 1) + 6 ∗ x(n − 2) + 2
Continuous dynamic systems are described by differential
equations.
x¨ = 5 ∗ x˙ + 10 ∗ x + 10
Linear vs Non-Linear Control
Although in practice many robotic systems follow non-linear
dynamics:
In many cases, a linear model can be developed
→ It is much easier to develop controllers
→ It is much easier to analyse mathematically
→ It is much easier to prove mathematically the stability
Not possible to linearise complex dynamic systems!
Fascination for mechanical devices such as automata. Complex
enough (for that time) to demonstrate some kind of life-like
behaviour.
Example: Vaucanson’s Digesting Duck (1739)
Flap its wings, eat grain and defecate.
History and Terminology
The word robot comes from the word Robota in Slav languages,
first introduced by the Czech playwright Karel Capek in his play
Rossum’s Universal Robots (RUR) in 1920.
The original word meaning was worker.
Robots in many:
Science fiction books
Movies
In the mid twentieth century (1950s), the term cybernetics was
used for an exciting science field to understand life and create
intelligent machines.
The 3 Laws of Robotics — Isaac Asimov
1. A robot may not injure a human being or, through inaction,
allow a human being to come to harm.
2. A robot must obey orders given it by human beings except
where such orders would conflict with the First Law.
3. A robot must protect its own existence as long as such
protection does not conflict with the First or Second Law.
Videos on the history of Robots
Robots - A 50 year journey video
Robots - The journey continues video
Robots Today
Robotic automation mostly used in Manufacturing (millions of
arm-type robots), e.g.:
Welding
Painting
Machine loading/unloading
Electronic assembly
Packaging
Categorising Robots and Definition
or according to the operation they perform:
Manufacturing
Service robots (cleaning, personal care, medical)
Field robots (agriculture, mining, construction)
Humanoid robots having the form of a human being and
perform more than one tasks
A robot can be defined as (according to Peter Corke):
A goal oriented machine that can sense, plan and act.
Characteristics of Robots
Consistent
Accurate
Reliable
Do things that people
Rigid Bodies and Robotic Manipulators
The most common form of an industrial robot is a mechanical
manipulator.
A mechanical manipulator consists of rigid bodies (called
links) connected by joints.
The joints are moved by actuators (e.g. electric motors).
An end-effector (gripper or hand) is usually attached to a
specific link.
A rigid body is a solid body which cannot be deformed (or
deformations are so small that can be neglected). The distance
between any 2 points in a rigid body remains constant even if
forces are applied.
A Modern Robotic Manipulator
The above has 7 degrees of freedom as explained a bit later.
Robotic Manipulators and Configuration Space
How can we specify the exact position of a Robot
(Manipulator)?
Configuration Space
The answer to How can we specify the exact position of a Robot is:
Robot Configuration: a specification of the positions of ALL
points of a robot.
Robot links are rigid: therefore only a few numbers are required to
represent its configuration (position).
Configuration Examples
How can the configuration of a door be represented?
→ angle θ about its hinge.
How can the configuration of a point on a plane be
described?
→ 2 coordinates (x, y)
Configuration Examples (cont’d)
How can the configuration of a coin lying on a table be
represented?
→ 2 coordinates (x, y) to specify the position of a fixed point on
the coin (e.g. centre) and one coordinate θ to specify the
coin’s orientation.
Degrees of Freedom of a Body (Robot)
The number of degrees of freedom (DOF) of a body (e.g.
robot) is the smallest number of real-valued variables needed to
represent its configuration.
C-space (Configuration space): the n-dimensional space of
all possible configurations of a robot
Degrees of freedom n is the dimension of the C-space
Degrees of Freedom of a Spatial Rigid Body
A rigid body in a 3-dimensional space (spatial rigid body ) has 6
degrees of freedom.
(x, y, z) coordinates for specifying the position of the centre
of mass.
3 angles (θ, ϕ, ψ) specifying the orientation (attitude) of the body.
