Background

Kinematics – the geometric behavior of the robot.

Robot modeling – the dynamics Dynamics – full consideration of the forces / torques
necessary to produce the motion.

Lecture 3
Mikael Norrlöf

PhDCourse Robot modeling and control

Lecture 3

Up till now Why are the dynamic models so important?

§ Lecture 1 § Important in the manipulator design

Ÿ Introduction Ÿ Virtual prototyping
Ÿ Rigid body motion Ÿ Life time estimation
Ÿ Representation of rotation Ÿ Design optimization
Ÿ Homogenous transformation § Fundamental for control design
Ÿ Identification
§ Lecture 2 Ÿ Model based control
Ÿ Kinematics § “Must have” in optimal trajectory planning
• Position
§ …
• Jacobians
• DH parameterization

PhDCourse Robot modeling and control PhDCourse Robot modeling and control
Lecture 3 Lecture 3
 Systematic ways to derive the dynamic equations Limitations

§ Analytical mechanics § Consider open chain robot structures

Ÿ Lagrange’s equation Ÿ This constraint can be relaxed …
Ÿ Newton – Euler iterative technique § Actuator dynamics are neglected (e.g., motors are
Ÿ Kane’s method assumed to be ideal, torque reference in – torque
Ÿ … out)
Joseph Louis Lagrange-
Isaac Newton
§ Graphic / Component modeling 1643 - 1727 Giuseppe Luigi Lagrange
1736 - 1813 Ÿ Can also be relaxed with an additional modeling effort
Ÿ Modelica Available
torque
Ÿ SimMechanics
Ÿ …
§ FEM – modeling
§ …
Sir William Rowan Leonhard Euler
Hamilton 1707 - 1783
Motor
1805 - 1865
speed
PhDCourse Robot modeling and control PhDCourse Robot modeling and control
Lecture 3 Lecture 3

Basic assumptions Constraint forces

Consider a system of n particles. Newton’s second law, § The constraints form a smooth surface in R3k
Fi = mi r&&i , ri Î R 3 ,i = 1,...,k § Constraint forces act to keep the system velocity tangent to
this surface – hence they are normal to the surface
The particles are connected. Introduce constraints Ÿ The constraint forces do not produce any work!
§ The system equations can be written as
g j ( r1 ,...,rk ) = 0 j = 1,...,l
æ m1I 0 öæ r&&i ö
ç ÷ç ÷ l
Holonomic constraint: Algebraic relation between positions. F =ç O ÷ç M ÷ + å Gj l j , g j ( r1 ,...,rk ) = 0 j = 1,...,l
ç 0 mk I ÷øçè r&&k ÷ø
j =1
è
Non holonomic constraints …
where G1, ..., Gk Î R3k are a basis for the constraint forces and
lj are scale factors.

Gj can be chosen as gradient of the constraints gj.

PhDCourse Robot modeling and control PhDCourse Robot modeling and control
Lecture 3 Lecture 3
 Better system representation Virtual work

§ For a system of k particles with l constraints, find Principal of virtual work: “The work done by external forces
n = 3k – l variables q1, …, qn and functions f1, …, fk
corresponding to any set of virtual displacements is zero.”
ri = f i ( q1 ,..., qn ) g j ( r1 ,..., rk ) = 0 (Spong etal, p247)
i =1,..., k j = 1,...,l
fi
qi are called generalized coordinates. k
dr1
åF i
T
dri = 0
§ Generalized forces are forces acting along the generalized r1 å F (ri ) =Fi = 0 i =1

å ( f ) dr = 0
coordinates. k
f 1( a ) (a) T
i i
Example: Robot manipulator with rotational joints. The i =1
generalized forces are torques acting around the joints. l
k

åf i
T
dri = 0
§ The dynamic equations can be expressed in terms of the new i =1
variables.
f 2( a ) r2
dr2

PhDCourse Robot modeling and control PhDCourse Robot modeling and control
Lecture 3 Lecture 3

Dynamic case (D’Alembert’s principle) Dynamic case (D’Alembert’s principle)

k k

åf
i =1
i
T
dri - å p&
i =1
i
T
dri = 0
“if one introduces a
fictitious additional force - p& i
¶ri
k k n n
fi on particle i for each i,
where pi is the
fi
å
i =1
f i T dri = åå
i =1 j = 1
fiT
¶q j å
dq j = y j dq j
j =1
dr1 momentum of the particle, dr1
r1 å F (r ) =F = 0 then each particle will be r1 å F (r ) =F = 0 k
¶ri
åf
i i i i
in equilibrium” yj = i
T

f 1( a ) f 1( a ) i =1
¶q j

l l

f 2( a ) r2 f 2( a ) r2
dr2 dr2

PhDCourse Robot modeling and control PhDCourse Robot modeling and control
Lecture 3 Lecture 3
 Dynamic case (D’Alembert’s principle) Dynamic case (D’Alembert’s principle)
k
¶ri å2mv
k k n 1
K= T
å p& i dri = åå mi &r&i dq j vi
T T i i

