Intelligent Control Systems for Humanoid Robot: Master Thesis_Owen Chih-Hsuan Chen

SOPC DO : M: L
HC
Design and Implementationof IntelligentControl
System forAutonomousHumanoid Robot based
on SOPC Technology
-. 9. 121 1.
/. 121 2 0 2
.
2018/7/27 1Control System Lab, Dept. of EE. , YZU

Agenda
• Chapter 1 Introduction
• Chapter 2 Mechanism and Hardware of the Humanoid Robot
• Chapter 3 Modeling and Reference Trajectories Planning of
Humanoid Robot
• Chapter 4 Adaptive CMAC-BasedDynamic-Balancingand ZMP
CompensationDesign
• Chapter 5 Simulation and Experimental Results
• Chapter 6 Conclusions and Future Works
2018/7/27 2Control System Lab, Dept. of EE. , YZU

Chapter 1 Introduction
2018/7/27 Control System Lab, Dept. of EE. , YZU 3

Why do people make robot?
– An aging society is coming
– To avoid human being doing high risk tasks
– To replace duplicating jobs
– For education
– Entertainments: pets
– Robot will soon become part of our day to day,
Bill gates said.

Why makes humanoid robot?
– Advantages
• A humanoid robot has a structure similar to human’s leg
and has higher mobility than conventionalwheeled
robots.
• Involve numerous field and only one platform relates to
wild areas
– Disadvantagesfor designer
• Balance should be concerned
• Complicated to design mechanism
• Has complex dynamics and many non-linear factors.
• It is very difficult to move a robot stably.

Introduction of SOPC
• System on Programmable chip (SOPC)
FPGA
- Nios II Plus All Peripherals Written In HDL
- Can Be TargetedFor All Altera FPGAs
- Synthesis Using Quartus II Integrated Synthesis
Avalon Switch Fabric
UART
GPIO
Timer
SPI
SDRAM
Controller
On-Chip
ROM
On-Chip
RAM
Nios II
CPU
Debug
Cache

Chapter 2 Mechanism and Hardware of the
Humanoid Robot

Design of Mechanism Criterion I
• Different link length will cause displacementof COG (Center
of Gravity)
COG Displacement
r
r
2r
r
r
2r
h1
h2
21 hh ≠21 hh =
h1
h2
Chapter 2 Mechanism andHardware of the HumanoidRobot

Design of Mechanism Criterion II
• Different leg length will cause large displacement whiling
COG moving to sole of the foot.
COG COG COG

The Hardware of the Robot

Actuators
• The whole trunk of robot is composed by 27 servo motors with
different size based on requested torque.
TYPE Torque
(kg/cm)
Speed
(sec/60°)
Operate
range(°)
Communic
ation
method
KRS-
4014HV
40.8 0.19 270 PWM
KRS-
2350HV
29.5 0.13 180 PWM
NARO 0.7 0.12 180 PWM
PICO 1.4 0.19 180 PWM

Processor
• Collects information
• Computes robot dynamic
• Performs human brain
• High hardware acceleration
Nios II
Processor 1
Nios II
Processor 2
Avalon bridge
Hardware
Mutex
Shared
Memory
DDR SDRAM
Controller
SSRAM
Controller
FLASH
Controller
On-Chip
Ram
HSMC
User Define
Interface
Timer
UART
SSRAM
DDR
SDRAM
FLASH

External circuits
Power management circuitData Acquisition circuit
peripheral circuits
Isolation circuit

Accelerometers and gyroscopes
• The accelerometer has been converted to an acceleration that
varies between –1 g and +1 g,
• TheADXRS300 is a complete angular rate sensor (gyroscope)

Force-sensing resistor
• Exhibits a decrease in resistance with an increase in force
applied normal to the device surface

CMOS sensor
• In practical situation high speed module without hyper speed
processor will be in vain.

