Robot Toolbox MATLAB
Robot Toolbox MATLAB
TOOLBOX
for MATLAB
(Release 7.1)
z
y x
5.5
4.5
4
I11
Puma 560
3.5
2.5
2
4
2 4
0 2
0
!2
0.8 !2
0.6 q3 !4 !4
q2
!2002
c by Peter I. Corke.
3
1
Preface
1 Introduction
This, the seventh release of the Toolbox, represents nearly a decade of tinkering and a sub-
stantial level of maturity. It finally includes many of the features I’ve been planning to add
for some time, particularly MEX files, Simulink support and modified Denavit-Hartenberg
support. The previous release has had thousands of downloads and the mailing list has
hundreds of subscribers.
The Toolbox provides many functions that are useful in robotics including such things as
kinematics, dynamics, and trajectory generation. The Toolbox is useful for simulation as
well as analyzing results from experiments with real robots.
The Toolbox is based on a very general method of representing the kinematics and dynam-
ics of serial-link manipulators. These parameters are encapsulated in Matlab objects. Robot
objects can be created by the user for any serial-link manipulator and a number of examples
are provided for well know robots such as the Puma 560 and the Stanford arm. The Toolbox
also provides functions for manipulating and converting between datatypes such as vec-
tors, homogeneous transformations and unit-quaternions which are necessary to represent
3-dimensional position and orientation.
The routines are written in a straightforward manner which allows for easy understanding,
perhaps at the expense of computational efficiency. My guide in all of this work has been
the book of Paul[1], now out of print, but which I grew up with. If you feel strongly about
computational efficiency then you can always rewrite the function to be more efficient,
compile the M-file using the Matlab compiler, or create a MEX version.
This release has some significant new functionality as well as some bug fixes.
• Full support for modified (Craig’s) Denavit-Hartenberg notation, forward and inverse
kinematics, Jacobians and forward and inverse dynamics.
• Simulink blockset library and demonstrations included, see Section 2
• MEX implementation of recursive Newton-Euler algorithm written in portable C.
Speed improvements of at least 1000. Tested on Solaris, Linux and Windows. See
Section 1.9.
• Fixed still more bugs and missing files in quaternion code.
• Remove ‘@’ notation from fdyn to allow operation under Matlab 5 and 6.
• Fairly major update of documentation to ensure consistency between code, online
help and this manual.
1 INTRODUCTION 4
All code is now under CVS control which should eliminate many of the versioning problems
I had previously due to developing the code across multiple computers.
1.2 Contact
http://www.cat.csiro.au/cmst/staff/pic/robot
This page will always list the current released version number as well as bug fixes and new
code in between major releases.
A Mailing List is also available, subscriptions details are available off that web page.
The Robotics Toolbox is freely available from the Toolbox home page at
http://www.cat.csiro.au/cmst/staff/pic/robot
The files are available in either gzipped tar format (.gz) or zip format (.zip). The web page
requests some information from you regarding such as your country, type of organization
and application. This is just a means for me to gauge interest and to help convince my
bosses (and myself) that this is a worthwhile activity.
The file robot.pdf is a comprehensive manual with a tutorial introduction and details
of each Toolbox function. A menu-driven demonstration can be invoked by the function
rtdemo.
The Toolbox works with MATLAB version 6 and greater and has been tested on a Sun with
version 6.
The Toolbox does not function under MATLAB v3.x or v4.x since those versions do not
support objects. An older version of the Toolbox, available from the Matlab4 ftp site is
workable but lacks some features of this current Toolbox release.
1.5 Acknowledgements
I am grateful for the support of my employer, CSIRO, for supporting me in this activity and
providing me with the Matlab tools and web server.
I have corresponded with a great many people via email since the first release of this Tool-
box. Some have identified bugs and shortcomings in the documentation, and even better,
some have provided bug fixes and even new modules, thankyou.
1 INTRODUCTION 5
I’m always happy to correspond with people who have found genuine bugs or deficiencies
in the Toolbox, or who have suggestions about ways to improve its functionality. However
I draw the line at providing help for people with their assignments and homework!
Many people are using the Toolbox for teaching and this is something that I would encour-
age. If you plan to duplicate the documentation for class use then every copy must include
the front page.
If you want to cite the Toolbox please use
@ARTICLE{Corke96b,
AUTHOR = {P.I. Corke},
JOURNAL = {IEEE Robotics and Automation Magazine},
MONTH = mar,
NUMBER = {1},
PAGES = {24-32},
TITLE = {A Robotics Toolbox for {MATLAB}},
VOLUME = {3},
YEAR = {1996}
}
Many people are not aware that there are two quite different forms of Denavit-Hartenberg
representation for serial-link manipulator kinematics:
1. Classical as per the original 1955 paper of Denavit and Hartenberg, and used in text-
books such as by Paul[1], Fu etal[2], or Spong and Vidyasagar[3].
2. Modified form as introduced by Craig[4] in his text book.
Both notations represent a joint as 2 translations (A and D) and 2 rotation angles (! and ").
However the expressions for the link transform matrices are quite different. In short, you
must know which kinematic convention your Denavit-Hartenberg parameters conform to.
Unfortunately many sources in the literature do not specify this crucial piece of information.
Most textbooks cover only one and do not even allude to the existence of the other. These
issues are discussed further in Section 3.
The Toolbox has full support for both the classical and modified conventions.
Let’s take a simple example like the two-link planar manipulator from Spong & Vidyasagar[3]
(Figure 3-6, p73) which has the following (standard) Denavit-Hartenberg link parameters
1 INTRODUCTION 6
Link ai !i di "i
1 1 0 0 "∗1
2 1 0 0 "∗2
where we have set the link lengths to 1. Now we can create a pair of link objects:
L1 =
L2 =
r =
>>
The first few lines create link objects, one per robot link. Note the second argument to
link which specifies that the standard D&H conventions are to be used (this is actually the
default). The arguments to the link object can be found from
which shows the order in which the link parameters must be passed (which is different to
the column order of the table above). The fifth element of the first argument, sigma, is a
flag that indicates whether the joint is revolute (sigma is zero) or primsmatic (sigma is non
zero).
1 INTRODUCTION 7
1.5
0.5 yz x
0
Z
!0.5 noname
!1
!1.5
!2
2
1 2
0 1
0
!1
!1
Y !2 !2
X
The link objects are passed as a cell array to the robot() function which creates a robot
object which is in turn passed to many of the other Toolbox functions.
Note that displays of link data include the kinematic convention in brackets on the far right.
(std) for standard form, and (mod) for modified form.
The robot just created can be displayed graphically by
The Robotics Toolbox Release 7 includes portable C source code to generate a MEX file
version of the rne function.
The MEX file runs upto 500 times faster than the interpretted version rne.m and this is
critical for calculations involving forward dynamics. The forward dynamics requires the
calculation of the manipulator inertia matrix at each integration time step. The Toolbox uses
a computationally simple but inefficient method that requires evaluating the rne function
n + 1 times, where n is the number of robot axes. For forward dynamics the rne is the
bottleneck.
The Toolbox stores all robot kinematic and inertial parameters in a robot object, but ac-
cessing these parameters from a C language MEX file is somewhat cumbersome and must
be done on each call. Therefore the speed advantage increases with the number of rows in
the q, qd and qdd matrices that are provided. In other words it is better to call rne with a
trajectory, than for each point on a trajectory.
To build the MEX file:
2. On a Unix system just type make. For other platforms follow the Mathworks guide-
lines. You need to compile and link three files with a command something like mex
frne.c ne.c vmath.c.
3. If successful you now have a file called frne.ext where ext is the file extension and
depends on the architecture (mexsol for Solaris, mexlx for Linux).
4. From within Matlab cd into this same directory and run the test script
>> cd ROBOTDIR/mex
>> check
***************************************************************
************************ Puma 560 *****************************
***************************************************************
************************ normal case *****************************
DH: Fast RNE: (c) Peter Corke 2002
Speedup is 17, worst case error is 0.000000
MDH: Speedup is 1565, worst case error is 0.000000
************************ no gravity *****************************
DH: Speedup is 1501, worst case error is 0.000000
MDH: Speedup is 1509, worst case error is 0.000000
************************ ext force *****************************
DH: Speedup is 1497, worst case error is 0.000000
MDH: Speedup is 637, worst case error is 0.000000
***************************************************************
********************** Stanford arm ***************************
***************************************************************
************************ normal case *****************************
DH: Speedup is 1490, worst case error is 0.000000
MDH: Speedup is 1519, worst case error is 0.000000
************************ no gravity *****************************
DH: Speedup is 1471, worst case error is 0.000000
MDH: Speedup is 1450, worst case error is 0.000000
************************ ext force *****************************
DH: Speedup is 417, worst case error is 0.000000
MDH: Speedup is 1458, worst case error is 0.000000
>>
This will run the M-file and MEX-file versions of the rne function for various robot
models and options with various options. For each case it should report a speedup
greater than one, and an error of zero. The results shown above are for a Sparc Ultra
10.
5. Copy the MEX-file frne.ext into the Robotics Toolbox main directory with the
name rne.ext. Thus all future references to rne will now invoke the MEX-file
instead of the M-file. The first time you run the MEX-file in any Matlab session it
will print a one-line identification message.
9
2
Using the Toolbox with Simulink
2 Introduction
Simulink is the block diagram editing and simulation environment for Matlab. Until its
most recent release Simulink has not been able to handle matrix valued signals, and that has
made its application to robotics somewhat clumsy. This shortcoming has been rectified with
Simulink Release 4. Robot Toolbox Release 7 and higher includes a library of blocks for
use in constructing robot kinematic and dynamic models.
