ME1035 ADVANCED ROBOTICS

Independent Joint Control

Jee-Hwan Ryu

School of Mechanical Engineering

Korea University of Technology and Education

Introduction to Control
Control: determining the time history of joint inputs to do a
commanded motion
Control methods are depend on hardware and application
z Cartesian vs. Elbow
z Motor with gear reduction vs. High torque motor without gear
z Continuous path vs. P-to-P
Control methods have been advanced with the
development of complicated hardware
z The more complicated hardware, the more advanced control
methods

Korea University of Technology and Education

Independent Joint Control
Each Axis -> SISO
Coupling effect -> disturbance
Objectives: tracking and disturbance rejection

Korea University of Technology and Education

Actuator Dynamics
DC Motor: Simple and easy to use

F = i
m = K1ia
Vb = K 2m

Korea University of Technology and Education

Actuator Dynamics

dia
L + Ria = V Vb
dt
m = K1ia = K mia
d m
Vb = K 2m = K bm = K b
dt
Korea University of Technology and Education

Torque Speed Relation

Korea University of Technology and Education

Torque Constant
When motor is stalled

R 0
Vr = Ria =
Km
R 0
Km =
Vr

Korea University of Technology and Education

Independent Joint Model

d 2 m d m
Jm + Bm = m l / r
dt 2 dt
= K mia l / r

Korea University of Technology and Education

Independent Joint Model

(Ls + R )I a (s ) = V (s ) K b s m (s )
(J m s 2 + Bm s )m (s ) = K m I a (s ) l (s ) / r

Korea University of Technology and Education

Independent Joint Motion

m (s ) Km
= , with l = 0
V (s ) s[(Ls + R )( J m s + Bm ) + K b K m ]
m (s ) (Ls + R ) / r
= , with V = 0
l (s ) s[(Ls + R )( J m s + Bm ) + K b K m ]

Effect of the load torque (disturbance) is reduced by the gear reduction

m (s ) Km / R
=
L J V (s ) s[J m s + Bm + K b K m / R ]
<< m
R Bm m (s ) 1/ r
=
l (s ) s[J m s + Bm + K b K m / R ]

Electrical time constant << Mechanical time constant

Korea University of Technology and Education

Independent Joint Motion

J m&&m (t ) + (Bm + K b K m / R )&m (t ) = (K m / R )V (t ) l (t ) / r

J&&(t ) + B&(t ) = u (t ) d (t )

B: effective damping

Korea University of Technology and Education

PD Compensator for Set Point Tracking

Set point tracking: tracking a constant or step reference

KP + KDs d
(s ) = (s ) D(s )
1
(s ) (s )
(s ) = Js 2 + (B + K D )s + K P

Korea University of Technology and Education

PD Compensator

E (s ) = d (s ) (s )
Js 2 + Bs d
(s ) + D (s )
1
=
(s ) (s )
For a step reference input and a constant disturbance

d
(s ) = , D (s ) =
d D
s s

ess = lim sE (s ) =
D
s 0 KP
Larger gear reduction and large P-gain can reduce the steady-state error

Korea University of Technology and Education

PD Compensator
Closed-loop characteristic polynomial

s2 +
(B + K D ) s + K P = s 2 + 2s + 2
J J
K P = 2 J , K D = 2J B

For robotic applications, critically damped, fastest

nonoscillatory response
=1

determined the speed of response

Korea University of Technology and Education

Example 6.1

Korea University of Technology and Education

Example 6.1

Korea University of Technology and Education

Example 6.2

Korea University of Technology and Education

PID Compensator

(s ) =
(K D )
s2 + KPs + KI d
(s ) +
s
D(s )
Rouths criteria
2 (s ) 2 (s )
KI <
(B + K D )K P
2 = Js + (B + K D )s + K P s + K I
3 2
J

Korea University of Technology and Education

PID Compensator

Korea University of Technology and Education

Design rule-of-thumb for PID Compensator

First, set K_I =0;
Desing PD gain to achieve the desired transient behavior
z Rise time, settling time, etc)
Design K_I within the limits

Korea University of Technology and Education

The Effect of Saturation and Flexibility
In theory, arbitrary fast response and arbitrary small steady
state error to a constant disturbance can be achieved by
simply increasing the gains
In practice, however, there is a maximum speed of
response achievable from the system
Two major factors
z Saturation
z Joint flexibility

Korea University of Technology and Education

Saturation

Korea University of Technology and Education

Flexibility
Should avoid resonant frequency
Can not increase w arbitrary

Korea University of Technology and Education

Feedforward Control
To track time-varying trajectories

q(s ) c (s ) a(s ) Y (s ) q(s )(c(s )b(s ) + a(s )d (s ))

G (s ) = , H (s ) = , F (s ) = T (s ) = =
p (s ) d (s ) b(s ) R(s ) b(s )( p(s )d (s ) + q(s )c(s ))

Feedforward system and closed-loop system should be stable

Korea University of Technology and Education

Feedforward Control

F (s ) = 1 / G (s )