i =1 i =1 j =1 ¶q j i =1

k ì
k
¶ri ïd é ¶v ù ¶v üï
fi
å
k
m &
r&
¶riT
= å
k ì
ï d é T ¶ri ù
í ê m &
r ú - m &
r T d
é ¶ri ù üï
ê úý
fi
å mi &r&iT
i =1
= å í êmi viT i ú - mi viT i ý
¶q j i =1 ïî dt ëê ¶q& j ûú ¶q j ïþ
¶q j i =1 ïî dt ëê ¶q j ûú dt êë ¶q j ûú ïþ
i i i i i i
dr1 i =1 dr1
r1 å F (r ) =F = 0 r1 å F (r ) =F = 0 d ¶K ¶K
i i
¶r
n
¶v ¶r
i i
= -
vi = r&i = å i q& j Þ i = i dt ¶q& j ¶q j
j =1 ¶q j ¶q& j ¶q j
f 1( a ) f 1( a )

n ì
l d é ¶ri ù n ¶ 2 ri ¶ n
¶ri ¶v l k
ï d ¶K ¶K üï
ú=å q&l = å q&l = i å d = å - ýdq j
T
ê &
p r í
dt êë ¶q j úû l =1 ¶q j ¶ql ¶q j l =1 ¶ql ¶q j i =1
i i
j =1 ï
î dt ¶q& j ¶q j ïþ
k ì
k
¶ri ïd é ¶v ù ¶v üï ìï d ¶K üï
å mi &r&iT = å í êmi viT i ú - mi viT i ý ¶K
n

å íïî dt ¶q&
f 2( a ) r2 f 2( a ) r2
- - y j ýdq j = 0
dr2 i =1 ¶q j i =1 ïî dt êë ¶q& j úû ¶q j ïþ dr2
j =1 j ¶q j ïþ
PhDCourse Robot modeling and control PhDCourse Robot modeling and control
Lecture 3 Lecture 3

Lagrangian and Lagrange’s equation Hamilton’s principle

§ The Lagrangian is defined as Hamilton's principle states that the true evolution of a
system described by m generalized coordinates
L( q ,q& ) =K ( q , q& ) - P( q ) between two specified states and at two specified
Kinetic energy Potential energy times t1 and t2 is an extremum of the action functional
S ( q )= ò L( q ,q& )dt , ¶S ( q )= 0
t
2

§ Lagrange’s equation
t1 ¶q
d ¶L( q , q& ) ¶L( q , q& ) Trajectory q(t) is a stationary point of S. Assume e is a
- = t i , i = 1,..., n
dt ¶q&i ¶qi perturbation (0 at t1 and t2)
t2 t2
¶L ¶L
¶S = ò L( q + e ,q& + e& ) - L( q ,q& )dt = ò e + e& dt
§ Newton’s law in generalized coordinates t1 t1 ¶q ¶q&
t2
d ¶L( q , q& ) ¶L( q , q& ) é ¶L ù æ ¶L d ¶L ö
t2

= +t d
(momentum) = applied force = êe ú + ò e çç - ÷÷ dt
dt ¶q& ¶q dt ë ¶q& û t1 t1 è ¶q dt ¶q& ø
1424 3
=0

PhDCourse Robot modeling and control PhDCourse Robot modeling and control
Lecture 3 Lecture 3
 Example Kinetic and potential energy

§ Kinetic energy for body B

1 T 1
K= mr& r& + w T Iw c
2 2 B
z
m is the mass and I is the inertia r
tensor.
0
§ Inertia tensor y
Ÿ 3x3 matrix
x
Ÿ Symmetric
Ÿ Positive definite
Ÿ Constant in a body fixed coordinate
system

PhDCourse Robot modeling and control PhDCourse Robot modeling and control
Lecture 3 Lecture 3

Inertia tensor Inertia tensor

§ Rotational part of K § Computation of the inertia

Ÿ w expressed in inertial tensor. r(x,y,z) is the mass
frame c B density as a function of c B
z z
Ÿ With Ib the inertia tensor in position.
r r
body fixed coordinate
system 0
y é I xx I xy I xz ù 0
y
w Iw = w T RI b R T w I = êê I yx I yz úú
T
I yy
x êë I zx I zx I zz úû x

PhDCourse Robot modeling and control PhDCourse Robot modeling and control
Lecture 3 Lecture 3
 Example: Uniform rectangular solid Kinetic energy for an n-link manipulator

Body frame attached to center of gravity. From lecture 2 we know

where vi and wi can be for any point on the manipulator

(depends on the Jacobian).
The kinetic energy can now be expressed as

Notice that
rabc = m where D(q) is the inertia matrix.
Inertia of two bodies expressed in the same
coordinate frame can be added (subtracted) Properties: Symmetric and positive definite.