To use this new feature it is neccessary to include the Toolbox Simulink block directory in
your Matlab path:
>> roblocks
noname noname x x
q
q
y T T y
noname q n J
tau qd qd
J z z
qdd
qdd xyz2T T2xyz
plot jacobn
jtraj roll roll
Robot
pitch T T pitch
0
J !1 yaw yaw
q J J
noname J Ji
q TODO rpy2T T2rpy
noname
qd tau
a a
jacob0
qdd ijacob
b T T b
rne noname
T1 c c
q T dx eul2T T2eul
T2
fkine tr2diff
Copyright (c) 2002 Peter Corke
[0 0 0 0 0 0]’ tau qd
plot
Zero qdd
torque
Robot
0
simout
Clock
To Workspace
Simple dynamics demo
pic
11!Feb!2002 14:19:49
Figure 3: Robotics Toolbox example demo1, Puma robot collapsing under gravity.
3 Examples
The Simulink model, demo1, is shown in Figure 3, and the two blocks in this model would
be familiar to Toolbox users. The Robot block is similar to the fdyn() function and repre-
sents the forward dynamics of the robot, and the plot block represents the plot function.
Note the parameters of the Robot block contain the robot object to be simulated and the
initial joint angles. The plot block has one parameter which is the robot object to be dis-
played graphically and should be consistent with the robot being simulated. Display options
are taken from the plotbotopt.m file, see help for robot/plot for details.
To run this demo first create a robot object in the workspace,typically by using the puma560
command, then start the simulation using Simulation/Start option from the model toolbar.
>> puma560
>> demo1
The Simulink model, demo2, is shown in Figure 4, and represents a simple 2-link robot with
flexible or compliant transmission. The first joint receives a step position demand change at
time 1s. The resulting oscillation and dynamic coupling between the two joints can be seen
clearly. Note that the drive model comprises spring plus damper, and that the joint position
control loops are simply unity negative feedback.
To run this demo first create a 2-link robot object in the workspace,typically by using the
twolink command, then start the simulation using Simulation/Start option from the model
toolbar.
>> twolink
>> demo2
3 EXAMPLES 11
Constant Robot
du/dt 20
assume the motor Scope
is infinitely "stiff" Derivative B
0
transmission comprises
spring + damper Clock simout
2!link demo
pic To Workspace
Mon Apr 8 11:37:04 2002
The Simulink model, demo3, shown in Figure 5, is for a Puma560 with a computed torque
control structure. This is a “classical” dynamic control technique in which the rigid-body
dynamic model is inverted to compute the demand torque for the robot based on current joint
angles and joint angle rates and demand joint angle acceleration. This model introduces the
rne block which computes the inverse dynamics using the recursive Newton-Euler algo-
rithm (see rne function), and the jtraj block which computes a vector quintic polynomial.
jtraj has parameters which include the initial and final values of the each output element
as well as the overall motion time. Initial and final velocity are assumed to be zero.
In practice of course the dynamic model of the robot is not exactly known, we can only
invert our best estimate of the rigid-body dynamics. In the simulation we can model this
by using the perturb function to alter the parameters of the dynamic model used in the
rne block — note the ’P/’ prefix on the model name displayed by that block. This means
that the inverse dynamics are computed for a slightly different dynamic model to the robot
under control and shows the effect of model error on control performance.
To run this demo first create a robot object in the workspace,typically by using the puma560
command, then start the simulation using Simulation/Start option from the model toolbar.
Kp
Kd
0
Puma560 computed torque control
pic Clock
11!Feb!2002 14:18:39 simout
To Workspace
>> puma560
>> demo3
The Simulink model demo4 demonstrates torque feedforward control, another “classical”
dynamic control technique in which the demanded torque is computed using the rne block
and added to the error torque computed from position and velocity error. It is instructive to
compare the structure of this model with demo3. The inverse dynamics are not in the for-
ward path and since the robot configuration changes relatively slowly, they can be computed
at a low rate (this is illustrated by the zero-order hold block sampling at 20Hz).
To run this demo first create a robot object in the workspace,typically by using the puma560
command, then start the simulation using Simulation/Start option from the model toolbar.
>> puma560
>> demo4
The Simulink model, demo5, shown in Figure 6, demonstrates Cartesian space motion con-
trol. There are two conventional approaches to this. Firstly, resolve the Cartesian space
demand to joint space using inverse kinematics and then perform the control in joint space.
The second, used here, is to compute the error in Cartesian space and resolve that to joint
space via the inverse Jacobian. This eliminates the need for inverse kinematics within the
control loop, and its attendent problems of multiple solutions. It also illustrates some addi-
tional Simulink blocks.
This demonstration is for a Puma 560 robot moving the tool in a circle of radius 0.05m
centered at the point (0.5, 0, 0). The difference between the Cartesian demand and the
current Cartesian pose (in end-effector coordinates) is computed by the tr2diff block
which produces a differential motion described by a 6-vector. The Jacobian block has as
its input the current manipulator joint angles and outputs the Jacobian matrix. Since the
differential motion is with respect to the end-effector we use the JacobianN block rather
than Jacobian0. We use standard Simulink block to invert the Jacobian and multiply it by
Cartesian control
q
!1
J
Bad Link J Ji
Puma 560
Cartesian circle jacob0 ijacob
Clock Matrix 1
y T T1 !0.6 q T
Multiply s q
dx
z
0.05*sin(u) T2
xyz2T Rate fkine x
controlled
tr2diff
robot T y
axes XY Graph
z
T2xyz
0
the differential motion. The result, after application of a simple proportional gain, is the
joint space motion required to correct the Cartesian error. The robot is modelled by an
integrator as a simple rate control device, or velocity servo.
This example also demonstrates the use of the fkine block for forward kinematics and the
T2xyz block which extracts the translational part of the robot’s Cartesian state for plotting
on the XY plane.
This demonstration is very similar to the numerical method used to solve the inverse kine-
matics problem in ikine.
To run this demo first create a robot object in the workspace,typically by using the puma560
command, then start the simulation using Simulation/Start option from the model toolbar.
>> puma560
>> demo5
The Simulink model, demo6, shown in Figure 7, demonstrates image-based visual servoing
(IBVS)[5]. This is quite a complex example that simulates not only the robot but also a
camera and the IBVS algorithm. The camera is assumed to be mounted on the robot’s end-
effector and this coordinate is passed into the camera block so that the relative position of
the target with respect to the camera can be computed. Arguments to the camera block
include the 3D coordinates of the target points. The output of the camera is the 2D image
plane coordinates of the target points. The target points are used to compute an image
Jacobian matrix which is inverted and multiplies the desired motion of the target points on
the image plane. The desired motion is simply the difference between the observed target
points and the desired point positions. The result is a velocity screw which drives the robot
to the desired pose with respect to a square target.
When the simulation starts a new window, the camera view, pops up. We see that initially
the square target is off to one side and somewhat oblique. The image plane errors are
mapped by an image Jacobian into desired Cartesian rates, and these are futher mapped by a
Puma 560
!1
[6x6] J [6x6] Puma 560
q n J J Ji
6 [6x6] Matrix 6 1 6 [4x4] [4x2] [8x6] MATLAB [6x8]
J q T T uv uv J
Multiply s q 6 [4x2] [8x6] Function Matrix 6
jacobn 6 pinv Multiply
ijacob
Rate fkine visual 8
controlled Jacobian
camera
robot
axes
256 456 [4x2]
456 456 [4x2] [4x2] MATLAB 8
456 256 Function feature vel
256 256
desired
0.21 85.07
image plane MATLAB 274
0.28 coordinates 107.89 8 Function
!0.32 112.87 feature error
6 norm
!0.00 109.40
0.04 108.90 8
!0.01 80.92 8
81.10 feature
error
Cartesian velocity dmd 80.81
manipulator Jacobian into joint rates which are applied to the robot which is again modelled
as a rate control device. This closed-loop system is performing a Cartesian positioning task
using information from a camera rather than encoders and a kinematic model (the Jacobian
is a weak kinematic model). Image-based visual servoing schemes have been found to be
extremely robust with respect to errors in the camera model and manipulator Jacobian.
15
3
Tutorial
3 Manipulator kinematics
Kinematics is the study of motion without regard to the forces which cause it. Within kine-
matics one studies the position, velocity and acceleration, and all higher order derivatives of
the position variables. The kinematics of manipulators involves the study of the geometric
and time based properties of the motion, and in particular how the various links move with
respect to one another and with time.
Typical robots are serial-link manipulators comprising a set of bodies, called links, in a
chain, connected by joints 1 . Each joint has one degree of freedom, either translational or
rotational. For a manipulator with n joints numbered from 1 to n, there are n + 1 links,
numbered from 0 to n. Link 0 is the base of the manipulator, generally fixed, and link n
carries the end-effector. Joint i connects links i and i − 1.
A link may be considered as a rigid body defining the relationship between two neighbour-
ing joint axes. A link can be specified by two numbers, the link length and link twist, which
define the relative location of the two axes in space. The link parameters for the first and
last links are meaningless, but are arbitrarily chosen to be 0. Joints may be described by
two parameters. The link offset is the distance from one link to the next along the axis of the
joint. The joint angle is the rotation of one link with respect to the next about the joint axis.
To facilitate describing the location of each link we affix a coordinate frame to it — frame i
is attached to link i. Denavit and Hartenberg[6] proposed a matrix method of systematically
assigning coordinate systems to each link of an articulated chain. The axis of revolute joint
i is aligned with zi−1 . The xi−1 axis is directed along the normal from z i−1 to zi and for
intersecting axes is parallel to z i−1 × zi . The link and joint parameters may be summarized
as:
link length ai the offset distance between the z i−1 and zi axes along the
xi axis;
link twist !i the angle from the z i−1 axis to the zi axis about the x i axis;
link offset di the distance from the origin of frame i − 1 to the x i axis
along the zi−1 axis;
joint angle "i the angle between the x i−1 and xi axes about the zi−1 axis.
