Professional Documents
Culture Documents
TheDeltaParallelRobot KinematicsSolutionsR.L.Williams PDF
TheDeltaParallelRobot KinematicsSolutionsR.L.Williams PDF
The Delta Robot has 4-degrees-of-freedom (dof), 3-dof for XYZ translation, plus a fourth inner
leg to control a single rotational freedom at the end-effector platform (about the axis perpendicular to the
platform). The remainder of this document will focus only on the 3-dof XYZ translation-only Delta
Robot since that is being widely applied by 3D printers and Arduino hobbyists.
Presented is a description of the 3-dof Delta Robot, followed by kinematics analysis including
analytical solutions for the inverse position kinematics problem and the forward position kinematics
problem, and then examples for both, snapshots and trajectories. The velocity equations are also derived
This is presented for both revolute-input and prismatic-input Delta Robots.
1
R. Clavel, 1991, “Conception d'un robot parallèle rapide à 4 degrés de liberté”, Ph.D. Thesis, EPFL, Lausanne, Switzerland.
2
R. Clavel, 1990, “Device for the Movement and Positioning of an Element in Space”, U.S. Patent No. 4,976,582.
2
Table of Contents
The three-dof Delta Robot is capable of XYZ translational control of its moving platform within
its workspace. Viewing the three identical RUU chains as legs, points Bi , i 1, 2,3 are the hips, points
Ai , i 1, 2,3 are the knees, and points Pi , i 1,2,3 are the ankles. The side length of the base
equilateral triangle is sB and the side length of the moving platform equilateral triangle is sP. The
moving platform equilateral triangle is inverted with respect to the base equilateral triangle as shown, in
a constant orientation.
The fixed base Cartesian reference frame is {B}, whose origin is located in the center of the base
equilateral triangle. The moving platform Cartesian reference frame is {P}, whose origin is located in
the center of the platform equilateral triangle. The orientation of {P} is always identical to the
orientation of {B} so rotation matrix P R I3 is constant. The joint variables are Θ 1 2 3 , and
B T
the Cartesian variables are PP x y z . The design shown has high symmetry, with three upper
B T
leg lengths L and three lower lengths l (the parallelogram four-bar mechanisms major lengths).
The Delta Robot fixed base and platform geometric details are shown on the next page.
5
The fixed-base revolute joint points Bi are constant in the base frame {B} and the platform-fixed
U-joint connection points Pi are constant in the base frame {P}:
3 3
wB wB
0 2 2
1 1
B
B1 wB B
B2 wB B
B3 wB
0 2 2
0 0
sP sP
0 2 2
P
P1 u P
P
P2 wP P
P3 wP
0 0 0
sB sB
2 0 2
B
b1 wB B
b2 uB
B
b3 wB
0 0 0
where:
3 3 3 3
wB sB uB sB wP sP uP sP
6 3 6 3
The model values above are for a specific commercial delta robot, the ABB FlexPicker IRB 360-1/1600,
scaled from a figure (new.abb.com/products). Though Delta Robot symmetry is assumed, the following
methods may be adapted to the general case.
7
M 6( N 1) 5J1 4 J 2 3J 3
where:
M is the mobility, or number of degrees-of-freedom
N is the total number of links, including ground
J1 is the number of one-dof joints
J2 is the number of two-dof joints
J3 is the number of three-dof joints
N 17
J1 21 M 6(17 1) 5(21) 4(0) 3(0)
J2 0 M 9 dof
J3 0
As often happens, the Kutzbach equation fails because the result must obviously be 3-dof. This result
predicts the Delta is a severely overconstrained statically indeterminate structure, which is incorrect.