Chapter 3 Modeling and Reference Trajectories
Planning of Humanoid Robot

Kinematics Analysis
• Forward kinematics for position/orientation
– determinethe position and orientation of the end effector
given the values for the joint variables of the robot
• Inverse kinematics for position/orientation
– determinethe values for the joint variables given the end
effector’s position and orientation
Chapter 3 Modelingand Reference Trajectories Planningof HumanoidRobot

Forward kinematics
• The position sense of the twist angle and the joint angle
are shown as below:
• The four principal homogeneous transforms are involved:
iθiα
ix ia
Joint axis i
Joint 1+i
iO
iα
1−ix
id
1−iz
iz
Normal 1−i
Joint axis 1−i
Joint i
1−iO
iθ
Normal i
iiiiiiii αxraxtdztθzr
i
i ,,,,
1
11
HHHHH −−
=−
(3.1)

Forward kinematics
• Homogeneous transform parameters are defined in appendix C
and is shown as
• For simplicity, the homogeneous transform denotes as .
⎥
⎥
⎥
⎥
⎦
⎤
⎢
⎢
⎢
⎢
⎣
⎡
−
−
=
⎥
⎥
⎥
⎥
⎦
⎤
⎢
⎢
⎢
⎢
⎣
⎡
⎥
⎥
⎥
⎥
⎦
⎤
⎢
⎢
⎢
⎢
⎣
⎡
⎥
⎥
⎥
⎥
⎦
⎤
⎢
⎢
⎢
⎢
⎣
⎡
⎥
⎥
⎥
⎥
⎦
⎤
⎢
⎢
⎢
⎢
⎣
⎡
= ××− −
1000
cossin0
sincossincoscossin
cossinsinsincoscos
1000
0
0
0
1000
0
0
1000
0
0
1000
0
0
0
,3333,1 1
iii
iiiiiii
iiiiiii
αx
i
i
θzi
i
dαα
θaθαθαθ
θaθαθαθ
a
d
iiii
RUUR
H
i
i
i AH =−1
(3.2)

Forward kinematics
• The end-link expressed by the homogeneous transform is
where
• The D-H Parameters of a Leg-Part of Humanoid Robot are
shown in Table 3.1.
nn
n
nn qqq AAAHHHqH !! 21
1
2
1
21
0
1
0
)()()()( == −
[ ]T
nqqq 21 !=q (3.4)
(3.3)

Forward kinematics
• leg-part of humanoid robot is given by
• The D-H parameters form is shown as:
• where
654321
5
6
4
5
3
4
2
3
1
2
0
1
0
6 )()()()()()()( AAAAAAHHHHHHqH 654321 == qqqqqq
⎥
⎥
⎥
⎥
⎦
⎤
⎢
⎢
⎢
⎢
⎣
⎡
=
1000
)(0
6
zzzz
yyyy
xxxx
paon
paon
paon
qH
(3.6)
(3.5)
(3.4)
6234165152341 )( ssccsssccnx +−−=
6234165152341 )( ssscccscsny ++−=
623465234 sccssnz −−=
6234165152341 )( cscscssccox ++=
(3.7)
(3.8)
(3.9)

Forward kinematics
• whenever convenient shorthand notation defines as
for trigonometricfunctions.
6234165152341 )( csssccscsoy +−=
623465234 ccsssoz −=
5152341 sscccax +−=
5152341 scccsay −−=
5234csaz −=
)()( 232321623411651523411 cacacsscaccssccapx ++++−=
)()( 232321623411651523411 cacassssacccscsapy ++++−=
2323262341652341 sasascacssapz ++−−=
θsθ sin=
)cos( φψc φψ +=+
(3.14)
(3.15)
(3.16)
(3.17)
(3.10)
(3.11)
(3.12)
(3.13)

Inverse kinematics
• One subchain which comprises joint variables , ,
in which denotes the two argument arctangent
function.
• Applying the cosine theorem to obtain
• Hence, is given by
• Similarly is given as
),(Atan21 cc xyθ =
32
2
3
2
2
222
3
2
cos
aa
aazyx
θ ccc −−++
=
)cos,cos1(Atan2 33
2
3 θθθ −±=
)cos,sin(Atan2),(Atan2 32332
22
2 θaaθayxzθ ccc +++=
1θ 2θ 3θ
),(Atan2 xy
3θ
2θ
(3.20)
(3.19)
(3.18)
(3.21)

Inverse kinematics
• Hence, is given by
• Then and are given respectively as
• It will get 8 different solutions. Then, preserve the one solution
by determining movablerange.
),(Atan2 33232323113231332323231132314 rsrcsrccrcrssrscθ +++−−=
),(Atan2 2111112211216 rcrsrcrsθ +−−=
))(1,(Atan2 2
2311312311315 rcrsrcrsθ −−±−=
4θ 6θ
5θ
(3.25)
(3.26)
(3.27)

Zero Moment Point
Center-of-Gravity Shift
Load Point (ZMP)
Center of Gravity
Load Point (ZMP)
Center of Gravity
Gravity
Inertial Force
Total Inertial Force
Out In
Fig. 3.2 The center of gravity shifting
N
M
P
Fig. 3.3 Concept of the ZMP