For a revolute axis " i is the joint variable and d i is constant, while for a prismatic joint d i
is variable, and "i is constant. In many of the formulations that follow we use generalized
coordinates, q i , where !
"i for a revolute joint
qi =
di for a prismatic joint
1
Parallel link and serial/parallel hybrid structures are possible, though much less common in industrial manip-
ulators.
3 MANIPULATOR KINEMATICS 16
1. Frame i has its origin along the axis of joint i + 1, as described by Paul[1] and Lee[2,
7].
3 MANIPULATOR KINEMATICS 17
2. Frame i has its origin along the axis of joint i, and is frequently referred to as ‘modi-
fied Denavit-Hartenberg’ (MDH) form[8]. This form is commonly used in literature
dealing with manipulator dynamics. The link transform matrix for this form differs
from (1).
Figure 8 shows the notational differences between the two forms. Note that a i is always the
length of link i, but is the displacement between the origins of frame i and frame i + 1 in
one convention, and frame i − 1 and frame i in the other 2. The Toolbox provides kinematic
functions for both of these conventions — those for modified DH parameters are prefixed
by ‘m’.
For an n-axis rigid-link manipulator, the forward kinematic solution gives the coordinate
frame, or pose, of the last link. It is obtained by repeated application of (2)
0 0
Tn = A1 1 A2 · · · n−1 An (3)
= K (q) (4)
which is the product of the coordinate frame transform matrices for each link. The pose
of the end-effector has 6 degrees of freedom in Cartesian space, 3 in translation and 3 in
rotation, so robot manipulators commonly have 6 joints or degrees of freedom to allow
arbitrary end-effector pose. The overall manipulator transform 0 Tn is frequently written as
Tn , or T6 for a 6-axis robot. The forward kinematic solution may be computed for any
manipulator, irrespective of the number of joints or kinematic structure.
Of more use in manipulator path planning is the inverse kinematic solution
q = K −1 (T) (5)
which gives the joint angles required to reach the specified end-effector position. In general
this solution is non-unique, and for some classes of manipulator no closed-form solution
exists. If the manipulator has more than 6 joints it is said to be redundant and the solution
for joint angles is under-determined. If no solution can be determined for a particular ma-
nipulator pose that configuration is said to be singular. The singularity may be due to an
alignment of axes reducing the effective degrees of freedom, or the point T being out of
reach.
The manipulator Jacobian matrix, J " , transforms velocities in joint space to velocities of
the end-effector in Cartesian space. For an n-axis manipulator the end-effector Cartesian
velocity is
0 0
ẋn = J" q̇ (6)
tn tn
ẋn = J" q̇ (7)
in base or end-effector coordinates respectively and where x is the Cartesian velocity rep-
resented by a 6-vector. For a 6-axis manipulator the Jacobian is square and provided it is
not singular can be inverted to solve for joint rates in terms of end-effector Cartesian rates.
The Jacobian will not be invertible at a kinematic singularity, and in practice will be poorly
2 Many papers when tabulating the ‘modified’ kinematic parameters of manipulators list a
i−1 and !i−1 not ai
and !i .
4 MANIPULATOR RIGID-BODY DYNAMICS 18
conditioned in the vicinity of the singularity, resulting in high joint rates. A control scheme
based on Cartesian rate control
q̇ = 0 J−1
"
0
ẋn (8)
was proposed by Whitney[9] and is known as resolved rate motion control. For two frames
A and B related by A TB = [n o a p] the Cartesian velocity in frame A may be transformed to
frame B by
B
ẋ = B JA A ẋ (9)
where the Jacobian is given by Paul[10] as
[n o a]T [p × n p × o p × a] T
( )
B
JA = f (A TB ) = (10)
0 [n o a]T
• inverse dynamics in which the manipulator’s equations of motion are solved for given
motion to determine the generalized forces, discussed further in Section 4.1, and
• direct dynamics in which the equations of motion are integrated to determine the
generalized coordinate response to applied generalized forces discussed further in
Section 4.2.
where
q is the vector of generalized joint coordinates describing the pose of the manipulator
q̇ is the vector of joint velocities;
q̈ is the vector of joint accelerations
M is the symmetric joint-space inertia matrix, or manipulator inertia tensor
C describes Coriolis and centripetal effects — Centripetal torques are proportional to q̇ 2i ,
while the Coriolis torques are proportional to q̇ i q̇ j
F describes viscous and Coulomb friction and is not generally considered part of the rigid-
body dynamics
G is the gravity loading
Q is the vector of generalized forces associated with the generalized coordinates q.
The equations may be derived via a number of techniques, including Lagrangian (energy
based), Newton-Euler, d’Alembert[2, 12] or Kane’s[13] method. The earliest reported work
was by Uicker[14] and Kahn[15] using the Lagrangian approach. Due to the enormous com-
putational cost, O(n 4 ), of this approach it was not possible to compute manipulator torque
for real-time control. To achieve real-time performance many approaches were suggested,
including table lookup[16] and approximation[17, 18]. The most common approximation
was to ignore the velocity-dependent term C, since accurate positioning and high speed
motion are exclusive in typical robot applications.
4 MANIPULATOR RIGID-BODY DYNAMICS 19
Table 1: Comparison of computational costs for inverse dynamics from various sources.
The last entry is achieved by symbolic simplification using the software package ARM.
Orin et al.[19] proposed an alternative approach based on the Newton-Euler (NE) equations
of rigid-body motion applied to each link. Armstrong[20] then showed how recursion might
be applied resulting in O(n) complexity. Luh et al.[21] provided a recursive formulation of
the Newton-Euler equations with linear and angular velocities referred to link coordinate
frames. They suggested a time improvement from 7.9s for the Lagrangian formulation
to 4.5 ms, and thus it became practical to implement ‘on-line’. Hollerbach[22] showed
how recursion could be applied to the Lagrangian form, and reduced the computation to
within a factor of 3 of the recursive NE. Silver[23] showed the equivalence of the recursive
Lagrangian and Newton-Euler forms, and that the difference in efficiency is due to the
representation of angular velocity.
“Kane’s equations” [13] provide another methodology for deriving the equations of motion
for a specific manipulator. A number of ‘Z’ variables are introduced, which while not
necessarily of physical significance, lead to a dynamics formulation with low computational
burden. Wampler[24] discusses the computational costs of Kane’s method in some detail.
The NE and Lagrange forms can be written generally in terms of the Denavit-Hartenberg
parameters — however the specific formulations, such as Kane’s, can have lower compu-
tational cost for the specific manipulator. Whilst the recursive forms are computationally
more efficient, the non-recursive forms compute the individual dynamic terms (M, C and
G) directly. A comparison of computation costs is given in Table 1.
The recursive Newton-Euler (RNE) formulation[21] computes the inverse manipulator dy-
namics, that is, the joint torques required for a given set of joint angles, velocities and
accelerations. The forward recursion propagates kinematic information — such as angu-
lar velocities, angular accelerations, linear accelerations — from the base reference frame
(inertial frame) to the end-effector. The backward recursion propagates the forces and mo-
ments exerted on each link from the end-effector of the manipulator to the base reference
frame3. Figure 9 shows the variables involved in the computation for one link.
The notation of Hollerbach[22] and Walker and Orin [26] will be used in which the left
superscript indicates the reference coordinate frame for the variable. The notation of Luh et
al.[21] and later Lee[7, 2] is considerably less clear.
3 It
should be noted that using MDH notation with its different axis assignment conventions the Newton Euler
formulation is expressed differently[8].
4 MANIPULATOR RIGID-BODY DYNAMICS 20
Figure 9: Notation used for inverse dynamics, based on standard Denavit-Hartenberg nota-
tion.
Outward recursion, 1 ≤ i ≤ n.
If axis i + 1 is rotational
* +
i+1
$i+1 = i+1 Ri i $i + z0 q̇i+1 (12)
, * +-
i+1
$̇i+1 = i+1 Ri i $̇i + z0 q̈i+1 + i $i × z0 q̇i+1 (13)
i+1
vi+1 = i+1
$i+1 × i+1 p∗i+1 + i+1 Ri i vi (14)
, -
i+1
v̇i+1 = i+1
$̇i+1 × i+1 p∗i+1 + i+1 $i+1 × i+1
$i+1 × i+1 p∗i+1 + i+1 Ri i v̇i (15)
If axis i + 1 is translational
i+1
$i+1 = i+1 Ri i $i (16)
i+1 i+1 i
$̇i+1 = Ri $̇i (17)
* +
i+1
vi+1 = i+1
Ri z0 q̇i+1 + i vi + i+1 $i+1 × i+1 p∗i+1 (18)
* + * +
i+1
v̇i+1 = i+1
Ri z0 q̈i+1 + i v̇i + i+1 $̇i+1 × i+1 p∗i+1 + 2 i+1 $i+1 × i+1 Ri z0 q̇i+1
* +
+i+1 $i+1 × i+1 $i+1 × i+1 p∗i+1 (19)
i i
$̇i × si + i $i × i $i × si + i v̇i
. /
v̇i = (20)
i i
Fi = mi v̇i (21)
i
= Ji $̇i + $i × Ji i $i
i i
0 1
Ni (22)
Inward recursion, n ≥ i ≥ 1.
i i
fi = Ri+1 i+1 f i+1 + i F i (23)
, * + - * +
i
ni = i
Ri+1 i+1 ni+1 + i+1 Ri i p∗i × ii+1 f i+1 + i p∗i + si × i F i + i N i (24)
0 1 0
i ni T i Ri+1 z
1
0 if link i + 1 is rotational
Qi = * +T 0 (25)
if iR
1
i i+1 z0 if link i + 1 is translational
where
4 MANIPULATOR RIGID-BODY DYNAMICS 21
i p∗ is the displacement from the origin of frame i − 1 to frame i with respect to frame i.
i
ai
i ∗
pi = di sin !i (28)
di cos !i
Note that the COM linear velocity given by equation (14) or (18) does not need to be com-
puted since no other expression depends upon it. Boundary conditions are used to introduce
the effect of gravity by setting the acceleration of the base link
v̇0 = −g (29)
where g is the gravity vector in the reference coordinate frame, generally acting in the
negative Z direction, downward. Base velocity is generally zero
v0 = 0 (30)
$0 = 0 (31)
$̇0 = 0 (32)
At this stage the Toolbox only provides an implementation of this algorithm using the stan-
dard Denavit-Hartenberg conventions.
