1
DLR 02/05/2012
Feedback control of humanoid robots: balancing and walkingDr.-Ing. Christian Ott
German Aerospace Center (DLR)
Institute for Robotics and Mechatronics
Folie 2
Overview
Part 0: Short overview of (biped robots at) DLR
Part I: Modeling
Part II: Balancing
Part III: Walking Control
Folie 3
German Aerospace Center (DLR)
National research laboratoryPart of the Helmholtz associationResearch fields:
o aerospace, o space technologies, o energy and traffic
~6300 researchers13 locations29 institutes
Institute of Robotics and Mechatronics: Former director: Prof. HirzingerDirector: Prof. Albu-Schäffer
Folie 4
Institute of Robotics & Mechatronics
LBR-II 1999/2000
ROKVISS 2007-2010
LBR-I,199x
Space Robotics
Manipulation
ROTEX 1993
GETEX
ESS
Torque sensors at the power outputafter the bearings
Torque sensors afterthe gears, but beforethe last bearings
Modular design,load/weight ratio ~ 1:1
LBR-M, 2003
Commercial torque controlled arm (KUKA)
Compliant ManipulationJoint torque sensing & control for manipulation
Robustness: Passivity Based Control
Performance:Joint Torque Feedback
(noncollocated)
MotorDynamics
Rigid-BodyDynamics
TorqueControl
Environ-ment
ComplianceControl
extq
um
Compliant Manipulation
Robustness: Passivity Based Control
Performance:Joint Torque Feedback
(noncollocated)
MotorDynamics
Rigid-BodyDynamics
TorqueControl
Environ-ment
ComplianceControl
extq
um
Folie 8
2009-2013
Anthropomorphic Hand-Arm System[Grebenstein, Albu-Schäffer et al, Humanoids 2010]
• Compliant actuation
• Antagonistic actuation for fingers
• Variable stiffness actuation in arm
• Robustness to shocks and impacts
Biped RobotJoint torque sensing & control for manipulation
Folie 9
Bipedal Walking Robots at DLR
DLR-Biped(2010-2012)
TORO, preliminary version (2012)
TORO (2013)TOrque controlled humanoid RObot
• Drive technology of the DLR arm Allow for position controlled walking (ZMP) and joint torque control!
• Small foot design: 19 x 9,5 cm
• Sensors:- joint torque sensors- force/torque sensors in the feet- IMU in the trunk
Folie 10
First experiments with DLR-Biped
First experiment at Automatica Fair in 06-2010: ZMP preview control [Kajita, 2003]Current approach: Walking control based on the Capture Point
[Englsberger, Ott, Roa, et al. IROS 2011]
Folie 11
TORO
Folie 12
Overview
Part I: Modeling• Multibody dynamics• ZMP• Simplified models for control• Capture Point• Centroidal Moment Pivot
Part II: Balancing
Part III: Walking Control
Models of Legged (Humanoid) Robots
Multi-Body-Models Conceptual Models
Fixed Base Models(predefined contact state)
Floating Base Models Walking Running
Dynamical Models (Mechanical)
Complexity
Specialization
Free-Floating vs. Fixed Base Models
Fixed base modelsIn each contact state the model is different:
• Single support (right, left)• Double support• Heel Off• Toe Touch Down• …
Transition between contact states
double supportover-contrained
single supportserial kin. chain
Free-floating model
Components:• Lagrangian dynamics• Constraints due to contact forces• Transition equations (impacts)
underactuated
Folie 15
Free-Floating vs. Fixed Base Models
Fixed base modelsIn each contact state the model is different:
• Single support (right, left)• Double support• Heel Off• Toe Touch Down• …
Transition between contact states
double supportover-contrained
single supportserial kin. chain
Free-floating model
Components:• Lagrangian dynamics• Constraints due to contact forces• Transition equations (impacts)
underactuated
Planning & control must ensure that the considered contact state is valid! ground reaction force must fulfill constraints
Configuration Space
)3(SEHb Qq
n
111 SSST n
Configuration Space: )3(SEQ
Configuration Space
)3(SEHb Qq
n
111 SSST n
Configuration Space: )3(SEQ
Using local coordinates: n6
6bx
Folie 18
Modeling
)3(6 seFext
extT FqJqgqqqCqqM )()(),()(
,q
Folie 19
Modeling
extF
extT FqJqgqqqCqqM )()(),()(
extF
,q
6bx
6bF
Folie 20
Modeling
extF
extT FqJqgqqqCqqM )()(),()(
extF
extT
Tbb
bb
bb
qx
xqx FqJqJF
qxgqx
qxqCqx
qMqMqMqM
)()(
),(),,()()()()(
,q
6bx
6bF
Folie 21
Modeling
extF
extT FqJqgqqqCqqM )()(),()(
extF
extT
Tbtb
sbbbqb
bqb FqJ
qAdWqHg
qqqC
qqMqMqMqM
)()(
),(),,()()()()(
,q
)3()3(
sexSEHx
b
sbb
body twist
Global representation!