ZMP
• If ZMP is in the stable region for double support phase and
single support phase, then the robot will not go falling down
situation.
• where represents the ZMP vector, is the each
link
∑
∑ ∑
=
= =
−
−−−
= n
i
zii
n
i
n
i
ixiiizii
zmp
gzm
zgxmxgzm
x
1
1 1
)(
)()(
!!
!!!!
∑
∑ ∑
=
= =
−
−−−
= n
i
zii
n
i
n
i
ixiiizii
zmp
gzm
zgymygzm
y
1
1 1
)(
)()(
!!
!!!!
[ ]T
zyx ppp ,,=P im
(3.28)
(3.29)

Actual ZMP
• Thus the actual ZMP can be computed by
• To realize the actual ZMP via FSR measurement each sensor
reaction is shown as
∑
∑
=
=
= n
i
i
n
i
ii
ZMP
f
1
1
r
r
P
1f
2f
3f
5f6f
7f 8f
4f
x
y
o
1r
2r
3r 4r
5r
6r
7r
8r
Fig. 3.4 The vector of FSR located in sole of feet
(3.30)

Walking Analysis
Movement of
Center of Gravity
Sole of Foot
Movement of
Center of Gravity
Sole of Foot
Fig. 3.5 Walking pattern: (a): static walking; (b): dynamic walking.

Static walking & Dynamic walking
Gravity
Inertial Force
Total Inertial Force
Floor Reaction
Load Point (ZMP) Load Point (ZMP) Load Point (ZMP)
Fig. 3.6 The walking floor reaction of total inertial force

Linear inverted pendulum model
• To extract a dominant feature which is high-order and
nonlinear and to use this dominant factor to explain the
dynamics of the system
X
Y
Z
m
C

LIPM
• The position of the body can be calculated via the moment
of mass equation:
where is the position of the k-th particle, is the total
mass.
cr
∑=
=
n
k
kkc mm
1
rr
kr ∑=
=
n
k
kmm
1
1m
nm
km
CoM
m
cr
kr
x y
o
z
(3.31)

LIPM
• The dynamics equations of the inverted pendulum can be
derived as follows:
• To attain linear equations assume the z-coordinates of the
inverted pendulum is assumed to be constant
)(
)(
gcm
ccmcgcm
p
z
zxxz
x
−
−−
=
!!
!!!!
)(
)(
gcm
ccmcgcm
p
z
zyyz
y
−
−−
=
!!
!!!!
x
c
xx c
g
z
cp !!−=
y
c
yy c
g
z
cp !!−=
(3.33)
(3.32)
(3.34)
(3.35)

Table-cart model
• The moment around the ZMP must be zero the following
condition hold.
0=ZMPτ
O
mg−
xp
xm !!
x
cz
0)( =−−= cxzmp zxmpxmg !!τ (3.36)
Fig. 3.9 Table-cart model

Ideal ZMP
0Τ 02Τ 03Τ 04Τ 05Τ
ref
xP
ref
yP
0Τ 02Τ 03Τ 04Τ 05Τ
t
t
L
L2
L3
L4
L5
B
B−
Dotted : Single SupportPhase
Solid : Double SupportPhase
ref
xP
ref
yP
Single Support Phase
Double SupportPhase
Step Position
L L2 L3 L4 L5
B
B−

Fourier approximation CoM trajectories
• The ZMP reference trajectories can be expressed as follows.
• Finally, the CoM can be obtained by applying inverse Laplace
transformations as follows.
∑
∞
=
−=
1
0 )()(
k
ref
x kTtuLtp
∑
∞
=
−⋅−+⋅=
1
0 )()1(2)()(
K
Kref
y kTtuBtuBtp
[ ] )(1)(cosh1)( 00 kTtkTtωLtc
k
nx −⋅−−= ∑
∞
[ ] )(1)(cosh1)1(2)( 00 kTtkTtωBtc
k
n
k
x −⋅−−−= ∑
∞
(3.42)
(3.41)
(3.45)
(3.46)

