Upload
lythien
View
222
Download
3
Embed Size (px)
Citation preview
Trajectory Planning for Robots:The Challenges of Industrial Considerations
Joonyoung Kim1,2 and Elizabeth A. Croft2
1Hyundai Heavy Industries Co. Ltd., S. Korea2The University of British Columbia, Canada
1/19ICRA 2014
2
‐ Multi‐purpose and efficient (6DOF, easy to program, fast with good repeatability) ‐> provides various applications (different from other automated machines)
‐ Trends: more intelligence (vision & force sensors), safety (collision, human‐robot interaction)competitive market (cost reduction) → light weight design (less stiff) → the high motion performance is dependent on control methods (the main theme)
Hyundai Motors, Czech Republic LG Display, S. Korea
‐ 1.5 million industrial robots in the world (2013, IFR); used in large industries (automotive, display)
Introduction market, applications, and trends
ICRA 2014
ICRA 2014 3
Purpose
‐ To introduce a typical motion planning method for industrial robots
‐ To review trajectory planning methods proposed in the literature(related to motion performance of industrial robots)
‐ To review some common but the most important constraints for generic industrial robot controllers
‐ To explain why some of well‐known methods are not used in practice
Assumptions‐ 6DOF industrial robots‐ A generic robot controller (architecture, servo control loop)‐ A good trajectory:
; simple (implementation, maintenance); efficient (small CPU burden); being able to address important constraints; maximizes motion performance (high speed and high accuracy)
4
0S1S
2S
3S
S1 MOVE P, S= 100%, A=5, T=1
S2 MOVE L, S= 80%, A=3, T=1
S3 MOVE L, S=100%, A=0, T=1
Teach Pendant
Control module
Planning module(path & trajectory)
S1
S2
S3
Robot Moving
ProgramRunning
Trajectory Generator
Reference Input
Robot ControllerRobot Task Space
HHI (Hyundai Heavy Industries Co.., Ltd.) robot system
Introduction motion generation
ICRA 2014
5
fast (time-optimal)
Control
accurate tracking
Equal at all times!
planning
V
t
low speed
Introduction An ideal motion control for industrial robots
Path‐Invariant Time‐Optimal Motion
Requirements for trajectories‐ time‐optimal‐ generate a path independent of trajectories‐ band‐limited
6
HHI’s HS165 (6DOF, payload: 165kg)
Task space
joint space
3 (3)SO
6
vector space curved space (Lie Group)(not a vector space)
3: , (3)nf SO
Multi‐dimensional nonlinear map
f
n 1f
3 (3)SO
Challenges kinematics
Multi‐dimensional, nonlinear
ICRA 2014
‐ Need to address both positions and orientations‐ Orientations: much more complex than positions
Kinematic constraints: path, maximum speed/acceleration/jerk
7
Multi‐dimensional, nonlinear, highly coupled
12( ) ( , ) ( ) τ M q q + B q q q + N q
m mJ
mq rqr
k rJ
mb
each joint
Equation of motion
Inertia‐ position‐dependent‐ coupled
Coriolis‐Centrifugal and friction‐ Pos. & vel. dependent‐ coupled
Joint compliance,gravity‐ position dependent‐ coupled
Natural frequencies at steady‐state: 3 ~ 25 Hz (1~2nd mode) (very low, position dependent)
Joint (link) positions: not measurable!
Harmonic Drive (r=100 ~ 200)
Challenges dynamics
ICRA 2014
Constraints‐maximum torques (hard constraint)‐ not exciting natural modes
8
Complex geometry + complex dynamics (too complicated)
Divide & Conquer (Planning / Control)
‐ Increase speed; time‐optimal trajectory planning
‐ minimize vibrations; smooth motion trajectory planning
‐ Track accurately, respond quickly, ‐ Obtain stable motions
; various control methods (linear, nonlinear, SISO, MIMO)
Planning Task (goals) Control Task (goals)
Traditional Approach
ICRA 2014
9
Time‐Optimal Trajectory Planning
Shin and McKay (1985) Bobrow et al. (1985)
Constantinescu and Croft (2000)
3( ) [ , ]{ | 0 1}
f ss s
A parameterized path
s
s
( , )o os s ( , )f fs sforward in time backward in time
Phase Plane
switching point
Find the time‐optimal trajectory
Pros: problem simplified by the path parameter very close to true time‐optimal trajectories
Cons: high CPU burden, infinite jerk (in general),dynamic model never perfect
not considered
Well‐Known Methods Time‐optimal trajectories
ICRA 2014
EOM ( ) ( , ) ns s s s τ m c
max max( ) ( , )s s s s τ m c τ Constraints
max,max,
( , )min
( )i i
acci
c s ss
m s
10
Meckl and Woods (1985)Singer (1990)
Macfalane and Croft (2003)Kroger and Walh (2010)
‐ Argument ; time‐optimal: too complicated, high frequencies (advanced control methods required); smooth trajectories can obtain faster settling times due to small residual vibrations.