Folie 22
Modeling
extF
rF lF
l
r
q
6bx
Folie 23
Modeling
extF
rF lF
lT
l
Tbl
rT
r
Tbr
bb
bb
qx
xqx FqJ
qJFqJ
qJqxg
qx
qxqCqx
qMqMqMqM
)(0
)(
0)()(
0),(),,(
)()()()(
l
r
q
6bx
Folie 24
Bipedal Robot Model
rF lF
lT
l
Tbl
rT
r
Tbr
bb
bb
qx
xqx FqJ
qJFqJ
qJqxg
qx
qxqCqx
qMqMqMqM
)(0
)(
0)()(
0),(),,(
)()()()(
l
r
q
6bx
Properties for control:• Underactuated• Varying unilateral constraints
(single support, double support, edge contact)
• Constraints on the state & control
Folie 25
l
Tl
Tbl
rT
r
Tbr
bb
bb
qx
xqx FqJ
qJFqJ
qJqxg
qx
qxqCqx
qMqMqMqM
)(0
)(
0)()(
0),(),,(
)()()()(
pf p p
c
Mg
lríiT
i
FqJ
Iu
MgqqCq
cqM
M
, )ˆ(00
0)ˆ,ˆ(ˆ0
ˆ)(ˆ00
p
q
),( c
bx
Bipedal Robot Model
Folie 26
l
Tl
Tbl
rT
r
Tbr
bb
bb
qx
xqx FqJ
qJFqJ
qJqxg
qx
qxqCqx
qMqMqMqM
)(0
)(
0)()(
0),(),,(
)()()()(
pf p p
c
Mg
lríiT
i
FqJ
Iu
MgqqCq
cqM
M
, )ˆ(00
0)ˆ,ˆ(ˆ0
ˆ)(ˆ00
p
q
lri
ifMgcM,
),( c
bx
Conservation of momentum:
Bipedal Robot Model
Folie 27
l
Tl
Tbl
rT
r
Tbr
bb
bb
qx
xqx FqJ
qJFqJ
qJqxg
qx
qxqCqx
qMqMqMqM
)(0
)(
0)()(
0),(),,(
)()()()(
pf p p
c
Mg
lríiT
i
FqJ
Iu
MgqqCq
cqM
M
, )ˆ(00
0)ˆ,ˆ(ˆ0
ˆ)(ˆ00
p
q
lri
ifMgcM,
),( c
bx
iMgcL Conservation of angular momentum:
Conservation of momentum:
Bipedal Robot Model
Folie 28
l
Tl
Tbl
rT
r
Tbr
bb
bb
qx
xqx FqJ
qJFqJ
qJqxg
qx
qxqCqx
qMqMqMqM
)(0
)(
0)()(
0),(),,(
)()()()(
pf p p
c
Mg
lríiT
i
FqJ
Iu
MgqqCq
cqM
M
, )ˆ(00
0)ˆ,ˆ(ˆ0
ˆ)(ˆ00
p
q
lri
ifMgcM,
),( c
bx
iMgcL Conservation of angular momentum:
Conservation of momentum:
On a flat ground:
)( MgcMpMgcLp
Bipedal Robot Model
ppfp
Folie 29
l
Tl
Tbl
rT
r
Tbr
bb
bb
qx
xqx FqJ
qJFqJ
qJqxg
qx
qxqCqx
qMqMqMqM
)(0
)(
0)()(
0),(),,(
)()()()(
pf p p
c
Mg
lríiT
i
FqJ
Iu
MgqqCq
cqM
M
, )ˆ(00
0)ˆ,ˆ(ˆ0
ˆ)(ˆ00
),( c
bx
On a flat ground:
)( MgcMpMgcLp
Bipedal Robot Model
Zero Moment Point
Folie 31
Zero Moment Point[Vukobratovic and Stepanenko,1972]
)(x1x 2x
F
ZMP as a ground reference point: Distributed ground reaction force under the supporting foot can be replaced by a single force acting at the ZMP.