Fourier approximation CoM trajectories
• Assuming the reference trajectory of CoM has the form by
using Fourier series.
• As a result, the x-directionalreferencetrajectory of CoM can
be found as follows:
• the y-directionalreferencetrajectory of CoM can be found in a
similar manner as follows:
∑
∞
=
⎥
⎦
⎤
⎢
⎣
⎡
⎟⎟
⎠
⎞
⎜⎜
⎝
⎛
+
+
+⎟
⎠
⎞
⎜
⎝
⎛
−=
1 0
2222
0
22
00
0
sin
)(
)cos1(
2
)(
n n
nref
x t
T
nπ
πnωTnπ
nπωLTT
t
T
L
tc
∑
∞
=
⎥
⎦
⎤
⎢
⎣
⎡
⎟⎟
⎠
⎞
⎜⎜
⎝
⎛
+
−
=
1 0
2222
0
22
0
sin
)(
)cos1(2
)(
n n
nref
y t
T
nπ
πnωTnπ
nπωBT
tc
(3.47)
(3.48)

Swing Foot Reference Trajectories
• The hip joint fix on the same height. The movement
trajectories show in below:
Hip trajectory
Foot trajectory
),( hh zx
),( ff zx
fh
oh
L L
⎥
⎦
⎤
⎢
⎣
⎡
−= )2sin(2)(
ss
f
T
t
π
T
t
π
π
L
tx
Btyf ±=)(
⎥
⎦
⎤
⎢
⎣
⎡
−= )2cos(1
2
)(
s
f
f
T
t
π
h
tz
Fig. 3.14 The swing foot trajectories

Chapter 4 Adaptive CMAC-Based Dynamic-
Balancing and ZMP Compensation
Design

CMAC
• CMAC-based adaptive control system
– Self-learning control parameter
– Simple computation
– Better control performance
Chapter 4 Adaptive CMAC-Based Dynamic-BalancingandZMP CompensationDesign

RCMAC
Input Space
Receptive-Field
Space
Weight Memory
Space
Association Memory
Space
Recurrent Unit
k1
φ
nk
φ
∑
∑
Output Space
kow ! !
-
∑
∑
kpw! ! !
anp
1p k1µ
kna
µ
kφ
ikr Tikr T
Ono
1o
sI
sA
sR
sW
sO
Input Space
Receptive-Field
Space
Weight Memory
Space
Association Memory
Space
Recurrent Unit
k1
φ
nk
φ
∑
∑
Output Space
kow ! !
-
∑
∑
kpw! ! !
anp
1p k1µ
kna
µ
kφ
ikr Tikr Tikr TTikr T
Ono
1o
sI
sA
sR
sW
sO
⎥
⎦
⎤
⎢
⎣
⎡ −−
= 2
2
)(
ik
iki
ik
v
cp
expµ
Association Memory Space:
Receptive-Filed Space:
)()()( Ttrtptp ikikirik −+= µ
Recurrent Unit:
⎥
⎦
⎤
⎢
⎣
⎡ −−
== ∑∏ ==
aa n
i ik
ikrik
n
i
ikkkkk
v
cp
exp
1
2
2
1
)(
),,,( µφ rvcp
∑=
==
dn
k
kkp
T
pp wo
1
φΦw
Output Space:
a
a
nT
nppp ℜ∈= ],,,[ 21 !p
Input Space:
Architecture of an RCMAC

Adaptive CMAC Control System
Humanoid RobotSystem
CMACAdaptiveLaws
Adaptation
PropagationLaw dt
d
truckδ
Inverse
Kinematics
WalkingPattern
Generator ∑+
CMACAdaptiveLaws
Adaptation
PropagationLaw dt
d
1
1 −
− z
eΔ
ZMPδ
ZMP
Computation
ZMP
Planner
∑+
_
∑
+
_
∑
+
+
+
ω
ZMPCompensatorvia ACMAC
Dynamic-BalancingController based AdaptiveCMAC
*
dReference
Model
md
if
oτ
trunke
ZMPe
ZMPp
d
ZMPU
HipU
drdvdmdw ,η,η,ηη
zrzvzmzw ,η,η,ηη
zzz vmw ˆ,ˆ,ˆ
ddd vmw ˆ,ˆ,ˆ
τ
OPP

ACMAC-Based Dynamic-Balancing
• The tracking error of dynamic-balancing controller is defined
as
• And the output of the ACMAC-based dynamic-balancing is
• The total desired is defined as
• The tracking error of dynamic-balancing controller is defined
as
• And the output of the ACMAC-based dynamic-balancing is
dde mtruck −=
HipU
Hipo U+=ττ
ZMPZMPZMP pde −=
ZMPU

ACMAC-Based ZMP Compensator
• The total desired hip trajectories is defined as
ZMPo UPP +=