The Kutzbach equation knows nothing about special geometry – in the Delta Robot case, there
are three parallel four-bar mechanisms. The overall robot would work kinematically identically to the
original Delta Robot if we removed one of the long parallel four-bar mechanism links, along with two
revolute joints each. With this equivalent case, the Kutzbach equation yields:
N 14
J1 15 M 6(14 1) 5(15) 4(0) 3(0)
J2 0 M 3 dof
J3 0
which is correct. An alternative approach to calculating the Delta Robot mobility is by ignoring the
three parallel four-bar mechanisms, replacing each with a single link instead. In this we must count a
universal joint at either end of this virtual link. This approach follows the simplified Delta Robot
naming convention 3-RUU. The Kutzbach equation for this case also succeeds:
8
N 8
J1 3 M 6(8 1) 5(3) 4(6) 3(0)
J2 6 M 3 dof
J3 0
Either of the second two approaches works. The author prefers the former since it is closer to the actual
Delta Robot design.
9
B
B i B L i B l i B PP BP R P Pi B PP P Pi i 1,2,3
li l
B
i
B
PP P Pi B B i B Li i 1,2,3
It will be more convenient to square both sides of the constraint equations above to avoid the square-root
in the Euclidean norms:
l i 1,2,3
2
li2 B
i lix2 liy2 liz2
Again, the Cartesian variables are PP x y z . The constant vector values for points Pi
B T
Θ 1 2 3 :
T
3 3
L cos2 L cos 3
0 2 2
1 1
B
L1 L cos1 B
L2 L cos 2 B
L3 L cos 3
L sin 2 2
1
L sin 2 L sin 3
Substituting all above values into the vector-loop closure equations yields:
3 3
x L cos 2 b x L cos 3 b
x 2 2
B
l1 y L cos 1 a l 2 y 2 L cos 2 c
B 1
l3 y 2 L cos 3 c
B 1
z L sin
1 z L sin 2 z L sin 3
11
a wB uP
sP 3
where: b wB
2 2
1
c wP wB
2
And the three constraint equations yield the kinematics equations for the Delta Robot:
2 L ( y a ) cos 1 2 zL sin 1 x 2 y 2 z 2 a 2 L2 2 ya l 2 0
L ( 3( x b ) y c ) cos 2 2 zL sin 2 x 2 y 2 z 2 b 2 c 2 L2 2 xb 2 yc l 2 0
L ( 3( x b) y c ) cos 3 2 zL sin 3 x 2 y 2 z 2 b 2 c 2 L2 2 xb 2 yc l 2 0
The three absolute vector knee points are found using Ai Bi Li , i 1,2,3 :
B B B
3 3
( wB L cos 2 ) ( wB L cos3 )
0 2 2
1 1
B
A1 wB L cos1 B
A2 (wB L cos 2 ) B
A3 ( wB L cos 3 )
L sin1 2 2
L sin 2 L sin 3
The 3-dof Delta Robot inverse position kinematics (IPK) problem is stated: Given the Cartesian
position of the moving platform control point (the origin of {P}), PP x y z , calculate the three
B T
required actuated revolute joint angles Θ 1 2 3 . The IPK solution for parallel robots is often
T
straightforward; the IPK solution for the Delta Robot is not trivial, but can be found analytically.
Referring to the Delta Robot kinematic diagram above, the IPK problem can be solved independently for
each of the three RUU legs. Geometrically, each leg IPK solution is the intersection between a known
circle (radius L, centered on the base triangle R joint point Bi ) and a known sphere (radius l, centered
B
where:
12
E1 2 L ( y a )
F1 2 zL
G1 x 2 y 2 z 2 a 2 L2 2 ya l 2
E2 L ( 3( x b) y c ) E3 L( 3( x b) y c)
F2 2 zL F3 2 zL
G2 x 2 y 2 z 2 b 2 c 2 L2 2( xb yc ) l 2 G3 x 2 y 2 z 2 b 2 c 2 L2 2( xb yc) l 2
The equation Ei cos i Fi sin i Gi 0 appears a lot in robot and mechanism kinematics and is readily
solved using the Tangent Half-Angle Substitution.
i 1 ti2 2ti
If we define ti tan then cos i and sin i
2 1 ti2 1 ti2
Fi Ei2 Fi 2 Gi2
(Gi Ei )t (2Fi )ti (Gi Ei ) 0
2
quadratic formula: ti1,2
Gi Ei
i
i 2 tan 1 (ti )
Two i solutions result from the in the quadratic formula. Both are correct since there are two valid
solutions – knee left and knee right. This yields two IPK branch solutions for each leg of the Delta
Robot, for a total of 8 possible valid solutions. Generally the one solution with all knees kinked out
instead of in will be chosen.