Equation (11) may be used to compute the so-called inverse dynamics, that is, actuator
torque as a function of manipulator state and is useful for on-line control. For simulation
REFERENCES 22
the direct, integral or forward dynamic formulation is required giving joint motion in terms
of input torques.
Walker and Orin[26] describe several methods for computing the forward dynamics, and
all make use of an existing inverse dynamics solution. Using the RNE algorithm for in-
verse dynamics, the computational complexity of the forward dynamics using ‘Method 1’
is O(n3 ) for an n-axis manipulator. Their other methods are increasingly more sophisticated
but reduce the computational cost, though still O(n 3 ). Featherstone[27] has described the
“articulated-body method” for O(n) computation of forward dynamics, however for n < 9
it is more expensive than the approach of Walker and Orin. Another O(n) approach for
forward dynamics has been described by Lathrop[28].
• link mass, mi ;
• three first moments, which may be expressed as the COM location, s i , with respect to
some datum on the link or as a moment Si = mi si ;
• six second moments, which represent the inertia of the link about a given axis, typi-
cally through the COM. The second moments may be expressed in matrix or tensor
form as
Jxx Jxy Jxz
J = Jxy Jyy Jyz (33)
Jxz Jyz Jzz
where the diagonal elements are the moments of inertia, and the off-diagonals are
products of inertia. Only six of these nine elements are unique: three moments and
three products of inertia.
For any point in a rigid-body there is one set of axes known as the principal axes of
inertia for which the off-diagonal terms, or products, are zero. These axes are given
by the eigenvectors of the inertia matrix (33) and the eigenvalues are the principal
moments of inertia. Frequently the products of inertia of the robot links are zero due
to symmetry.
A 6-axis manipulator rigid-body dynamic model thus entails 60 inertial parameters. There
may be additional parameters per joint due to friction and motor armature inertia. Clearly,
establishing numeric values for this number of parameters is a difficult task. Many parame-
ters cannot be measured without dismantling the robot and performing careful experiments,
though this approach was used by Armstrong et al.[29]. Most parameters could be derived
from CAD models of the robots, but this information is often considered proprietary and
not made available to researchers.
References
[1] R. P. Paul, Robot Manipulators: Mathematics, Programming, and Control. Cam-
bridge, Massachusetts: MIT Press, 1981.
REFERENCES 23
[2] K. S. Fu, R. C. Gonzalez, and C. S. G. Lee, Robotics. Control, Sensing, Vision and
Intelligence. McGraw-Hill, 1987.
[3] M. Spong and M. Vidyasagar, Robot Dynamics and Control. John Wiley and Sons,
1989.
[4] J. J. Craig, Introduction to Robotics. Addison Wesley, 1986.
[5] S. Hutchinson, G. Hager, and P. Corke, “A tutorial on visual servo control,” IEEE
Transactions on Robotics and Automation, vol. 12, pp. 651–670, Oct. 1996.
[6] R. S. Hartenberg and J. Denavit, “A kinematic notation for lower pair mechanisms
based on matrices,” Journal of Applied Mechanics, vol. 77, pp. 215–221, June 1955.
[7] C. S. G. Lee, “Robot arm kinematics, dynamics and control,” IEEE Computer, vol. 15,
pp. 62–80, Dec. 1982.
[8] J. J. Craig, Introduction to Robotics. Addison Wesley, second ed., 1989.
[9] D. Whitney, “The mathematics of coordinated control of prosthetic arms and manipu-
lators,” ASME Journal of Dynamic Systems, Measurement and Control, vol. 20, no. 4,
pp. 303–309, 1972.
[10] R. P. Paul, B. Shimano, and G. E. Mayer, “Kinematic control equations for simple
manipulators,” IEEE Trans. Syst. Man Cybern., vol. 11, pp. 449–455, June 1981.
[11] J. M. Hollerbach, “Dynamics,” in Robot Motion - Planning and Control (M. Brady,
J. M. Hollerbach, T. L. Johnson, T. Lozano-Perez, and M. T. Mason, eds.), pp. 51–71,
MIT, 1982.
[12] C. S. G. Lee, B. Lee, and R. Nigham, “Development of the generalized D’Alembert
equations of motion for mechanical manipulators,” in Proc. 22nd CDC, (San Antonio,
Texas), pp. 1205–1210, 1983.
[13] T. Kane and D. Levinson, “The use of Kane’s dynamical equations in robotics,” Int. J.
Robot. Res., vol. 2, pp. 3–21, Fall 1983.
[14] J. Uicker, On the Dynamic Analysis of Spatial Linkages Using 4 by 4 Matrices. PhD
thesis, Dept. Mechanical Engineering and Astronautical Sciences, NorthWestern Uni-
versity, 1965.
[15] M. Kahn, “The near-minimum time control of open-loop articulated kinematic link-
ages,” Tech. Rep. AIM-106, Stanford University, 1969.
[16] M. H. Raibert and B. K. P. Horn, “Manipulator control using the configuration space
method,” The Industrial Robot, pp. 69–73, June 1978.
[17] A. Bejczy, “Robot arm dynamics and control,” Tech. Rep. NASA-CR-136935, NASA
JPL, Feb. 1974.
[18] R. Paul, “Modelling, trajectory calculation and servoing of a computer controlled
arm,” Tech. Rep. AIM-177, Stanford University, Artificial Intelligence Laboratory,
1972.
[19] D. Orin, R. McGhee, M. Vukobratovic, and G. Hartoch, “Kinematics and kinetic
analysis of open-chain linkages utilizing Newton-Euler methods,” Mathematical Bio-
sciences. An International Journal, vol. 43, pp. 107–130, Feb. 1979.
REFERENCES 24
2
Reference
For an n-axis manipulator the following matrix naming and dimensional conventions apply.
Symbol Dimensions Description
l link manipulator link object
q 1×n joint coordinate vector
q m×n m-point joint coordinate trajectory
qd 1×n joint velocity vector
qd m×n m-point joint velocity trajectory
qdd 1×n joint acceleration vector
qdd m×n m-point joint acceleration trajectory
robot robot robot object
T 4×4 homogeneous transform
T 4×4×m m-point homogeneous transform trajectory
Q quaternion unit-quaternion object
M 1×6 vector with elements of 0 or 1 corresponding to
Cartesian DOF along X, Y, Z and around X, Y, Z.
1 if that Cartesian DOF belongs to the task space,
else 0.
v 3×1 Cartesian vector
t m×1 time vector
d 6×1 differential motion vector
Object names are shown in bold typeface.
A trajectory is represented by a matrix in which each row corresponds to one of m time
steps. For a joint coordinate, velocity or acceleration trajectory the columns correspond
to the robot axes. For homogeneous transform trajectories we use 3-dimensional matrices
where the last index corresponds to the time step.
Units
All angles are in radians. The choice of all other units is up to the user, and this choice will
flow on to the units in which homogeneous transforms, Jacobians, inertias and torques are
represented.
Homogeneous Transforms
eul2tr Euler angle to homogeneous transform
oa2tr orientation and approach vector to homogeneous transform
rotvec homogeneous transform for rotation about arbitrary vector
rotx homogeneous transform for rotation about X-axis
roty homogeneous transform for rotation about Y-axis
rotz homogeneous transform for rotation about Z-axis
rpy2tr Roll/pitch/yaw angles to homogeneous transform
tr2eul homogeneous transform to Euler angles
tr2rot homogeneous transform to rotation submatrix
tr2rpy homogeneous transform to roll/pitch/yaw angles
transl set or extract the translational component of a homoge-
neous transform
trnorm normalize a homogeneous transform
Trajectory Generation
ctraj Cartesian trajectory
jtraj joint space trajectory
trinterp interpolate homogeneous transforms
Quaternions
/ divide quaternion by quaternion or scalar
* multiply quaternion by a quaternion or vector
inv invert a quaternion
norm norm of a quaternion
plot display a quaternion as a 3D rotation
q2tr quaternion to homogeneous transform
quaternion construct a quaternion
qinterp interpolate quaternions
unit unitize a quaternion
Manipulator Models
link construct a robot link object
nofriction remove friction from a robot object
perturb randomly modify some dynamic parameters
puma560 Puma 560 data
puma560akb Puma 560 data (modified Denavit-Hartenberg)
robot construct a robot object
showlink show link/robot data in detail
stanford Stanford arm data
twolink simple 2-link example
Kinematics
diff2tr differential motion vector to transform
fkine compute forward kinematics
ftrans transform force/moment
ikine compute inverse kinematics
ikine560 compute inverse kinematics for Puma 560 like arm
jacob0 compute Jacobian in base coordinate frame
jacobn compute Jacobian in end-effector coordinate frame
tr2diff homogeneous transform to differential motion vector
tr2jac homogeneous transform to Jacobian
Graphics
drivebot drive a graphical robot
plot plot/animate robot
Dynamics
accel compute forward dynamics
cinertia compute Cartesian manipulator inertia matrix
coriolis compute centripetal/coriolis torque
fdyn forward dynamics (motion given forces)
friction joint friction
gravload compute gravity loading
inertia compute manipulator inertia matrix
itorque compute inertia torque
rne inverse dynamics (forces given motion)
Other
ishomog test if matrix is 4 × 4
maniplty compute manipulability
rtdemo toolbox demonstration
unit unitize a vector
accel
Description Returns a vector of joint accelerations that result from applying the actuator torque to the
manipulator robot with joint coordinates q and velocities qd.