Online Learning Algorithm
• The energy function, E, is defined as
• with the energy function , the error term to be propagated is
given by
• based on backpropagation method can be derived as
2018/7/27 45
22
2
1
)(
2
1
mm eddE =−=
ZMP
m
mZMP
ZMP
U
d
d
e
e
E
U
E
δ
∂
∂
∂
∂
∂
∂
−=
∂
∂
−=
kZMPzw
z
ZMP
ZMP
zw
z
zwz δη
w
U
U
E
η
w
E
ηw φ−=
∂
∂
∂
∂
−=
∂
∂
−=Δ
2
)(2
ik
ikrik
kzZMPzm
ik
ik
ik
k
k
ZMP
ZMP
zmik
v
mp
wδη
m
µ
µ
U
U
E
ηm
−
−=
∂
∂
∂
∂
∂
∂
∂
∂
−= φ
φ
φ
Δ
3
2
2
ik
ikrik
kzZMPzv
ik
ik
ik
k
k
ZMP
ZMP
zvik
v
)m(p
wδη
v
µ
µ
U
U
E
ηv
−
−=
∂
∂
∂
∂
∂
∂
∂
∂
−= φ
φ
φ
Δ
)(
2
2
Ttµ
v
)p(m
wδη
r
µ
µ
U
U
E
ηr ik
ik
rikik
kzZMPzr
ik
ik
ik
k
k
ZMP
ZMP
zrik −
−
−=
∂
∂
∂
∂
∂
∂
∂
∂
−= φ
φ
φ
Δ
(4.13)
(4.14)
(4.15)
(4.16)
(4.17)
(4.18)

Online Learning Algorithm
• cannot be computer if the plant model is unknown, an
adaptationpropagation law is given by
• The connective weights, means and variance can now be
updated according to the following equation
ZMPUd ∂∂ /
mmmmZMP eeddddδ +=−+−≅ Δ)()( !!
)(Δ)()1( NwNwNw zzz +=+
)(Δ)()1( NmNmNm ikikik +=+
)(Δ)()1( NvNvNv ikikik +=+
)(Δ)()1( NrNrNr ikikik +=+
(4.20)
(4.21)
(4.22)
(4.23)
(4.19)

Chapter 5 Simulation and Experimental Results

Experimental environment
Digital
Oscilloscope
DC Power Supply
12V/18A
FPGACyclone III
Starter boardMonitor
Vision System
Emergency
Bottom
FSRs
Internal Circuits
and Sensors
Chapter 5 Simulationand Experimental Results

Trajectories planning-ZMP
• ZMP trajectories
0 1 2 3 4 5 6 7 8
-2
0
2
4
6
8
10
12
Time (sec)
(a)
X-direction[cm]
ref
xc ref
yp
1.95 2 2.05 2.1 2.15 2.2 2.25 2.3
20
22
24
26
28
30
32
ref
ypideal
0 1 2 3 4 5 6 7 8
-8
-6
-4
-2
0
2
4
6
8
Time (sec)
(b)
Y-direction[cm]
3.8 4 4.2 4.4 4.6 4.8 5
36
38
40
42
44
46
48
50
52
ref
yp
ref
ypideal
ref
xc
0 1 2 3 4 5 6 7 8 9 10
-6
-4
-2
0
2
4
6
X-direction [cm](c)
Y-direction[cm]
57.1 57.2 57.3 57.4 57.5 57.6 57.7 57.8
35
35.5
36
36.5
37
37.5
38
38.5
39
39.5
ref
xc
ref
ypideal
ref
yp
Fig. 5.2 ZMP and COM reference (a) x-direction (b) y-direction (c) x-y plane

Trajectories planning-Swing foot
• Swing foot trajectories:
0 0.5 1 1.5 2 2.5 3 3.5 4
0
2
4
6
8
10
12
Time (sec)
(a)
X-direction[cm]
0 0.5 1 1.5 2 2.5 3 3.5 4
0
0.5
1
1.5
2
2.5
3
Time (sec)
(b)
Z-direction[cm]
0 2 4 6 8 10 12
0
0.5
1
1.5
2
2.5
3
X-direction [cm]
(c)
Z-direction[cm]
Fig. 5.3 Swing foot reference trajectories (a) x-direction (b )y-direction (c) x-z plane

