Professional Documents
Culture Documents
Enpm808J Rehabilitation Robotics Lecture #2 Anindo Roy, PHD
Enpm808J Rehabilitation Robotics Lecture #2 Anindo Roy, PHD
ENPM808J
Rehabilitation Robotics
Lecture #2
The ReWalk
19
The ReWalk
20
• Potential causes
Protocol violation (safety) – unlikely (2 sites, different teams)
Hardware – unlikely (2 different robots)
• What then?
21
• Controller (likely)
Control strategy – wrong type of controller with respect to application
Controller settings – right type of controller but poorly tuned
• Trajectory-‐following control
• Trajectory-‐following control
Input(s) Output(s)
Physical
System
Basic
control
structure
Physical
System
Physical
System
System
stability
Input(s) Output(s)
Inverted
pendulum
Physical
Inherently
unstable
System
Some
systems
become
unstable
owing
to
nature
(e.g.
amplitude)
of
input(s)
Acquired
instability
The
need
for
control
System
stability
Input(s) Output(s)
System
performance
Physical
System
Mass-‐spring-‐damper
system:
• Amplitude,
frequency
response
• Steady-‐state,
rise
&
decay
Emes
Basic
control
ques>ons
Control
Law:
Relates
the
controller
output
to
a
system
variable
such
as
system
input
etc.
u
=
f(x)
u:
controller
output
(also:
control
signal)
f:
operator
or
mathema>cal
func>on
(the
“law”)
x:
system
variable
(e.g.
input)
Control
Law:
Feed-‐forward
System
u
x Controller y
Physical
System
u
=
f(x)
Control
Law:
Feedback
System
u
x Controller y
Physical
System
u
=
f(x,
y)
Robot
Control
Systems
• Trajectory-‐following control
f
=f
Natural
frequency
(only
k)
k:
spring
sEffness
b:
damping
constant
System
m:
mass
constants
f:
applied
force
x:
posiEon
along
axis
of
applied
Damping
constant
force
(both
k
and
b)
Physical
System:
No
Control
Vary
ωn
Fix
ζ
Decaying
oscillaEons
Physical
System:
No
Control
Fix
ωn
Vary
ζ
Decaying
oscillaEons
Physical
System:
No
Control
Fix
b
Vary
k
Fix
k
Vary
b
Our
objecEve:
Se0ling
3me
be
upper-‐bounded
That
is,
the
system
should
aCain
its
steady-‐state
response
within
desired
:me
(the
upper
bound).
SeCling
:me:
the
>me
it
takes
for
the
system
response
to
aaain
some
propor>on
(usually
90-‐100%)
of
its
steady-‐state
(final)
response
Physical
System:
No
Control
At t = ts,
x(t = ts) =
t = ts
=0
ts = (1/ωd)tan-1(α/β)
Physical
System:
No
Control
ts = (1/ωd)tan-1(α/β) ≤ c
ts = (1/ωd)tan-1(α/β) ≤ c
f
Closed-‐loop
Control
System
Choose control
f
Closed-‐loop
Control
System
Choose control
Propor>onal
Control
Part
Law
Deriva>ve
Part
Closed-‐loop
Control
System
Choose control
f
Closed-‐loop
Control
System
Choose control
Controller
Closed-‐loop
Control
System
Choose control
Controller
What
is
controlled
here?
Closed-‐loop
Control
System
Choose control
Controller
What
is
controlled
here?
Posi>on
Closed-‐loop
Control
System
Choose control
Controller
What
is
controlled
here?
Posi>on
Steady
state?
Closed-‐loop
Control
System
Choose control
Controller
What
is
controlled
here?
Posi>on
Steady
state?
x
=
0
Physical
System:
With
Control
f
=f
k = k + kp
k:
spring
sEffness
b:
damping
constant
m:
mass
f:
applied
force
x:
posiEon
along
axis
of
applied
force
b = b + kv
Physical
System:
With
Control
ts = (1/ωd)tan-1(α/β) ≤ c
Yes!!
Adjust PD controller parameters!
Physical
System:
With
Control
Cri:cal
Damping
(b + kv)2 = 4m(k + kp)
Cri:cal
Damping
(b + kv)2 = 4m(k + kp)
Problem:
Make
the
controller
parameters
independent
of
system
proper3es
(e.g.
mass)
25
Laplace Transforms
Transfer Functions
26
Modeling Systems
• Systems may be studied in the frequency or time domain
• Frequency-domain: how the amplitude of the signal changes
with respect to frequency
Transfer functions
Modeling Systems
• The relationship between the input and output is
described by a mathematical function called a transfer
function
• Transfer functions can be used in block diagrams to study
the flow of inputs, outputs, controllers, and so on
• Systems described by differential equations are difficult to
model with block diagrams
Laplace transform may be used
30
Damping:
dx
F = Dv = D
dt
k D xout Spring: F = kx
31
Fin
2
d x
∑ F = m dt 2 = Fin − FD − Fk
d 2x dx
∑ F = m dt 2 = Fin − D dt − kx
k D xout
d 2x dx
m 2 + D + kx = Fin
dt dt
Second order differential equation
44
xout ms X ( s ) + DsX ( s ) + kX ( s ) = F ( s )
2
k D
ms X ( s ) + DsX ( s ) + kX ( s ) = F ( s )
2
Fin
(ms + Ds + k ) X ( s ) = F ( s )
2
F (s)
= ms + Ds + k
2
k D xout X (s)
X (s) 1
= 2 = G (s)
F ( s ) ms + Ds + k
54
Fin
k D xout
X (s) 1
= 2 = G (s)
F ( s ) ms + Ds + k
34
Error Regulation
35
“Unity Feedback” (H = 1)
• Percent overshoot
• Settling time
• Peak time
• Rise time
37
Response
38
Response
42
c(t ) SS = Ke(t ) SS
1
e(t ) SS = c(t ) SS
Error diminishes K
as gain K increases
43
E ( s) = R( s) − C ( s)
C ( s ) = R ( s )T ( s )
E ( s ) = R( s )[1 − T ( s )]
How do we determine the steady-state error at time t?
45
E ( s) = R( s) − C ( s)
C ( s ) = R ( s )T ( s )
E ( s ) = R( s )[1 − T ( s )]
E ( s) = R( s) − C ( s)
C ( s ) = E ( s )G ( s )
R( s)
E (s) =
1 + G (s)
29
Stability Analysis
30
Response
31
• Marginally stable?
Poles on Im axis
37
1
K s ( s + 1)( s + 2)
1
K s ( s + 1)( s + 2)
40
1
K s ( s + 1)( s + 2)
41
1
K s ( s + 1)( s + 2)
42
1
K s ( s + 1)( s + 2)
43
1
K s ( s + 1)( s + 2)
44
1
K s ( s + 1)( s + 2)
45
1
K s ( s + 1)( s + 2)