PhDCourse Robot modeling and control PhDCourse Robot modeling and control
Lecture 3 Lecture 3

Potential energy for an n-link manipulator Dynamic equations

In rigid dynamics, gravity is the only source of Lagrangian L = K - P = 12 å d ij ( q )q&i q& j - P( q )

i, j
potential energy.
n n Recall: Lagrange’s equation
P = å Pi = å mi g T rci
i =1 i =1 d ¶L( q , q& ) ¶L( q , q& )
- = t k , k = 1,..., n
dt ¶q& k ¶q k
If the robot contains elastic components the energy In terms of L above this gives
stored in the elasticities has to be included in the
¶L d ¶L ¶d kj
potential energy. = å d kj q& j Þ = å d kj q&& j + å q&i q& j
¶q& k j dt ¶q& k j i , j ¶qi

¶L 1 ¶d ij ¶P
= 2å q&i q& j -
¶qk i , j ¶q k ¶qk

PhDCourse Robot modeling and control PhDCourse Robot modeling and control
Lecture 3 Lecture 3
 Dynamic equations, cont’d Dynamic equations, cont’d

ì ¶dkj 1 ¶dij ü ¶P In matrix form

å dkj q&&j + å í -2 ýq&i q& j + =tk
i , j î ¶qi ¶qk þ ¶qk
j
D( q )q&& + C( q ,q& )q& + g ( q ) = t

æ x ö æqö
ìï ¶d kj ¶d ki ¶d ij üï ¶P State space representation x& = f ( x ,u ), x = çç 1 ÷÷ = çç ÷÷
Þ å d kj q&& j + å í 1
+ - ýq&i q& j + =tk è x2 ø è q& ø
i , j ï ¶qi ¶ ¶ ï ¶
2
j
î q j q k þ qk
1444 424444 3
æ x&1 ö æ x2 ö
cijk
çç ÷÷ = çç -1 ÷÷
cijk = c jik è x& 2 ø è D ( x1 )(- C( x1 , x2 )x2 - g ( x1 ) + u )ø
Dynamic equations

å d kj q&& j + å cijk q&i q& j + g k = t k Extensions. Friction and joint flexibilities

j i,j
D( qa )q&&a + C( qa ,q& a )q& a + g( qa ) + fq& a = k ( qa - qm ) + d ( qa - qm )
Christoffel gravity Mq&&m + f m q& m = - k ( qa - qm ) - d ( qa - qm ) + t
symbols
PhDCourse Robot modeling and control PhDCourse Robot modeling and control
Lecture 3 Lecture 3

Some properties Newton-Euler

The following matrix is skew symmetric § Different approach to find the dynamic equations
§ A more local approach
N ( q ,q& ) = D& ( q ) - 2C( q ,q& )
Ÿ Each link is modeled separately
Ÿ Links interconnected, leads to a forward – backward iteration
The system is passive.
scheme

The inertia matrix is bounded (constant lower/upper

bound when only revolute joints)
lm I nxn £ D( q ) £ lM I nxn < ¥
Figure 7.12 form Spong etal.

Balance equations:

PhDCourse Robot modeling and control PhDCourse Robot modeling and control
Lecture 3 Lecture 3
 Newton-Euler, cont’d Newton-Euler, cont’d

Figure 7.12 form Spong etal.

Figure 7.12 form Spong etal.

Solve from i = 0 to n
Balance equations:

§ Find fi and ti by solving the equations from fn+1= 0 and

tn+1 = 0.

PhDCourse Robot modeling and control PhDCourse Robot modeling and control
Lecture 3 Lecture 3

Newton-Euler vs Langrangian Home assignment, part I

§ From a blue-print of a robot (IRB1600-8/1.45) find a
Ÿ Kinematic model
§ Solves the same problem. Ÿ Assume the robot to consist of hollow uniform rectangular beams
§ Lagrangian technique gives the dynamic equations made out of metal (steel, aluminum, iron, …)
Ÿ Compute the inertia matrix for each link
“directly”. Ÿ Derive a dynamic model using for example Lagrange’s equation
§ Newton-Euler gives all torques / forces, not just the
generalized torques. Will be very important later … § The dynamic model can be restricted to 3-DOF while the
kinematics shall be derived for a full 6-DOF manipulator.
§ … § Inverse kinematic should be implemented (can be numerical)
§ Include gear-box in the model (gear ratio [-100 100 100 -60 -60
40]:1), motor inertia can be assumed to be 50 – 100 % link
inertia when transformed to the arm-side, i.e., after the gear-
box)
§ Motor torque max, [6 10 5 0.6 0.6 0.5] Nm
DEAD-LINE: JAN 24

PhDCourse Robot modeling and control PhDCourse Robot modeling and control
Lecture 3 Lecture 3
 The robot in the assignment Kinematic data (pdf-files on the homepage)

A small/mid size robot

Main applications
• Arc Welding
• Machine tending
• Material handling
• Continuous processes

PhDCourse Robot modeling and control PhDCourse Robot modeling and control
Lecture 3 Lecture 3