Digital Filter
0 1 2 3 4 5 6 7 8 9 10
-50
-40
-30
-20
-10
0
10
20
30
40
50
Angle(deg)
Pitch anglewithout filter
(a)
Time (sec)
0 1 2 3 4 5 6 7 8 9 10
-20
-15
-10
-5
0
5
10
15
20
Angularvelocity(deg/sec)
Pitch angular velocitywithoutKalmanfilter
Noise
measurement
Time (sec)
(d)
0 1 2 3 4 5 6 7 8 9 10
-50
-40
-30
-20
-10
0
10
20
30
40
50
Angle(deg)
Pitch anglewith filter
(c)
Time (sec)
0 1 2 3 4 5 6 7 8 9 10
-20
-15
-10
-5
0
5
10
15
20
Time (sec)
Pitch angular velocitywith Kalmanfilter
Kalman estimate
with Q=0.0005
(f)
Fig. 5.4 Pitch angle and angular velocity signals: (a) original pitch angle. (c) pitch angle pass through 1K
Hz cut-off frequency complementary filter. (d) original pitch angular velocity. (f) pitch angular velocity
pass through Q=0.0005 Kalman filter.

Experimental Results
• Dynamic balancing
– Condition (i): without dynamic balancingcontroller
– Condition (ii): with dynamic balancing controller
• Dynamic walking
– Condition (i): walking on surface without ZMP compensator
– Condition (ii): walking on surface with ZMP compensator
– Condition (iii): walking on uphill with ZMP compensator

Dynamic balancing for condition(i)
0 1 2 3 4 5 6 7
-15
-10
-5
0
5
10
15
(a)
Time (sec)
Angle(deg)
Pitch angle
Inverse
disturbance
Forward
disturbance
0 1 2 3 4 5 6 7
-15
-10
-5
0
5
10
15
Rollangle
Time (sec)
(c)
Angle(deg)
Forward
disturbance
Inverse
disturbance
0 1 2 3 4 5 6 7
-15
-10
-5
0
5
10
15
Pitch angular velocity
Time (sec)
(b)
0 1 2 3 4 5 6 7
-15
-10
-5
0
5
10
15
(d)
Time (sec)
Roll angular velocity
Fig. 5.7 Snapshot of dynamic balancingFig. 5.6 Dynamic balancing for condition (i)

Dynamic Balancing Demonstration

Dynamic balancing for condition(ii)
0 1 2 3 4 5 6 7
-15
-10
-5
0
5
10
15
(a)
Time (sec)
Angle(deg)
Pitch angle
Inverse
disturbance
Forward
disturbance
0 1 2 3 4 5 6 7
-15
-10
-5
0
5
10
15
Rollangle
Time (sec)
(d)
Angle(deg)
Forward
disturbance
Inverse
disturbance
0 1 2 3 4 5 6 7
-15
-10
-5
0
5
10
15
Pitch angular velocity
Time (sec)
(b)
0 1 2 3 4 5 6 7
-15
-10
-5
0
5
10
15
(e)
Time (sec)
Roll angular velocity
0 1 2 3 4 5 6 7
-10
-8
-6
-4
-2
0
2
4
6
8
10
Controleffort ofvirtualpitch angle
Angle(deg)
Time (sec)
(c)
0 1 2 3 4 5 6 7
-10
-8
-6
-4
-2
0
2
4
6
8
10
Angle(deg)
Time (sec)
Controleffort of virtualrollangle
(f)
Fig. 5.8 Dynamic balancing for condition (ii)

Dynamic walking for condition (i)
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
30
0 2 4 6 8 10 12 14 16
-50
-40
-30
-20
-10
0
Joint Angle2
Angle(deg)
Time (sec)
Desired
Actual
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
30
0 2 4 6 8 10 12 14 16
-50
-40
-30
-20
-10
0
Joint Angle3
Angle(deg)
Time (sec)
Actual
Desired
0 2 4 6 8 10 12 14 16
-60
-40
-20
0
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
Joint Angle4
Angle(deg)
Time (sec)
Actual
Desired
0 2 4 6 8 10 12 14 16
-60
-40
-20
0
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
Joint Angle5
Angle(deg)
Time (sec)
Actual
Desired
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
30
Joint Angle6
Angle(deg)
Time (sec)
Actual
Desired
0 2 4 6 8 10 12 14 16
-50
-40
-30
-20
-10
0
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
30
Joint Angle11
Angle(deg)
Time (sec)
Actual
Desired
0 2 4 6 8 10 12 14 16
-50
-40
-30
-20
-10
0
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
30
Joint Angle10
Angle(deg)
Time (sec)
Desired
Actual
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
0 2 4 6 8 10 12 14 16
10
20
30
40
50
60
70
Joint Angle9
Angle(deg)
Time (sec)
Actual
Desired
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
0 2 4 6 8 10 12 14 16
10
20
30
40
50
60
70
Joint Angle8
Angle(deg)
Time (sec)
Actual
Desired
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
30
Joint Angle7
Angle(deg)
Time (sec)
Actual
Desired
Fig. 5.10 Joint angle of left legFig. 5.9 Joint angle of right leg