z
x
),( yx
y
ZMP = CoP (Center of Pressure)
p1x 2x
F0
pyx 0),( in convex hull of the support polygon.
Folie 32
Some facts about the ZMP
Can ZMP leave the support polygon? NOCan ZMP location be used as a stability criterion NO
If ZMP reaches the border of the support polygon foot rotation possible.
ZMP is defined on flat contact (no uneven surface).ZMP gives no information about sliding.
)(x1x 2x
F
First usage of the ZMP
• Motion of the legs is predefined.• Upper body controls the ZMP in the center of the supporting foot
ensure proper foot contact during walking
How to obtain the ZMP?
Measurement e.g. by Force/Torque Sensor
Dynamics Computation
How to obtain the ZMP?
Measurement e.g. by Force/Torque Sensor
Dynamics Computation
sss fppp )()(
z
sfs
z
zsyyzszxy
z
zsxxzszyx
ffpfpp
p
ffpfpp
p
)(
)(
sp
p
How to obtain the ZMP?
Measurement e.g. by Force/Torque Sensor
Dynamics Computation
sss fppp )()(
z
zsyyzszxy
z
zsxxzszyx
ffpfpp
p
ffpfpp
p
)(
)(
pf p p
ppfp
z
sfs
sp
p
How to obtain the ZMP?
Dynamics Computation
pf p p
ppfp
MgcLfMgP
c
0
0
py
px
)( MgPpMgcLp
z
xyzyy
z
yxzxx
PMgLPpMgc
p
PMgLPpMgc
p
A simplified walking model based on the ZMP
Mass concentrated model
pf p p
ppfp
Mass concentrated model
pf p p
ppfp
c0
zp
cMcLcMP
Mass concentrated model
pf p p
ppfp
c
z
xyzyy
z
yxzxx
PMgLPpMgc
p
PMgLPpMgc
p
0
zp
cMcLcMP
z
yzyy
z
xzxx
cgcc
cp
cgcccp
ZMP of a mass concentrated model
Mass concentrated model
p
c
z
yzyy
z
xzxx
cgcc
cp
cgcccp
gcccp xz
xx
0 zz cc
xc
Cart-Table Model [Kajita]
xx cp
Mass concentrated model
p
c
gcccp xz
xx
xc
Cart-Table Model [Kajita]Linear Inverted Pendulum Model [Sugihara]
xxz
x pccgc
p
c
xx pc xx cp
Capture Point(Extrapolated Center of Mass)
((Divergent Component of Motion))
Capture PointDefinition of the “Capture Point” (Pratt 2006, Hof 2008):
Point to step in order to bring the robot to stand.
constp
0
0* xxp
ptxtxttx ))cosh(1()0()sinh()0()cosh()(
ptx )(
c
p *p
cc ,
Computation of the Capture Point:
zcg
ZMP
pxx 2
Can be computed exactly for simple models, e.g. linear inverted pendulum model:
Capture Point Dynamics
xx
Coordinate transformation: ),(),( xxx
)(2 pxx p
xx
COMcapturepoint
xp
System structure: Cascaded system
exp. stableopen loopunstable
Dual use of the capture point for robotics1. step planning2. control
Capture Point in Human Measurements
Linear Inverted Pendulum Human
xx
y yData from [*]
[*] Hof, The extrapolated center of mass concept suggests a simple control of balance in walking, Human Movement Science 27, pp.112-125, 2008.