Pros: small oscillation, no special controllers neededCons: slow (no dynamics)
too many cases for non‐zero boundary conditionscannot completely suppress high freq.natural freq. must be known
0 0.2 0.4 0.6 0.8 1 1.2 1.40
200
400
600
800
1000
1200
1400t-dp trajectory
t [s]
dp [mm/s]
0 5 10 15 20 25 300
1
2
3
4
Frequency (Hz)
|Y(f)
|
TVP
S‐curveTVP
S‐curve
Well‐Known Methods Smooth‐Trajectories
ICRA 2014
Challenges online planning
Online trajectory planning: A robot’s trajectory must be planned (replanned) during motion for any (stationary) target given kinematic and dynamic constraints.
Non-zero boundary conditions
S1
replan
torquesaturation
S2 input
11
Trapezoid Velocity
S‐curves or higher polynomials?‐ too many cases to consider‐ prone to generate s/w errors‐> difficult to implement and maintain
ICRA 2014
bad good
V 10 %
V 100 %
V 10 %
V 100 %
Speed change = collision
12
Challenges path and speed
ICRA 2014
Path must be the same regardless of speed changes.
S1 S2
S3Stop/Restart inManual Playback Mode
StopMinimum distance to the original path
Robot position at power‐failure
Motion command at power‐failure
Return to the path
S1
S2
Position after Power‐failure
Power‐failure during motion
Path unchanged
return
jogging
Planned path
Real path
13
Challenges path invariance for all operation conditions
ICRA 2014
A programmed path must not be changed under any circumstances unless such arequest is made by the user.
No powerfailure
Power failure Path Recovery
Path RecoveryImagine you are in charge of this factory!And there is a chance of black-out!
14ICRA 2014
15
TVP S‐curve
0 t
( )r t
aT aT
A A
0 t
( )r t
jT jT
J J
2( ) 1 ajTTVP
AR j e
3( ) 1 jjTS curve
JR j ej
Frequency Analysis via Fourier Transform
( ) 0m x k x r
2
2 2
( ) ( )( )
n
n
X j H jR j
2
2 2 2
( ) ( ) ( )
1 a
TVP TVP
jTn
n
X j H j R j
A e
2
3 2 2
( ) ( ) ( )
1 j
S curve S curve
jTn
n
X j H j R j
J ej
‐ Big A, Big J ‐> large Oscillation‐ TVP Larger vibration than S‐curve‐ Vibration can be 0 at some ‐ S‐curves still have high freq.
0 5 10 15 20 25 30 35 40 45 50-160
-140
-120
-100
-80
-60
-40
-20
0
frequency [Hz]
power
spe
ctru
m o
f R(jw
)
,a jT T
FRF of TVP
Trajectories Vs. Residual Vibrations
ICRA 2014
16
,1rq
,2rq1L
,2cL
2L
,1cL
2m
1m
X
Y
,1cJ
,2cJm mJ
mq rqr
k rJ
mb
Parameter Joint 1 Joint 2 Parameter Joint 1 Joint 2
Mass [kg] 80 170 k [Nm/rad] 1010700 505350
Jr [kg.m2] 9.97 13.7 bm [Nm s/rad] 0.0281 0.0088
L [m] 0.87 1.05 Max. motor torque [Nm] 49 35.86
Lc [m] 0.38 0.2 r (gear ratio) 201 145
Robot: 2DOF industrial Robot with flexible jointsInitial position: qr0 = [90, ‐90] [deg]Natural frequencies: 7 Hz, 21 Hz at qr0
Simulation dynamic model
Joint model
ICRA 2014
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 189
90
91
92
93
94
95
96
97
time [s]
qr1 [d
eg]
TVP (infinite jerk)
S‐curve (large jerk – tuned at the lowest natural frequency, 7 Hz)
S‐curve (small jerk)
0 0.2 0.4 0.6 0.8 1-50
0
50
m,cm
d 1 [Nm
]
Torque saturation
Oscillation at 7 Hz (1st mode)
‐ Control performance is affected by trajectories.‐ Trajectories must be band‐limited.‐ Such limitation of the band cannot be determined arbitrarily.
Simulation trajectories Vs. vibrations
Controller: Decentralized PD controllers
First joint position
17ICRA 2014
Simulation trajectories Vs. controller
0 0.1 0.2 0.3 0.4 0.588
90
92
94
96
time [s]
q r1 [d
eg]
0 0.1 0.2 0.3 0.4 0.5-20
0
20
40
60
80
time [s]
dqr1
[deg
/s]
0 0.1 0.2 0.3 0.4 0.5-96
-94
-92
-90
-88
time [s]
q r2 [d
eg]
0 0.1 0.2 0.3 0.4 0.5-80
-60
-40
-20
0
20
time [s]
dqr2
[deg
/s]
‐ Large tracking errors‐ Large oscillations ‐> Decentralized and linear control methods do not work well.‐> A more advanced control method is needed.
18
Trajectory: TVPController: decentralized
& linear full‐state feedback (linear state‐observer)
fixed gain by pole‐placement
19
Conclusions
‐ Trajectory planning for industrial robots is challenging.; online planning, safety, efficiency, complex kinematics and dynamics
‐ Traditionally, a good trajectory is regarded as either time‐optimal or smooth.
‐ Time‐optimal trajectories; high CPU burden and high frequency components
‐ Smooth trajectories; too slow, still complex (high order polynomials), vague jerk limitation
‐ What is the best trajectory for industrial robots?; not clear but TVP is still widely used!; it depends on servo controller.
ICRA 2014