Degrees of Freedom for Robots
degrees of freedom = (sum of freedoms of the bodies) − (number of independent
constraints)
Types of Robot Joints
Robot Joints: Degrees of Freedom and Constraints
A joint provides freedoms to allow one rigid body to move
relative to another
Equivalently: a joint provides constraints on the possible
motions of 2 rigid bodies it connects.
The number of degrees of freedom of a rigid body (3 for
planar bodies, 6 for spatial bodies) minus the number of
constraints provided by a joint, must equal the number of
freedoms provided by that joint.
Gr¨ubler’s Formula
The number of degrees of freedom of a mechanism (robot) with
links and joints can be calculated using Gr¨ubler’s formula which is
another expression of equation (1).
Consider a mechanism (robot) with:
N links (ground is also considered a link)
J joints
m is the degrees of freedom of a rigid body (m = 3 for planar
mechanisms, m = 6 for spatial mechanisms)
fi the number of freedoms provided by joint i
ci the number of constraints provided by joint i, where
fi + ci = m for all joints i.
assumes all joints constraints are independent.
Derivation of Gr¨ubler’s Formula
dof = m · (N − 1)
Example of Gr¨ubler’s Formula - Open Chain
3R Serial (3 revolute)
Example of Gr¨ubler’s Formula - Closed Chain mechanism
Four-Bar Linkage
Example of Gr¨ubler’s Formula - Slider-crank mechanism
3 revolute joints and 1 prismatic joint.
Coordinate Reference Frames
Fixed to the body of the robot (moving)
Fixed somewhere in space (stationary)
Kinematics
The science of motion that treats motion without considering the
forces which cause it.
Forward Kinematics: Given the angles for the joints, calculate
(compute) the position and orientation of the end-effector of
a robot manipulator.
Kinematics (cont’d)
Inverse Kinematics: Given the position and orientation of the
end-effector, compute all possible sets of joint angles that can
be used to achieve this position and orientation.
Robots: Social Issues and Ethics
Social Issues:
Workers losing their jobs and income
Economic consequences
Legal rights of robots (as they become more advanced and
intelligent)
Ethical Issues:
Self driving cars:
→ Choosing to save the driver over pedestrians or other drivers, etc?
→ Who is to blame, the robot, the software developer, the
manufacturer or the owner of the car?
Robotic healthcare: who to blame (surgery failure, etc.)?
Robots looking after elderly people: is this right as their
quality of life is affected by removing human contact?
Robots looking after children?
Using robots in armies to kill human beings?
Pose is the position and orientation of one coordinate frame with
respect to another reference coordinate frame.
Multiple coordinate frames are used in robotics to facilitate
the computations for motion and different types of
functionality.
NASA is using them to simplify calculations!
Pose (cont’d)
The robotic hand needs to grasp something located in a specific
point in space.
The orientation of the hand needs to be described
A coordinate frame is attached to the body (hand)
The coordinate frame attached to the body needs to be
described with respect to a reference coordinate frame
(possible the world coordinate frame)
Reference Frames in Real World Robots
How to Specify Pose
Using transformations:
Rotation
→ represents orientation
→ changes the reference frame in which a vector or frame is
represented
→ rotate a vector or a frame
Translation (linear move along one of the axes)
Terminology of Coordinate Frames
Pose in the 2D Space
Rotation:
A new coordinate frame {B} with the same origin as {A} but
rotated counter-clockwise by angle θ (positive angle)
Transforms vectors (their coordinates) from new frame {B} to
the old frame {A}:
cosθ −sinθ
sinθ cosθ
Properties of the Rotation Matrix
The inverse matrix is the same as the Transpose! R
→ easy to compute
The determinant is 1: det(R) = 1
→ the length of a vector is unchanged after the rotation
Creating a rotation matrix in the Python Robotics Toolbox
>>> R = rot2(math.pi/2)
>>> rot2(90, 'deg')
Visualising Rotation
The orientation represented by a rotation matrix can be visualised
as a coordinate frame:
R2 = rot2(-math.pi/2)
trplot2(R2)
Operations for Matrix Rotations
The product of two rotation matrices is also a rotation matrix:
R2=rot2(-math.pi/2)
R=rot2(math.pi/2)
R@R2
theta = Symbol('theta')
R = Matrix(rot2(theta))
>>> R*R
>>> R.det()
>>> R.det().simplify()
How to Represent Translation
Just a vector with 2 elements corresponding to how much we move
along the x and y axes.