R (s ) = Y (s )

Forward plant is stable -> system is minimum phase

Korea University of Technology and Education

Feedforward Control

q(s )d (s )
E (s ) = D(s )
p(s )d (s ) + q(s )c(s )

Korea University of Technology and Education

Feedforward Control

Differentiation of a actual signal is not required

Independent of the reference trajectory
With PID, steady-state error to a step disturbance is zero

( ) (
V (t ) = J&&d + B& d + K D d + K P d )
= f (t ) + K D e&(t ) + K P e(t )

Korea University of Technology and Education

Drive Train Dynamics

Popular for use in robots due to low backlash, high torque
transmission, compact
Joint flexibility is significant

Korea University of Technology and Education

Harmonic Drive
The Flexspline has two less teeth than the Circular Spline
The gear ratio is calculated by {#Flexspline Teeth} /
{#Flexspline Teeth - #Circular Spline Teeth}.

Korea University of Technology and Education

Drive Train Dynamics

Flexibility is the limiting factor to the achievable performance
in many cases

J l&&l + Bl&l + k ( l m ) = 0
J && + B & k ( ) = u
m m m m l m

Korea University of Technology and Education

Drive Train Dynamics

pl (s )l (s ) = k m (s ) pl (s ) = J l s 2 + Bl s + k
pm (s ) m (s ) = kl (s ) + U (s ) pm (s ) = J m s 2 + Bm s + k

l (s ) k
=
U (s ) pl (s ) pm (s ) k 2

Korea University of Technology and Education

Drive Train Dynamics

In practice, the stiffness of harmonic drive is large and the
damping is small
Neglect damping

J l J m s 4 + k ( J l + J m )s 2
Frequency of imaginary poles increases with increasing joint
stiffness
Difficult to Control

Korea University of Technology and Education

Drive Train Dynamics

Stable, but undesirable oscillation

Korea University of Technology and Education

Drive Train Dynamics

Korea University of Technology and Education

State Space Design

x1 = l x2 = &l l (s ) T
G (s ) = = c (sI A) b
1

x = x = &
3 m 4 m
U (s )
x& = Ax + bu Poles of the G(s) are
eigenvalues of the matrix A
0 1 0 0 0
k B k 0
J l
Jl Jl
0

A= l , b = 0
0 0 0 1
1
k 0
k Bm
J
J m Jm J m m

y = cT x
cT = [1 0 0 0]

Korea University of Technology and Education

State Feedback Control

4
u (t ) = k x + r = ki xi + r
T

i =1

Compare with previous PD/PID ?

( )
x& = A bk T x + br

More free parameters

Korea University of Technology and Education

Controllability
Definition 6.1: A linear system is said to be completely controllable, or
controllable for short, if for each initial state x(t_0) and each final state
x(t_f) there is a control input u(t) that transfer the system from x(t_0) at
time t_0 to x(t_f) at time t_f.
Lemma 6.1: A linear system of the form (6.50) is controllable if and only
if
det b [ Ab A2b L An 1b 0 ]
Theorem 1: Let (s ) = s n + n s n 1 + L + 2 s + 1 be an arbitrary
polynomial of degree n with real coefficients. Then there exists a state
feedback control law of the form Eq. (6.55) such that

( )
det sI A + bk T = (s )
if and only if the system (6.50) is controllable.
We may achieve arbitrary closed-loop poles using state feedback

Korea University of Technology and Education

Pole Assignment
How to choose an appropriate set of closed-loop poles
based on the desired performance, the limits on the available
torque, etc.
Optimal Control

{ }
J = xT (t )Qx(t ) + Ru 2 (t ) dt
0

u = k T x
1
k = bT P
R
1 T
AT P + PA Pb bP + Q = 0
R

Korea University of Technology and Education

Observer
Control law must be a function of all of the states
Observer: dynamical system (constructed in software),
attempts to estimate the full state using the system model
and output.

(
x& = Ax + bu + l y cT x )
Assumptions: given the system model, dont know the
initial condition
e(t ) = x x
(
e& = A lcT e )
Observability: the eignevalue of (A lc ) can be assigned
T

arbitrary

Korea University of Technology and Education

Observability
Definition 6.2 A linear system is completely observable, or observable
for short, if every initial state x(t_0) can be exactly determined from
measurements of the output y(t) and the input u(t) in a finite time
interval.
Theorem 2 the pair (A,c) is observable if and only if

det c[ AT c L AT
n 1
c 0 ]

Korea University of Technology and Education

Seperation Principle

x& = Ax + bu x& A bk T bk T x
e& =
u = k T x 0 A lc T e

Allows us to separate the design of the state feedback control

law from the design of the state estimator
Place the observer poles to the left of the poles of feedback
control law
Drawbacks
z Large observer gains can amplify the measurement noise
z Large gains of state feedback control law can result in saturation of
the input
z Uncertainties in the system parameters
z nonlinearities

Korea University of Technology and Education