Centroidal Moment Pivot
Centroidal Moment Pivot
• Observation in human data: For normal level-ground walking, the human body‘s angular momentum (and the angular excursions) about the COM remains small through the gait cycle.
• The centroidal moment pivot was introduced as a ground reference point to address the effects of angular momentum about the COM in connection with postural balance strategies.
Centroidal Moment Pivot
Forces in the LIP model
p
x
z
Mg
xM
F
Effect of an additional hip torque
CMP
zgM
xM
F
pxzgx
zFpCMP
Mz
pxz
zgx
Interpretation
• The distance between CMP and ZMP corresponds to the angular momentum about the COM.
• While the ZMP cannot leave the support polygon (by definition), the CMP can leave it.
• The distance between CMP and the support polygon has been proposed as an indicator which balance strategy should dominate (via ZMP or via angular momentum).
Folie 52
Overview
Part I: Modeling
Part II: Balancing1. Basics2. ZMP based balancing (concentrated mass model)3. Torque based balancing (multi body dynamics)
Part III: Walking Control
Folie 53
Humanoid Balance
Vestibular system
Vision
Somatosensory system
IMU
Vision
force sensors
joint sensing
“Balance” is a generic term describing the ability to control the body posturein order to prevent falling.
Folie 54
Humanoid Balance
Small push:Ankle strategy
force controlZMP control
angular momentum control
Medium push:Hip strategy
Large Push: Step strategy
Human
Robot
Strategies for human push recovery:
Folie 55
mass concentrated model
Strategies for gait stabilization: Effect of an additional hip torque
p
Mg
xM
F
Mz
pxzgx
1. Controlling ZMP (constraints!)
2. Angular momentum control
3. Step adaptation
Folie 56
Overview
Part I: Modeling
Part II: Balancing1. Basics2. ZMP based balancing (concentrated mass model)3. Torque based balancing (multi-body model)
Part III: Walking Control
Folie 57
Motivation for compliant control
completely stiff fully compliantcompliant control
Folie 58
ZMP based balancing
Joint Position Control
Feedback Stabilization
dx dqrefx
refpInverse
Kinematicsdq
RobotDynamics
Forward Kinematics
ZMPComputation
p
x q
LR FF ,
position/velocity controlled robot
p
x
z
Mg
xM
F
MgFxM
zxp ext
extF
Folie 59
ZMP based balancing
Joint Position Control
Feedback Stabilization
dx dqrefx
refpInverse
Kinematicsdq
RobotDynamics
Forward Kinematics
ZMPComputation
p
x q
LR FF ,
position/velocity controlled robot
)()( refPrefdXrefd ppKxxKxx
Control law for stabilization:
Stability condition [*]: 0 PX KK
Stability condition [*]: [Choi, Kim, Oh, and You, Posture/Walking Control for Humanoid Robot Based on Kinematic Resolution of CoM Jacobian With Embedded Motion, TRO, 2007].
Effective Stiffness:
zMg
KKK
P
X
1
p
x
z
Mg
xM
F
MgFxM
zxp ext
extF
ZMP Based balancing
Balancing + Vertical Motion
Balancing + Vertical Motion+ Compliant Orientation
Folie 61
Overview
Part I: Modeling
Part II: Balancing1. Basics2. ZMP based balancing (concentrated mass model)3. Torque based balancing (multi-body model)
Part III: Walking Control
Folie 62
Balancing & Posture Control
Trunk orientation Control
)()( dDdPCOM ccKccKMgF
Mg
extF
COMF
HIPT
)3(SOR
)(),(dR
RHIP DKRVT
extT
c
IMU measurements
Compliant COM control [Hyon & Cheng, 2006]
Folie 63
Balancing & Posture Control
Compliant COM control [Hyon & Cheng, 2006]
Trunk orientation Control
)()( dDdPCOM ccKccKMgF
Mg
extF
COMF
HIPT
)3(SOR
)(),(dR
RHIP DKRVT
extT
),( HIPCOMd TFW Desired wrench:
IMU measurements
Folie 64
Balancing & Posture Control
Compliant COM control [Hyon & Cheng, 2006]
Trunk orientation Control
)()( dDdPCOM ccKccKMgF
)(),(dR
RHIP DKRVT
),( HIPCOMd TFW Desired wrench:
IMU measurements
dW
extFMg
Folie 65
Grasping and Balancing
Force distribution: Similar problems!