V = vx/vy
Assuming P is the position of some object in a 2D space then we
can apply transformation TV by simply adding V to P:
TV (P) = P + V (3)
Homogeneous Form
To represent both rotation and translation using a single matrix:
cosθ −sinθ Vx
sinθ cosθ Vy
0 0 1
The left part is the rotation matrix and the right column is the
translation vector!
A row [0, 0, 1] is appended in the end.
The above represents first a translation (Vx , Vy ) followed by a
rotation with angle θ.
Derivation of the Homogeneous Form
Derivation of the Homogeneous Form (cont’d)
The homogeneous transformation can be considered as the
relative pose which first translates the coordinate frame by
AtB with respect to frame {A} and then is rotated by ARB (θ)
Working with the Toolbox for Homogeneous
Transformations
>>> trot2(0.3)
followed by a rotation of 0.3 radians:
>>> transl2(0, 0) @ trot2(0.3)
An example of a translation of (1, 2) followed by a rotation of 30
degrees:
>>> TA = transl2(1,2) @ trot2(30, "deg")
A coordinate frame representing the above pose can be plotted:
plotvol2([0, 5]);
trplot2(TA, frame="A", color="b");
T0 = transl2(0, 0);
trplot2(T0, frame="0", color="k");
Working with the Toolbox for Homogeneous
Transformations (cont’d)
Pose in the 3D Space
Rotation:
A new coordinate frame {B} with the same origin as {A} but
rotated with respect to {A}
Transforms vectors from new frame {B} to the old frame {A}:
Elementary Rotation Matrices in 3D
Rotation about the x-axis:
Rx (θ) =
1 0 0
0 cosθ −sinθ
0 sinθ cosθ
Rotation about the y-axis:
Ry (θ) =
cosθ 0 sinθ
0 1 0
−sinθ 0 cosθ
Rotation about the z-axis:
Rz (θ) =
cosθ −sinθ 0
sinθ cosθ 0
0 0 1
Properties of the 3D Rotation Matrix
Similarly with the 2D case:
The inverse matrix is the same as the Transpose! R
−1 = R(T)
→ easy to compute
The determinant is 1: det(R) = 1
→ the length of a vector is unchanged after the rotation
Rotations in 3D are not commutative (the order of rotation
matters!)
Representation of Rotation in 3D as an Axis-Angle
Combining:
a unit vector e indicating a single axis of rotation
an angle θ describing the magnitude of the rotation about the
a rotation of 90◦ about the z-axis.
Python Toolbox Example
Rx (π/2) can be represented as:
>>> R = rotx(math.pi / 2)
The orientation represented by a rotation matrix can be visualized
as a coordinate frame rotated with respect to the reference
coordinate frame:
How to Represent Translation in 3D
Just a vector with 3 elements corresponding to how much we move
along the x, y and z axes.
V = vx / vy /vz
Assuming P is the position of some object then we can apply
transformation TV by simply adding V to P:
TV (P) = P + V
Representing Pose in 3D
Different ways:
Vector and 3 angles (roll, pitch, yaw)
Homogeneous transformation (rotation and translation)
→ advantage of transformations calculations using matrix
multiplications!
Homogeneous Transformation in 3D
Construct a 4 × 4 array with the rotation matrix with 3 zeros (0) in
the row below it, and the translation vector with an extra element
of 1, as a column next to the rotation matrix:
e.g. rotation about x-axis with translation elements of vx , xy , vz
−→ Remember, the matrix-based transformations allow to apply
them (or even to combine them!) using matrix multiplication!
Homogeneous Transformation in 3D - Inverse Transformation
Although the inverse of the homogeneous transformation can be
calculated as normally by computing the inverse of the original
matrix (transformation), this can be done much faster.