Algorithm Uses the method 1 of Walker and Orin to compute the forward dynamics. This form is
useful for simulation of manipulator dynamics, in conjunction with a numerical integration
function.
References M. W. Walker and D. E. Orin. Efficient dynamic computer simulation of robotic mecha-
nisms. ASME Journal of Dynamic Systems, Measurement and Control, 104:205–211, 1982.
cinertia
Synopsis M = cinertia(robot, q)
Description cinertia computes the Cartesian, or operational space, inertia matrix. robot is a robot
object that describes the manipulator dynamics and kinematics, and q is an n-element vector
of joint coordinates.
Algorithm The Cartesian inertia matrix is calculated from the joint-space inertia matrix by
F = M(x)ẍ
References O. Khatib, “A unified approach for motion and force control of robot manipulators: the
operational space formulation,” IEEE Trans. Robot. Autom., vol. 3, pp. 43–53, Feb. 1987.
coriolis
Description coriolis returns the joint torques due to rigid-body Coriolis and centripetal effects for the
specified joint state q and velocity qd. robot is a robot object that describes the manipulator
dynamics and kinematics.
If q and qd are row vectors, tau c is a row vector of joint torques. If q and qd are matrices,
each row is interpreted as a joint state vector, and tau c is a matrix each row being the
corresponding joint torques.
Algorithm Evaluated from the equations of motion, using rne, with joint acceleration and gravitational
acceleration set to zero,
# = C(q, q̇)q̇
References M. W. Walker and D. E. Orin. Efficient dynamic computer simulation of robotic mecha-
nisms. ASME Journal of Dynamic Systems, Measurement and Control, 104:205–211, 1982.
ctraj
Description ctraj returns a Cartesian trajectory (straight line motion) TC from the point represented by
homogeneous transform T0 to T1. The number of points along the path is m or the length of
the given vector r. For the second case r is a vector of distances along the path (in the range
0 to 1) for each point. The first case has the points equally spaced, but different spacing may
be specified to achieve acceptable acceleration profile. TC is a 4 × 4 × m matrix.
Examples To create a Cartesian path with smooth acceleration we can use the jtraj function to create
the path vector r with continuous derivitives.
1.5
0.5
!0.5
!1
0 1 2 3 4 5 6 7 8 9 10
Time (s)
diff2tr
Description Returns a homogeneous transform representing differential translation and rotation corre-
sponding to Cartesian velocity x = [v x vy vz $x $y $z ].
References R. P. Paul. Robot Manipulators: Mathematics, Programming, and Control. MIT Press,
Cambridge, Massachusetts, 1981.
drivebot
Synopsis drivebot(robot)
drivebot(robot, q)
Description Pops up a window with one slider for each joint. Operation of the sliders will drive the
graphical robot on the screen. Very useful for gaining an understanding of joint limits and
robot workspace.
The joint coordinate state is kept with the graphical robot and can be obtained using the
plot function. If q is specified it is used as the initial joint angle, otherwise the initial value
of joint coordinates is taken from the graphical robot.
eul2tr
Description eul2tr returns a homogeneous transformation for the specified Euler angles in radians.
Cautionary Note that 12 different Euler angle sets or conventions exist. The convention used here is the
common one for robotics, but is not the one used for example in the aerospace community.
fdyn
Description fdyn integrates the manipulator equations of motion over the time interval t0 to t1 us-
ing MATLAB’s ode45 numerical integration function. Manipulator kinematic and dynamic
chacteristics are given by the robot object robot. It returns a time vector t, and matrices of
manipulator joint state q and joint velocities qd. These matrices have one row per time step
and one column per joint.
Actuator torque may be specified by a user function
where t is the current time, and q and qd are the manipulator joint coordinate and velocity
state respectively. Optional arguments passed to fdyn will be passed through to the user
function. Typically this function would be used to implement some axis control scheme
as a function of manipulator state and passed in setpoint information. If torqfun is not
specified then zero torque is applied to the manipulator.
Initial joint coordinates and velocities may be specified by the optional arguments q0 and
qd0 respectively.
Algorithm The joint acceleration is a function of joint coordinate and velocity given by
q̈ = M(q)−1 # − C(q, q̇)q̇ − G(q) − F(q̇)
. /
Example The following example shows how fdyn() can be used to simulate a robot and its controller.
The manipulator is a Puma 560 with simple proportional and derivative controller. The
simulation results are shown in the figure, and further gain tuning is clearly required. Note
that high gains are required on joints 2 and 3 in order to counter the significant disturbance
torque due to gravity.
Note the use of qz a zero vector of length 6 defined by puma560 pads out the two initial condition
arguments, and we place the control gains and the path as optional arguments. Note also the use of
the nofriction() function, see Cautionary note below. The invoked function is
%
% taufunc.m
%
% user written function to compute joint torque as a function
% of joint error. The desired path is passed in via the global
% matrix qt. The controller implemented is PD with the proportional
% and derivative gains given by the global variables Pgain and Dgain
% respectively.
%
function tau = taufunc(t, q, qd, Pgain, Dgain, qt)
0.05
Joint 1 (rad)
!0.05
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (s)
2
Joint 2 (rad)
!1
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (s)
1
Joint 3 (rad)
!1
!2
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (s)
Results of fdyn() example. Simulated path shown as solid, and reference path as dashed.
Cautionary The presence of non-linear friction in the dynamic model can prevent the integration from converging.
The function nofriction() can be used to return a Coulomb friction free robot object.
References M. W. Walker and D. E. Orin. Efficient dynamic computer simulation of robotic mechanisms. ASME
Journal of Dynamic Systems, Measurement and Control, 104:205–211, 1982.
fkine
Synopsis T = fkine(robot, q)
Description fkine computes forward kinematics for the joint coordinate q giving a homogeneous transform for
the location of the end-effector. robot is a robot object which contains a kinematic model in either
standard or modified Denavit-Hartenberg notation. Note that the robot object can specify an arbitrary
homogeneous transform for the base of the robot and a tool offset.
If q is a vector it is interpreted as the generalized joint coordinates, and fkine returns a homogeneous
transformation for the final link of the manipulator. If q is a matrix each row is interpreted as a joint
state vector, and T is a 4 × 4 × m matrix where m is the number of rows in q.
Cautionary Note that the dimensional units for the last column of the T matrix will be the same as the dimensional
units used in the robot object. The units can be whatever you choose (metres, inches, cubits or
furlongs) but the choice will affect the numerical value of the elements in the last column of T. The
Toolbox definitions puma560 and stanford all use SI units with dimensions in metres.
References R. P. Paul. Robot Manipulators: Mathematics, Programming, and Control. MIT Press, Cambridge,
Massachusetts, 1981.
J. J. Craig, Introduction to Robotics. Addison Wesley, second ed., 1989.
link/friction
Description friction computes the joint friction torque based on friction parameter data, if any, in the link
object link. Friction is a function only of joint velocity qd
If qd is a vector then tau f is a vector in which each element is the friction torque for the the
corresponding element in qd.
Algorithm The friction model is a fairly standard one comprising viscous friction and direction dependent
Coulomb friction
Bi q̇ + #− ˙
i , "<0
!
Fi (t) = +
Bi q̇ + #i , "˙ > 0
robot/friction
Description friction computes the joint friction torque vector for the robot object robot with a joint velocity
vector qd.
ftrans
Synopsis F2 = ftrans(F, T)
Description Transform the force vector F in the current coordinate frame to force vector F2 in the second coordi-
nate frame. The second frame is related to the first by the homogeneous transform T. F2 and F are
each 6-element vectors comprising force and moment components [Fx Fy Fz Mx My Mz ].
gravload
Description gravload computes the joint torque due to gravity for the manipulator in pose q.
If q is a row vector, tau g returns a row vector of joint torques. If q is a matrix each row is interpreted
as as a joint state vector, and tau g is a matrix in which each row is the gravity torque for the the
corresponding row in q.
The default gravity direction comes from the robot object but may be overridden by the optional grav
argument.
References M. W. Walker and D. E. Orin. Efficient dynamic computer simulation of robotic mechanisms. ASME
Journal of Dynamic Systems, Measurement and Control, 104:205–211, 1982.
ikine
Synopsis q = ikine(robot, T)
q = ikine(robot, T, q0)
q = ikine(robot, T, q0, M)
Description ikine returns the joint coordinates for the manipulator described by the object robot whose end-
effector homogeneous transform is given by T. Note that the robot’s base can be arbitrarily specified
within the robot object.
If T is a homogeneous transform then a row vector of joint coordinates is returned. The estimate for
the first step is q0 if this is given else 0.
Note that the inverse kinematic solution is generally not unique, and depends on the initial value q0
(which defaults to 0).
For the case of a manipulator with fewer than 6 DOF it is not possible for the end-effector to satisfy
the end-effector pose specified by an arbitrary homogeneous transform. This typically leads to non-
convergence in ikine. A solution is to specify a 6-element weighting vector, M, whose elements
are 0 for those Cartesian DOF that are unconstrained and 1 otherwise. The elements correspond
to translation along the X-, Y- and Z-axes and rotation about the X-, Y- and Z-axes respectively.
For example, a 5-axis manipulator may be incapable of independantly controlling rotation about the
end-effector’s Z-axis. In this case M = [1 1 1 1 1 0] would enable a solution in which the end-
effector adopted the pose T except for the end-effector rotation. The number of non-zero elements
should equal the number of robot DOF.