Dynamic walking for condition (i)
0 2 4 6 8 10 12 14 16
-10
-8
-6
-4
-2
0
2
4
6
8
10
Time (sec)
Angle(deg)
Pitch angle
(a)
0 2 4 6 8 10 12 14 16
-10
-8
-6
-4
-2
0
2
4
6
8
10
Time (sec)
Angle(deg)
Rollangle
(c)
0 2 4 6 8 10 12 14 16
-5
0
5
10
15
20
25
Time (sec)
ZMPx-coordinate(cm)
(b)
Stableregion
0 2 4 6 8 10 12 14 16
-8
-6
-4
-2
0
2
4
6
8
Time (sec)
ZMPy-coordinate(cm)
(d)
Stable region
-2 0 2 4 6 8 10 12 14 16 18
-8
-6
-4
-2
0
2
4
6
8
ZMP x-coordinate (cm)
ZMPy-coordinate(cm)
(e)
Fig. 5.11 (a),(c) Attitude of humanoid robot without ZMP compensator for condition
(i). (b),(d)-(e) Actual ZMP for condition (i)

2018/7/27 Control System Lab, Dept. of EE. , YZU
58
Dynamic walking for condition (ii)
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
30
0 2 4 6 8 10 12 14 16
-50
-40
-30
-20
-10
0
Joint Angle2
Angle(deg)
Time (sec)
Desired
Actual
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
30
0 2 4 6 8 10 12 14 16
-50
-40
-30
-20
-10
0
Joint Angle3
Angle(deg)
Time (sec)
Desired
Actual
0 2 4 6 8 10 12 14 16
-60
-40
-20
0
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
Joint Angle4
Angle(deg)
Time (sec)
Desired
Actual
0 2 4 6 8 10 12 14 16
-60
-40
-20
0
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
Joint Angle5
Angle(deg)
Time (sec)
Desired
Actual
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
30
Joint Angle6
Angle(deg)
Time (sec)
Desired
Actual
0 2 4 6 8 10 12 14 16
-50
-40
-30
-20
-10
0
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
30
Joint Angle11
Angle(deg)
Time (sec)
Desired
Actual
0 2 4 6 8 10 12 14 16
-50
-40
-30
-20
-10
0
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
30
Joint Angle10
Angle(deg)
Time (sec)
Desired
Actual
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
0 2 4 6 8 10 12 14 16
10
20
30
40
50
60
70
Joint Angle9
Angle(deg)
Time (sec)
Desired
Actual
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
0 2 4 6 8 10 12 14 16
10
20
30
40
50
60
70
Joint Angle8
Angle(deg)
Time (sec)
Desired
Actual
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
30
Joint Angle7
Angle(deg)
Time (sec)
Desired
Actual
Fig. 5.13 Joint angle of left leg Fig. 5.13 Joint angle of left leg

Dynamic walking for condition (ii)
0 2 4 6 8 10 12 14 16
-10
-8
-6
-4
-2
0
2
4
6
8
10
Time (sec)
Angle(deg)
Pitch angle
(a)
0 2 4 6 8 10 12 14 16
-10
-8
-6
-4
-2
0
2
4
6
8
10
Time (sec)
Angle(deg)
Rollangle
(d)
0 2 4 6 8 10 12 14 16
-10
-8
-6
-4
-2
0
2
4
6
8
10
Angle(deg)
Time (sec)
(b)
0 2 4 6 8 10 12 14 16
-10
-8
-6
-4
-2
0
2
4
6
8
10
Angle(deg)
Time (sec)
Controleffort ofvirtualrollangle
(e)
0 2 4 6 8 10 12 14 16
-5
0
5
10
15
20
25
Time (sec)
ZMPx-coordinate(cm)
(c)
Stableregion
0 2 4 6 8 10 12 14 16
-8
-6
-4
-2
0
2
4
6
8
Time (sec)
ZMPy-coordinate(cm)
(f)
Time (sec)
ZMPy-coordinate(cm)
Stable region
-2 0 2 4 6 8 10 12 14 16 18
-8
-6
-4
-2
0
2
4
6
8
ZMPy-coordinate(cm)
(g)
Fig. 5.14 (a),(d) Attitude of humanoid robot for condition (ii). (b),(e) control effort of virtual angle for condition (iii).
(c),(f)-(g) Actual ZMP for condition (ii)