Folie 66
Force Distribution in Grasping
F
FGGFGFGWO
1
111Net wrench acting on the object:
TPiOi AdG
Grasp Map
if
)3(seFC
Well studied problem in grasping: Find contact wrenches such that a desired net wrench on the object is achieved.
FCFC
)3(se
friction cone
Folie 67
Force distribution
HIPT
COMF
f
fGGWd
1
1
ii
ii Rp
RG
ˆ
3if
Relation between balancing wrench & contact forces
Constraints:• Unilateral contact: (implicit handling of ZMP constraints)• Friction cone constraints
0, zif
Cf
T
F
GG
Folie 68
Force distribution
HIPT
COMF
f
fGGWd
1
1
ii
ii Rp
RG
ˆ
3if
Relation between balancing wrench & contact forces
Constraints:• Unilateral contact: (implicit handling of ZMP constraints)• Friction cone constraints
0, zif
Formulation as a constraint optimization problem
Cf
23
22
21minarg CCTHIPCFCOMC ffGTfGFf
T
F
GG
321
Folie 69
Contact force control via joint torques
ifMgcM
3if
c
lríiT
i
FqJ
Iu
MgqqCq
cqM
M
, )ˆ(00
0)ˆ,ˆ(ˆ0
ˆ)(ˆ00
iT
i fqJ )ˆ(
Multibody robot model:COM as a base coordinate system structure with decoupled COM dynamics.
[Space Robotics], [Wieber 2005, Hyon et al. 2006]
Passivity based compliance control (well suited for balancing)
Folie 70
ForceDistribution
Torque based balancing
Force Mapping
TorqueControl
RobotDynamics
Object ForceGeneration
IMU
cf
q
for orientation control and COM computation
Folie 71
Uncertain Foot Contact
[Ott, Roa, Humanoids 2011, best paper award]
Folie 72
Experiments on a Perturbation Platform
Leg perturbation setup
Movable elastic platform
Experimental evaluation of the robustness with respect to disturbances (frequency & amplitude) at the foot
Out of phase disturbance
synchronous disturbance2mm, up to 8 Hz
Comparisons
1) Impact experiments2) Whole body interaction3) Singularities
Comparison 1/31) Impact experiments
Position Based Control
Torque Based Control
Comparison 1/31) Impact experiments
Position Based Control
Torque Based Control
Observations:– Balancing after impact is comparable– Torque based controller does not control relative foot location
Comparison 2/33) Whole body interaction
Position Based Control
Torque Based Control
Comparison 2/33) Whole body interaction
Position Based Control
Torque Based Control
Observations:– Force sensor based controller depends on a reference frame– Torque based controller does not need information about the
point of contact
Comparison 3/34) Singular Configurations
Position Based Control
Torque Based Control
Comparison 3/34) Singular Configurations
Position Based Control
Torque Based Control
Observations:– Position based controller uses Inverse Kinematics, which
requires singularity handling– Torque based controller uses transposed Jacobian mapping, and
thus is not affected by singularities
Balancing: Summary
On flat floor both approaches allow for a compliant behavior
Torque based controller shows independence on precise ground contact (force mapping based on IMU information)
Admittance controller depends on a reference frame
Folie 82
Overview
Part I: Modeling
Part II: Balancing
Part III: Walking Control1. Walking pattern generation2. Feedback control
Robot control based on conceptual models
Footstep Generation
Pattern Generation
cx
pZMP-COMStabilizer
dxPos. Controlled
Robot
e.g. LQR Preview Control [Kajita, 2003]
Model Predictive Control [Wieber, 2006]
realtime
F
Mass concentrated model
p
c
x
Cart-Table Model [Kajita]Linear Inverted Pendulum Model [Sugihara]
pxzgx
p
c
px
xgzxp
xp
Mass concentrated model
p
c
xgzxp
x
Cart-Table Model [Kajita]
xp
xxx
x
xu
py
uxx
100
000100010
xgzy
01
)(/01)(
)(2/6/
)(100
102/1
)1( 2
32
kxgcky
kuT
TT
kxTTT
kx
z
Continuous time control model
Discrete time model
LQR Preview Control
How to use future information about the reference?