Algorithm The solution is computed iteratively using the pseudo-inverse of the manipulator Jacobian.
q̇ = J+ (q)& F (q) − T
0 1
where & returns the ‘difference’ between two transforms as a 6-element vector of displacements and
rotations (see tr2diff).
Cautionary Such a solution is completely general, though much less efficient than specific inverse kinematic
solutions derived symbolically.
This approach allows a solution to obtained at a singularity, but the joint coordinates within the null
space are arbitrarily assigned.
Note that the dimensional units used for the last column of the T matrix must agree with the dimen-
sional units used in the robot definition. These units can be whatever you choose (metres, inches,
cubits or furlongs) but they must be consistent. The Toolbox definitions puma560 and stanford
all use SI units with dimensions in metres.
References S. Chieaverini, L. Sciavicco, and B. Siciliano, “Control of robotic systems through singularities,” in
Proc. Int. Workshop on Nonlinear and Adaptive Control: Issues in Robotics (C. C. de Wit, ed.),
Springer-Verlag, 1991.
ikine560
Description ikine560 returns the joint coordinates corresponding to the end-effector homogeneous transform
T. It is computed using a symbolic solution appropriate for Puma 560 like robots, that is, all revolute
6DOF arms, with a spherical wrist. The use of a symbolic solution means that it executes over 50
times faster than ikine for a Puma 560 solution.
A further advantage is that ikine560() allows control over the specific solution returned. config
is a string which contains one or more of the configuration control letter codes
Cautionary Note that the dimensional units used for the last column of the T matrix must agree with the dimen-
sional units used in the robot object. These units can be whatever you choose (metres, inches, cubits
or furlongs) but they must be consistent. The Toolbox definitions puma560 and stanford all use
SI units with dimensions in metres.
References R. P. Paul and H. Zhang, “Computationally efficient kinematics for manipulators with spherical
wrists,” Int. J. Robot. Res., vol. 5, no. 2, pp. 32–44, 1986.
Author Robert Biro and Gary McMurray, Georgia Institute of Technology, [email protected]
inertia
Synopsis M = inertia(robot, q)
Description inertia computes the joint-space inertia matrix which relates joint torque to joint acceleration
# = M(q)q̈
robot is a robot object that describes the manipulator dynamics and kinematics, and q is an n-
element vector of joint state. For an n-axis manipulator M is an n × n symmetric matrix.
If q is a matrix each row is interpreted as a joint state vector, and I is an n × n × m matrix where m is
the number of rows in q.
Note that if the robot contains motor inertia parameters then motor inertia, referred to the link
reference frame, will be added to the diagonal of M.
Example To show how the inertia ‘seen’ by the waist joint varies as a function of joint angles 2 and 3 the
following code could be used.
5.5
4.5
4
I11
3.5
2.5
2
4
2 4
0 2
0
!2
!2
q3 !4 !4
q2
References M. W. Walker and D. E. Orin. Efficient dynamic computer simulation of robotic mechanisms. ASME
Journal of Dynamic Systems, Measurement and Control, 104:205–211, 1982.
ishomog
Synopsis ishomog(x)
itorque
Description itorque returns the joint torque due to inertia at the specified pose q and acceleration qdd which is
given by
#i = M(q)q̈
If q and qdd are row vectors, itorque is a row vector of joint torques. If q and qdd are matrices,
each row is interpreted as a joint state vector, and itorque is a matrix in which each row is the
inertia torque for the corresponding rows of q and qdd.
robot is a robot object that describes the kinematics and dynamics of the manipulator and drive. If
robot contains motor inertia parameters then motor inertia, referred to the link reference frame, will
be included in the diagonal of M and influence the inertia torque result.
jacob0
Synopsis jacob0(robot, q)
Description jacob0 returns a Jacobian matrix for the robot object robot in the pose q and as expressed in the
base coordinate frame.
The manipulator Jacobian matrix, 0 Jq , maps differential velocities in joint space, q̇, to Cartesian
velocity of the end-effector expressed in the base coordinate frame.
0
ẋ = 0 Jq (q)q̇
References R. P. Paul, B. Shimano and G. E. Mayer. Kinematic Control Equations for Simple Manipulators.
IEEE Systems, Man and Cybernetics 11(6), pp 449-455, June 1981.
jacobn
Synopsis jacobn(robot, q)
Description jacobn returns a Jacobian matrix for the robot object robot in the pose q and as expressed in the
end-effector coordinate frame.
The manipulator Jacobian matrix, 0 Jq , maps differential velocities in joint space, q̇, to Cartesian
velocity of the end-effector expressed in the end-effector coordinate frame.
n
ẋ = n Jq (q)q̇
# = n Jq (q)'n F
References R. P. Paul, B. Shimano and G. E. Mayer. Kinematic Control Equations for Simple Manipulators.
IEEE Systems, Man and Cybernetics 11(6), pp 449-455, June 1981.
jtraj
Purpose Compute a joint space trajectory between two joint coordinate poses
Description jtraj returns a joint space trajectory q from joint coordinates q0 to q1. The number of points is n
or the length of the given time vector t. A 7th order polynomial is used with default zero boundary
conditions for velocity and acceleration.
The trajectory is a matrix, with one row per time step, and one column per joint. The function can
optionally return a velocity and acceleration trajectories as qd and qdd respectively.
link
Synopsis L = link
L = link([alpha, a, theta, d], convention)
L = link([alpha, a, theta, d, sigma], convention)
L = link(dyn row, convention)
A = link(q)
show(L)
Description The link function constructs a link object. The object contains kinematic and dynamic parameters
as well as actuator and transmission parameters. The first form returns a default object, while the
second and third forms initialize the kinematic model based on Denavit and Hartenberg parameters.
The dynamic model can be initialized using the fourth form of the constructor where dyn row is a
1 × 20 matrix which is one row of the legacy dyn matrix.
By default the standard Denavit and Hartenberg conventions are assumed but this can be overridden
by the optional convention argument which can be set to either ’modified’ or ’standard’
(default). Note that any abbreviation of the string can be used, ie. ’mod’ or even ’m’.
The second last form given above is not a constructor but a link method that returns the link transfor-
mation matrix for the given joint coordinate. The argument is given to the link object using paren-
thesis. The single argument is taken as the link variable q and substituted for " or D for a revolute or
prismatic link respectively.
The Denavit and Hartenberg parameters describe the spatial relationship between this link and the pre-
vious one. The meaning of the fields for each kinematic convention are summarized in the following
table.
variable DH MDH description
alpha !i !i−1 link twist angle
A Ai Ai−1 link length
theta "i "i link rotation angle
D Di Di link offset distance
sigma (i (i joint type; 0 for revolute, non-zero for prismatic
Since Matlab does not support the concept of public class variables methods have been written to
allow link object parameters to be referenced (r) or assigned (a) as given by the following table
The default is for standard Denavit-Hartenberg conventions, zero friction, mass and inertias.
The display method gives a one-line summary of the link’s kinematic parameters. The show
method displays as many link parameters as have been initialized for that link.
Examples
L =
-1.570796 0.020000 0.000000 0.150000 R (std)
>> show(L)
alpha = -1.5708
A = 0.02
theta = 0
D = 0.15
sigma = 0
mdh = 0
G = 100
Tc = 5 -5
>>
represents each link’s coordinate frame with respect to the previous link’s coordinate system. For a
revolute joint "i is offset by
References R. P. Paul. Robot Manipulators: Mathematics, Programming, and Control. MIT Press, Cambridge,
Massachusetts, 1981.
J. J. Craig, Introduction to Robotics. Addison Wesley, second ed., 1989.
maniplty
Synopsis m = maniplty(robot, q)
m = maniplty(robot, q, which)
Description maniplty computes the scalar manipulability index for the manipulator at the given pose. Manipu-
lability varies from 0 (bad) to 1 (good). robot is a robot object that contains kinematic and optionally
dynamic parameters for the manipulator. Two measures are supported and are selected by the optional
argument which can be either ’yoshikawa’ (default) or ’asada’. Yoshikawa’s manipulability
measure is based purely on kinematic data, and gives an indication of how ‘far’ the manipulator is
from singularities and thus able to move and exert forces uniformly in all directions.
Asada’s manipulability measure utilizes manipulator dynamic data, and indicates how close the inertia
ellipsoid is to spherical.
Algorithm Yoshikawa’s measure is based on the condition number of the manipulator Jacobian
5
)yoshi = |J(q)J(q)' |
x' M(x)x = 1
and gives an indication of how well the manipulator can accelerate in each of the Cartesian directions.
The scalar measure computed here is the ratio of the smallest/largest ellipsoid axes
min x
)asada =
max x
Ideally the ellipsoid would be spherical, giving a ratio of 1, but in practice will be less than 1.
References T. Yoshikawa, “Analysis and control of robot manipulators with redundancy,” in Proc. 1st Int. Symp.
Robotics Research, (Bretton Woods, NH), pp. 735–747, 1983.
robot/nofriction
Description Return a new robot object with modified joint friction properties. The first form sets the Coulomb
friction values to zero in the constituent links The second form sets viscous and Coulomb friction
values in the constituent links are set to zero.
The resulting robot object has its name string prepended with ’NF/’.
This is important for forward dynamics computation (fdyn()) where the presence of friction can
prevent the numerical integration from converging.
link/nofriction
Description Return a new link object with modified joint friction properties. The first form sets the Coulomb
friction values to zero. The second form sets both viscous and Coulomb friction values to zero.
This is important for forward dynamics computation (fdyn()) where the presence of friction can
prevent the numerical integration from converging.
oa2tr
Synopsis oa2tr(o, a)
Description oa2tr returns a rotational homogeneous transformation specified in terms of the Cartesian orienta-
tion and approach vectors o and a respectively.