Dynamic walking for condition (iii)
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
30
0 2 4 6 8 10 12 14 16
-50
-40
-30
-20
-10
0
Joint Angle2
Angle(deg)
Time (sec)
Desired
Actual
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
30
0 2 4 6 8 10 12 14 16
-50
-40
-30
-20
-10
0
Joint Angle3
Angle(deg)
Time (sec)
Desired
Actual
0 2 4 6 8 10 12 14 16
-60
-40
-20
0
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
Joint Angle4
Angle(deg)
Time (sec)
Desired
Actual
0 2 4 6 8 10 12 14 16
-60
-40
-20
0
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
Joint Angle5
Angle(deg)
Time (sec)
Desired
Actual
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
30
Joint Angle6
Angle(deg)
Time (sec)
Desired
Actual
0 2 4 6 8 10 12 14 16
-50
-40
-30
-20
-10
0
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
30
Joint Angle11
Angle(deg)
Time (sec)
Desired
Actual
0 2 4 6 8 10 12 14 16
-50
-40
-30
-20
-10
0
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
30
Joint Angle10
Angle(deg)
Time (sec)
Desired
Actual
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
0 2 4 6 8 10 12 14 16
10
20
30
40
50
60
70
Joint Angle9
Angle(deg)
Time (sec)
Desired
Actual
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
0 2 4 6 8 10 12 14 16
10
20
30
40
50
60
70
Joint Angle8
Angle(deg)
Time (sec)
Desired
Actual
0 2 4 6 8 10 12 14 16
-30
-20
-10
0
10
20
30
Joint Angle7
Angle(deg)
Time (sec)
Desired
Actual
Fig. 6.15 Joint angle of right leg Fig. 5.16 Joint angle of left leg

Dynamic walking for condition (iii)
0 2 4 6 8 10 12 14 16
-10
-8
-6
-4
-2
0
2
4
6
8
10
Time (sec)
Angle(deg)
Pitch angle
(a)
0 2 4 6 8 10 12 14 16
-10
-8
-6
-4
-2
0
2
4
6
8
10
Time (sec)
Angle(deg)
Rollangle
(d)
0 2 4 6 8 10 12 14 16
-10
-8
-6
-4
-2
0
2
4
6
8
10
Angle(deg)
Time (sec)
(b)
0 2 4 6 8 10 12 14 16
-10
-8
-6
-4
-2
0
2
4
6
8
10
Angle(deg)
Time (sec)
Virtualrollangle
(e)
0 2 4 6 8 10 12 14 16
-5
0
5
10
15
20
25
Time (sec)
ZMPx-coordinate(cm)
(c)
Stableregion
0 2 4 6 8 10 12 14 16
-8
-6
-4
-2
0
2
4
6
8
Time (sec)
ZMPy-coordinate(cm)
(f)
Time (sec)
ZMPy-coordinate(cm)
Stable region
-2 0 2 4 6 8 10 12 14 16 18
-8
-6
-4
-2
0
2
4
6
8
ZMPy-coordinate(cm)
(g)
Fig. 5.17 (a),(d) Attitude of humanoid robot for condition (iii). (b),(e) control effort of virtual angle for condition (iii).
(c),(f)-(g) Actual ZMP for condition (iii)

Dynamic walking- Siggital

Dynamic walking-Frontal

Chapter 6 Conclusions and Future Works

Conclusions
• Robot system has been designedand implementedsuccessfully
– Mechanism design
– Multi-sensors
– External circuit
– SOPC
• Trajectoriesplanning
– Swing foot trajectories
– CoM and ZMP trajectories
• Digital filter
– Kalman filter
– Complementaryfilter
• Dynamic balancingbasedon ACMAC
– Online learning algorithm
– Gradient descent method
– Virtual angle
• ACMAC-based ZMP compensator

Future Research
• Advanced particle swarm optimization for finding the optimal
learning rate
• Impendencecontrol
• Fast dynamic walking
• Build up middle size humanoid robot

Thanks for your attention

Intelligent Control Systems for Humanoid Robot: Master Thesis_Owen Chih-Hsuan Chen

More Related Content

Similar to Intelligent Control Systems for Humanoid Robot: Master Thesis_Owen Chih-Hsuan Chen (20)

Recently uploaded (20)

Intelligent Control Systems for Humanoid Robot: Master Thesis_Owen Chih-Hsuan Chen