LQR Preview Control
)()()()()1(
kCxkykBukAxkx
)()()()()()(0
kuRkukxQkxkeQkeJ Tx
T
ke
T
)1()()()1()()(
)()()(
kukukukxkxkx
kykyke ref
Time-discrete system:
Cost function:
• Uses differential control input integral action.
• Uses the output error compared to reference signal.
)(kyrefReference output: Assume known for N future time steps
Assume (A,B) is stabilizable & (C,A) is detectable, and [**] [**] Ensures that the system has no transmission zero at z=1.
p
n
kykx
)()(
RQe , positive definite.
npIAB
Crank
0Assumptions:
LQR Preview Control
)()()()()1(
kCxkykBukAxkx
Time-discrete system:
)1(0
)()(
)(0)1(
)1(
kyI
kuB
CBkx
keA
CAIkx
keref
Modified system representation:)()()()(
kukukxkx
[**] Ensures that this system is stabilizable.
LQR Preview Control
)()()()()1(
kCxkykBukAxkx
Time-discrete system:
)1(0
)()(
)(0)1(
)1(
kyI
kuB
CBkx
keA
CAIkx
keref
Modified system representation:)()()()(
kukukxkx
How to handle future reference input?
System augmentation(for next N reference input values)
)(,),1()( Nkykykx refrefd
)(
0000100000010
)1( kxkx d
A
d
d
Dynamics of the new state:
[**] Ensures that this system is stabilizable.
LQR Preview Control
)(
0)()(
)(
0000
0
)1()1(
)1(
)()1(
kuBCB
kxkx
ke
AA
ICAI
kxkx
ke
kz
dd
kz
d
Modified system representation:
)()()()()()(0
kuRkukxQkxkeQkeJ Tx
T
ke
T
Cost function:
)()()(0000000
)(0
kuRkukzQQ
kzJ T
kx
eT
Standard LQR Design for augmented system!
Preview Control
)()(
)()()(
kxkx
keKKKkKzku
d
de
N
irefd
k
irefe ikyiKkxKiyiyKku
10)()()()()()(
Control law:
Example Application: Walking Pattern Generation
Simplified Model: Cart Table Model
xxx
x
xu
p … ZMP (Zero Moment Point)loosely speaking: Point on the sole where the reduced contact force is acting.
x … Position of the CoM
)(/01)(
)(2/6/
)(100
102/1
)1( 2
32
kxgzky
kuT
TT
kxTTT
kx
py
(Kajita 2003)
Example Application: Walking Pattern Generation
Footstep planning
Walking Pattern
Generator
CoM IK
Joint Position Control
dp x dq
Preview Control:T = 5 msN = 400
Q_x = zeros(3,3);Q_e = 1.0;R = 1e-6;
x
Example Application: Walking Pattern Generation
Footstep planning
Walking Pattern
Generator
CoM IK
Joint Position Control
dp x dq
Preview Control:T = 5 msN = 400
Q_x = zeros(3,3);Q_e = 1.0;R = 1e-6;
x
Properties of Preview Control
• Efficient implementation, controller design can be computed offline
• Allows to incorporate predictive information• ZMP contstraints are not considered explicitely• Trajectory based approach
Extensions– Model predictive control (handle zmp constraints explicitely,
optimization over a finite control horizon)– Trajectory generation feedback control– Dynamic filter
Feedback Stabilization
Footstep planning
Walking Pattern
Generator
Joint Position Control
Feedback Stabilization
dp dx dqrefx
refpInverse
Kinematics
swing foot trajectory
dqRobot
Dynamics
Forward Kinematics
ZMPComputation
p
x q
LR FF ,
position/velocity controlled robot
)()( refPrefdXrefd ppKxxKxx
Control law for stabilization:
Stability condition [*]: 0 PX KK
Stability condition [*]: [Choi, Kim, Oh, and You, Posture/Walking Control for Humanoid Robot Based on Kinematic Resolution of CoM Jacobian With Embedded Motion, TRO, 2007].