Algorithm 6 7
ô × â ô â 0
T=
0 0 0 1
where ô and â are unit vectors corresponding to o and a respectively.
perturb
Description Return a new robot object with randomly modified dynamic parameters: link mass and inertia. The
perturbation is multiplicative so that values are multiplied by random numbers in the interval (1-p) to
(1+p).
Useful for investigating the robustness of various model-based control schemes where one model
forms the basis of the model-based controller and the peturbed model is used for the actual plant.
The resulting robot object has its name string prepended with ’P/’.
puma560
Synopsis puma560
Description Creates the robot object p560 which describes the kinematic and dynamic characteristics of a Uni-
mation Puma 560 manipulator. The kinematic conventions used are as per Paul and Zhang, and all
quantities are in standard SI units.
Also defines the joint coordinate vectors qz, qr and qstretch corresponding to the zero-angle,
ready and fully extended (in X-direction) poses respectively.
Details of coordinate frames used for the Puma 560 shown here in its zero angle pose.
References R. P. Paul and H. Zhang, “Computationally efficient kinematics for manipulators with spherical
wrists,” Int. J. Robot. Res., vol. 5, no. 2, pp. 32–44, 1986.
P. Corke and B. Armstrong-Hélouvry, “A search for consensus among model parameters reported for
the PUMA 560 robot,” in Proc. IEEE Int. Conf. Robotics and Automation, (San Diego), pp. 1608–
1613, May 1994.
puma560akb
Synopsis puma560akb
Description Creates the robot object p560m which describes the kinematic and dynamic characteristics of a Uni-
mation Puma 560 manipulator. It uses Craig’s modified Denavit-Hartenberg notation with the partic-
ular kinematic conventions from Armstrong, Khatib and Burdick. All quantities are in standard SI
units.
Also defines the joint coordinate vectors qz, qr and qstretch corresponding to the zero-angle,
ready and fully extended (in X-direction) poses respectively.
References B. Armstrong, O. Khatib, and J. Burdick, “The explicit dynamic model and inertial parameters of
the Puma 560 arm,” in Proc. IEEE Int. Conf. Robotics and Automation, vol. 1, (Washington, USA),
pp. 510–18, 1986.
qinterp
Description Return a unit-quaternion that interpolates between Q1 and Q2 as r varies between 0 and 1 inclusively.
This is a spherical linear interpolation (slerp) that can be interpreted as interpolation along a great
circle arc on a sphere.
>> q1 = quaternion(rotx(0.3))
q1 =
0.98877 <0.14944, 0, 0>
>> q2 = quaternion(roty(-0.5))
q2 =
0.96891 <0, -0.2474, 0>
ans =
0.98877 <0.14944, 0, 0>
ans =
0.96891 <0, -0.2474, 0>
ans =
0.99159 <0.10536, -0.075182, 0>
>>
References K. Shoemake, “Animating rotation with quaternion curves.,” in Proceedings of ACM SIGGRAPH,
(San Francisco), pp. 245–254, The Singer Company, Link Flight Simulator Division, 1985.
quaternion
Synopsis q = quaternion(qq)
q = quaternion(v, theta)
q = quaternion(R)
q = quaternion([s vx vy vz])
Description quaternion is the constructor for a quaternion object. The first form returns a new object with the
same value as its argument. The second form initializes the quaternion to a rotation of theta about
the vector v.
ans =
0.87758 <0.47943, 0, 0>
ans =
0.87758 <0.47943, 0, 0>
>>
The third form sets the quaternion to a rotation equivalent to the given 3 × 3 rotation matrix, or the
rotation submatrix of a 4 × 4 homogeneous transform.
The fourth form sets the four quaternion elements directly where s is the scalar component and [vx
vy vz] the vector.
All forms, except the last, return a unit quaternion, ie. one whose magnitude is unity.
method Returns
quaternion.d return 4-vector of quaternion elements
quaternion.s return scalar component
quaternion.v return vector component
quaternion.t return equivalent homogeneous transformation
matrix
quaternion.r return equivalent orthonormal rotation matrix
Examples
>> t = rotx(0.2)
t =
1.0000 0 0 0
0 0.9801 -0.1987 0
0 0.1987 0.9801 0
0 0 0 1.0000
>> q1 = quaternion(t)
q1 =
0.995 <0.099833, 0, 0>
>> q1.r
ans =
1.0000 0 0
0 0.9801 -0.1987
0 0.1987 0.9801
>> q1 * q2
ans =
>> q1*q1
ans =
0.98007 <0.19867, 0, 0>
>> q1ˆ2
ans =
0.98007 <0.19867, 0, 0>
>> q1*inv(q1)
ans =
1 <0, 0, 0>
>> q1/q1
ans =
1 <0, 0, 0>
>> q1/q2
ans =
0.98383 <0.098712, -0.14869, -0.014919>
>> q1*q2ˆ-1
ans =
0.98383 <0.098712, -0.14869, -0.014919>
Cautionary At the moment vectors or arrays of quaternions are not supported. You can however use cell arrays to
hold a number of quaternions.
References K. Shoemake, “Animating rotation with quaternion curves.,” in Proceedings of ACM SIGGRAPH,
(San Francisco), pp. 245–254, The Singer Company, Link Flight Simulator Division, 1985.
quaternion/plot
Synopsis plot(Q)
Description plot is overloaded for quaternion objects and displays a 3D plot which shows how the standard
axes are transformed under that rotation.
Examples A rotation of 0.3rad about the X axis. Clearly the X axis is invariant under this rotation.
>> q=quaternion(rotx(0.3))
q =
0.85303<0.52185, 0, 0>
>> plot(q)
0.5 Z
X
0
Z
!0.5
!1
1
0.5 1
0.5
0
0
!0.5
!0.5
!1 !1
Y
X
rne
Description rne computes the equations of motion in an efficient manner, giving joint torque as a function of joint
position, velocity and acceleration.
If q, qd and qdd are row vectors then tau is a row vector of joint torques. If q, qd and qdd are
matrices then tau is a matrix in which each row is the joint torque for the corresponding rows of q,
qd and qdd.
Gravity direction is defined by the robot object but may be overridden by providing a gravity acceler-
ation vector grav = [gx gy gz].
An external force/moment acting on the end of the manipulator may also be specified by a 6-element
vector fext = [Fx Fy Fz Mx My Mz] in the end-effector coordinate frame.
The torque computed may contain contributions due to armature inertia and joint friction if these are
specified in the parameter matrix dyn.
The MEX-file version of this function is over 1000 times faster than the M-file. See Section 1 of this
manual for information about how to compile and install the MEX-file.
where M is the manipulator inertia matrix, C is the Coriolis and centripetal torque, F the viscous and
Coulomb friction, and G the gravity load.
Cautionary The MEX file currently ignores support base and tool transforms.
References J. Y. S. Luh, M. W. Walker, and R. P. C. Paul. On-line computational scheme for mechanical manip-
ulators. ASME Journal of Dynamic Systems, Measurement and Control, 102:69–76, 1980.
robot
Synopsis r = robot
r = robot(rr)
r = robot(link ...)
r = robot(DH ...)
r = robot(DYN ...)
Description robot is the constructor for a robot object. The first form creates a default robot, and the second
form returns a new robot object with the same value as its argument. The third form creates a robot
from a cell array of link objects which define the robot’s kinematics and optionally dynamics. The
fourth and fifth forms create a robot object from legacy DH and DYN format matrices.
The last three forms all accept optional trailing string arguments which are taken in order as being
robot name, manufacturer and comment.
Since Matlab does not support the concept of public class variables methods have been written to
allow robot object parameters to be referenced (r) or assigned (a) as given by the following table
Some of these operations at the robot level are actually wrappers around similarly named link object
The offset vector is added to the user specified joint angles before any kinematic or dynamic function is
invoked (it is actually implemented within the link object). Similarly it is subtracted after an operation
such as inverse kinematics. The need for a joint offset vector arises because of the constraints of
the Denavit-Hartenberg (or modified Denavit-Hartenberg) notation. The pose of the robot with zero
joint angles is frequently some rather unusual (or even unachievable) pose. The joint coordinate offset
provides a means to make an arbitrary pose correspond to the zero joint angle case.
robot.name ’noname’
robot.manuf ”
robot.comment ”
robot.gravity [0 0 9.81] m/s2
robot.offset ones(n,1)
robot.base eye(4,4)
robot.tool eye(4,4)
robot.lineopt ’Color’, ’black’, ’Linewidth’, 4
robot.shadowopt ’Color’, ’black’, ’Linewidth’, 1
The multiplication operator, *, is overloaded and the product of two robot objects is a robot which is
the series connection of the multiplicands. Tool transforms of all but the last robot are ignored, base
transform of all but the first robot are ignored.
The plot function is also overloaded and is used to provide a robot animation.
Examples
>> r = robot(L)
r =
(2 axis, RR)
grav = [0.00 0.00 9.81]
standard D&H parameters
>>
robot/plot
Synopsis plot(robot, q)
plot(robot, q, arguments...)
z
y x
Puma 560
0.8
0.6
Description plot is overloaded for robot objects and displays a graphical representation of the robot given the
kinematic information in robot. The robot is represented by a simple stick figure polyline where
line segments join the origins of the link coordinate frames. If q is a matrix representing a joint-space
trajectory then an animation of the robot motion is shown.
GRAPHICAL ANNOTATIONS
The basic stick figure robot can be annotated with
All of these require some kind of dimension and this is determined using a simple heuristic from
the workspace dimensions. This dimension can be changed by setting the multiplicative scale factor
using the mag option below. These various annotations do slow the rate at which animations will be
rendered.
OPTIONS
Options are specified by a variable length argument list comprising strings and numeric values. The
allowed values are:
workspace w set the 3D plot bounds or workspace to the matrix [xmin xmax ymin ymax
zmin zmax]
perspective show a perspective view
ortho show an orthogonal view
base, nobase control display of base, a line from the floor upto joint 0
wrist, nowrist control display of wrist axes
name, noname control display of robot name near joint 0
shadow, noshadow control display of a ’shadow’ on the floor
joints, nojoints control display of joints, these are cylinders for revolute joints and boxes for
prismatic joints
xyz wrist axis labels are X, Y, Z
noa wrist axis labels are N, O, A
mag scale annotation scale factor
erase, noerase control erasure of robot after each change
loop, noloop control whether animation is repeated endlessly
The options come from 3 sources and are processed in the order:
1. Cell array of options returned by the function PLOTBOTOPT if found on the user’s current
path.
2. Cell array of options returned by the .plotopt method of the robot object. These are set
by the .plotopt method.
3. List of arguments in the command line.
This state is used, and adjusted, by the drivebot function. The current joint angle state can be
obtained by q = plot(robot). If multiple instances exist, that of the first one returned by find-
obj() is given.
>> clf
>> p560b = p560; % duplicate the robot
>> p560b.name = ’Another Puma 560’; % give it a unique name
>> p560b.base = transl([-.05 0.5 0]); % move its base
>> clf
>> figure % create a new figure
>> plot(p560, qz); % add a graphical robot
>> figure % create another figure
>> plot(p560, qz); % add a graphical robot
>> plot(p560, qr); % both robots should move
Now the two figures can be adjusted to give different viewpoints, for instance, plan and elevation.
Cautionary plot() options are only processed on the first call when the graphical object is established, they are
skipped on subsequent calls. Thus if you wish to change options, clear the figure before replotting.
rotvec
Description rotvec returns a homogeneous transformation representing a rotation of theta radians about the
vector v.
rotx,roty,rotz
Synopsis T = rotx(theta)
T = roty(theta)
T = rotz(theta)
Description Return a homogeneous transformation representing a rotation of theta radians about the X, Y or Z
axes.
rpy2tr
Description rpy2tr returns a homogeneous transformation for the specified roll/pitch/yaw angles in radians.
References R. P. Paul, Robot Manipulators: Mathematics, Programming, and Control. Cambridge, Mas-
sachusetts: MIT Press, 1981.
rtdemo
Synopsis rtdemo
Description This script provides demonstrations for most functions within the Robotics Toolbox.
Cautionary This script clears all variables in the workspace and deletes all figures.
showlink
Synopsis showlink(robot)
showlink(link)
Description Displays in detail all the parameters, including all defined inertial parameters, of a link. The first form
provides this level of detail for all links in the specified manipulator. roll/pitch/yaw angles in radians.
>> showlink(p560.link{2})
alpha = 0
A = 0.4318
theta = 0
D = 0
sigma = 0
mdh = 0
offset = 0
m = 17.4
r = -0.3638
0.006
0.2275
I = 0.13 0 0
0 0.524 0
0 0 0.539
Jm = 0.0002
G = 107.815
B = 0.000817
Tc = 0.126 -0.071
qlim =
>>
stanford
Synopsis stanford
0
Z
y
Stanford arm
z x
!1
!2
2
1 2
0 1
0
!1
!1
Y !2 !2
X
Description Creates the robot object stan which describes the kinematic and dynamic characteristics of a Stan-
ford manipulator. Specifies armature inertia and gear ratios. All quantities are in standard SI units.
References R. Paul, “Modeling, trajectory calculation and servoing of a computer controlled arm,” Tech. Rep.
AIM-177, Stanford University, Artificial Intelligence Laboratory, 1972.
tr2diff
Synopsis d = tr2diff(T)
d = tr2diff(T1, T2)
Description The first form of tr2diff returns a 6-element differential motion vector representing the incremental
translation and rotation described by the homogeneous transform T. It is assumed that T is of the form
0 −*z *y dx
*z 0 −*x dy
−*y *x 0 dz
0 0 0 0
The translational elements of d are assigned directly. The rotational elements are computed from the
mean of the two values that appear in the skew-symmetric matrix.
The second form of tr2diff returns a 6-element differential motion vector representing the dis-
placement from T1 to T2, that is, T2 - T1.
6 7
p2 − p1
d=
1/2 (n1 × n2 + o1 × o2 + a1 × a2 )
References R. P. Paul, Robot Manipulators: Mathematics, Programming, and Control. Cambridge, Mas-
sachusetts: MIT Press, 1981.
tr2eul
Synopsis [a b c] = tr2eul(T)
Description tr2eul returns a vector of Euler angles, in radians, corresponding to the rotational part of the homo-
geneous transform T.
Trot = RZ (a)RY (b)RZ (c)
Cautionary Note that 12 different Euler angle sets or conventions exist. The convention used here is the common
one for robotics, but is not the one used for example in the aerospace community.
References R. P. Paul, Robot Manipulators: Mathematics, Programming, and Control. Cambridge, Mas-
sachusetts: MIT Press, 1981.
tr2jac
Description tr2jac returns a 6 × 6 Jacobian matrix to map differential motions or velocities between frames
related by the homogeneous transform T.
References R. P. Paul, Robot Manipulators: Mathematics, Programming, and Control. Cambridge, Mas-
sachusetts: MIT Press, 1981.
tr2rpy
Synopsis [a b c] = tr2rpy(T)
Description tr2rpy returns a vector of roll/pitch/yaw angles, in radians, corresponding to the rotational part of
the homogeneous transform T
Trot = RZ (r)RY (p)RX (y)
References R. P. Paul, Robot Manipulators: Mathematics, Programming, and Control. Cambridge, Mas-
sachusetts: MIT Press, 1981.
transl
Synopsis T = transl(x, y, z)
T = transl(v)
v = transl(T)
xyz = transl(TC)
Description The first two forms return a homogeneous transformation representing a translation expressed as three
scalar x, y and z, or a Cartesian vector v.
The third form returns the translational part of a homogeneous transform as a 3-element column vector.
The fourth form returns a matrix whose columns are the X, Y and Z columns of the 4×4×m Cartesian
trajectory matrix TC.
trinterp
Description trinterp interpolates between the two homogeneous transforms T0 and T1 as r varies between 0
and 1 inclusively. This is generally used for computing straight line or ‘Cartesian’ motion. Rotational
interpolation is achieved using quaternion spherical linear interpolation.
>> t1=rotx(.2)
t1 =
1.0000 0 0 0
0 0.9801 -0.1987 0
0 0.1987 0.9801 0
0 0 0 1.0000
>> t2=transl(1,4,5)*roty(0.3)
t2 =
0.9553 0 0.2955 1.0000
0 1.0000 0 4.0000
-0.2955 0 0.9553 5.0000
0 0 0 1.0000
ans =
1.0000 0 0 0
0 0.9801 -0.1987 0
0 0.1987 0.9801 0
0 0 0 1.0000
ans =
0.9553 0 0.2955 1.0000
0 1.0000 0 4.0000
-0.2955 0 0.9553 5.0000
0 0 0 1.0000
ans =
0.9887 0.0075 0.1494 0.5000
0.0075 0.9950 -0.0998 2.0000
-0.1494 0.0998 0.9837 2.5000
0 0 0 1.0000
>>
References R. P. Paul, Robot Manipulators: Mathematics, Programming, and Control. Cambridge, Mas-
sachusetts: MIT Press, 1981.
trnorm
Synopsis TN = trnorm(T)
Description Returns a normalized copy of the homogeneous transformation T. Finite word length arithmetic can
lead to homogeneous transformations in which the rotational submatrix is not orthogonal, that is,
det(R) )= −1.
References J. Funda, “Quaternions and homogeneous transforms in robotics,” Master’s thesis, University of Penn-
sylvania, Apr. 1988.
twolink
Purpose Load kinematic and dynamic data for a simple 2-link mechanism
Synopsis twolink
yz x
0
Z
!1
!2
2
1 2
0 1
0
!1
!1
Y !2 !2
X
Description Creates the robot object tl which describes the kinematic and dynamic characteristics of a simple
two-link planar manipulator. The manipulator operates in the horizontal (XY) plane and is therefore
not influenced by gravity.
Mass is assumed to be concentrated at the joints. All masses and lengths are unity.
References Fig 3-6 of “Robot Dynamics and Control” by M.W. Spong and M. Vidyasagar, 1989.
unit
Synopsis vn = unit(v)
Algorithm
v
vn =
||v||
dh (legacy)
Description A dh matrix describes the kinematics of a manipulator in a general way using the standard Denavit-
Hartenberg conventions. Each row represents one link of the manipulator and the columns are assigned
according to the following table.
If the last column is not given, toolbox functions assume that the manipulator is all-revolute. For an
n-axis manipulator dh is an n × 4 or n × 5 matrix.
The first 5 columns of a dyn matrix contain the kinematic parameters and maybe used anywhere that
a dh kinematic matrix is required — the dynamic data is ignored.
Lengths Ai and Di may be expressed in any unit, and this choice will flow on to the units in which
homogeneous transforms and Jacobians are represented. All angles are in radians.
References R. P. Paul, Robot Manipulators: Mathematics, Programming, and Control. Cambridge, Mas-
sachusetts: MIT Press, 1981.
dyn (legacy)
Description A dyn matrix describes the kinematics and dynamics of a manipulator in a general way using the
standard Denavit-Hartenberg conventions. Each row represents one link of the manipulator and the
columns are assigned according to the following table.
For an n-axis manipulator, dyn is an n × 20 matrix. The first 5 columns of a dyn matrix contain
the kinematic parameters and maybe used anywhere that a dh kinematic matrix is required — the
dynamic data is ignored.
All angles are in radians. The choice of all other units is up to the user, and this choice will flow on to
the units in which homogeneous transforms, Jacobians, inertias and torques are represented.
See Also dh