Dynamic filter
Footstep planning
Preview Control
dpdqrefx Inverse
KinematicsRobot
Dynamics
Correction of the error due to model simplificationRequires computation of the multi-body dynamics
p p
Preview Control
refx Inverse Kinematics
DLR-Biped
ZMP basierte Gangregelung
Präsentiert auf der Industriemesse Automatica, Juni 2010
Overview
Part I: Modeling
Part II: Balancing
Part III: Walking Control1. Walking pattern generation2. Feedback control
Walking Control
State of the art walking control for fully actuated robots
– Pattern Generator for desired CoM and ZMP motion– ZMP based Stabilizer
Footstep Generation
Pattern Generation
cx
pZMP-COMStabilizer
dx InverseKinematics
Position Control
dq
e.g. Preview Control [Kajita, 2003]Model Predictive Control [Wieber]
realtime
Folie 102
Walking Control used at DLR
c
p
[Englsberger, capture point control]
Walking StabilizationCore concept: Capture point controlGeneralization (3D)Stairs, etc …
Predictive Control (MPC)Reactive step adaptation
COMcapturepoint
xp
exp. stable
(Pratt 2006, Hof 2008) ),(
),(
x
xx
xx ,)(2 pxx
xx
pxx
00
open loopunstable
Folie 103
Walking Control used at DLR
c
p
[Englsberger, capture point control]
Walking StabilizationCore concept: Capture point controlGeneralization (3D)Stairs, etc …
Predictive Control (MPC)Reactive step adaptation
COMcapturepoint
xp
exp. stable
(Pratt 2006, Hof 2008) ),(
),(
x
xx
CP control
xx ,)(2 pxx
xx
pxx
00
Folie 104
COM
ZMP
Capture Point • COM velocity always points towards CP
• ZMP „pushes away“ the CP on a line
• COM follows CP
Capture Point Dynamik
Folie 105
COM
ZMP
Capture Point
Capture Point Dynamik
Folie 106
COM kinematics
Capture Point Control
xx ,
pCP control
[Englsberger, Ott, et. al., IROS 2011]
ZMPControl
RobotDynamics
CP
qTrajectoryGenerator d
ZMP projection
Folie 107
COM kinematics
Capture Point Control
xx ,
pCP control
[Englsberger, Ott, et. al., IROS 2011]
ZMPControl
RobotDynamics
CP
qTrajectoryGenerator d
ZMP projection
MPC [SYROCO 2012]
Folie 108
Position based ZMP Control
COM kinematics
xx ,
pCP control
ZMPControl
RobotDynamics
CP
qTrajectoryGenerator d
ZMP projection
MPC
)(2 pxx dp
Desired ZMP implies a desired force acting on the COM:
)(2dd pxMF
Position based force control [Roy&Whitcomb,2002]:
)( FFkx dfd )(2dfd ppMkx
Position based ZMP Control
Folie 109
COM kinematics
Capture Point Control
xx ,
pCP control
[Englsberger, Ott, et. al., IROS 2011]
ZMPControl
RobotDynamics
CP
qTrajectoryGenerator d
ZMP projection
MPC [SYROCO 2012]
Applications1) Vision based walking
– stereo vision (Hirschmüller)– visual SLAM (Chilian, Steidel)– online footstep planning, collaboration with N. Perrin (IIT)
Applications2) Optimized swingfoot trajectories: collaboration with H.
Kaminaga (Univ. Tokyo)
• stride length: 70 cm• speed: 0.5 m/s• kinematically optimized swingfoot trajectory
• stride length: 70 cm• speed: 0.5 m/s• kinematically optimized torso motion
(no angular momentum conversation! slippery)
Folie 113
Summary
Modeling
Full Body ModelsSimplified Models
Balancing
Torque based BalancingZMP based balancing
Walking
Pattern generationFeedback control
Part I: Part II: Part III:
Folie 114
Thank you very much for your attention!
Dr. MaximoRoa
JohannesEnglsberger
AlexanderWerner
GianlucaGarofalo
Dr. Christian Ott
BerndHenze