The 3-dof Delta Robot forward position kinematics (FPK) problem is stated: Given the three
actuated joint angles Θ 1 2 3 , calculate the resulting Cartesian position of the moving platform
T
control point (the origin of {P}), PP x y z . The FPK solution for parallel robots is generally
B T
very difficult. It requires the solution of multiple coupled nonlinear algebraic equations, from the three
constraint equations applied to the vector loop-closure equations (derived previously). Multiple valid
solutions generally result.
13
Thanks to the translation-only motion of the 3-dof Delta Robot, there is a straightforward
analytical solution for which the correct solution set is easily chosen. Since Θ 1 2 3 are given,
T
we calculate the three absolute vector knee points using Ai Bi Li , i 1,2,3 . Referring to the
B B B
Delta Robot FPK diagram below, since we know that the moving platform orientation is constant,
always horizontal with PB R I3 , we define three virtual sphere centers Aiv Ai Pi , i 1,2,3 :
B B P
3 s 3 s
( wB L cos 2 ) P ( wB L cos 3 ) P
0 2 2 2 2
1 1
B
A1v wB L cos1 uP B
A2v ( wB L cos 2 ) wP B
A3v ( wB L cos 3 ) wP
2 2
L sin 1 L sin 2 L sin 3
and then the Delta Robot FPK solution is the intersection point of three known spheres. Let a sphere be
referred as a vector center point {c} and scalar radius r, ({c}, r). Therefore, the FPK unknown point
P is the intersection of the three known spheres:
B
P
( B
A1v , l) ( B
A2v , l) ( B
A3v , l).
Appendix A presents an analytical solution for the intersection point of the three given spheres,
from Williams et al.3 This solution also requires the solving of coupled transcendental equations. The
appendix presents the equations and analytical solution methods, and then discusses imaginary solutions,
singularities, and multiple solutions that can plague the algorithm, but all turn out to be no problem in
this design.
In particular, with this existing three-spheres-intersection algorithm, if all three given sphere
centers B
Aiv have the same Z height (a common case for the Delta Robot), there will be an
algorithmic singularity preventing a successful solution (dividing by zero). One way to fix this problem
is to simple rotate coordinates so all B
Aiv Z values are no longer the same, taking care to reverse this
coordinate transformation after the solution is accomplished. However, we present another solution
(Appendix B) for the intersection of three spheres assuming that all three sphere Z heights are identical,
to be used in place of the primary solution when necessary.
Another applicable problem to be addressed is that the intersection of three spheres yields two
solutions in general (only one solution if the spheres meet tangentially, and zero solutions if the center
distance is too great for the given sphere radii l – in this latter case the solution is imaginary and the
input data is not consistent with Delta Robot assembly). The spheres-intersection algorithm calculates
both solution sets and it is possible to automatically make the computer choose the correct solution by
ensuring it is below the base triangle rather than above it.
This three-spheres-intersection approach to the FPK for the Delta Robot yields results identical
to solving the three kinematics equations for PP x y z given Θ 1 2 3 .
B T T
3
R.L. Williams II, J.S. Albus, and R.V. Bostelman, 2004, “3D Cable-Based Cartesian Metrology
System”, Journal of Robotic Systems, 21(5): 237-257.
15
The revolute-input Delta Robot velocity kinematics equations come from the first time derivative
of the three position constraint equations presented earlier:
2 Ly cos 1 2 L ( y a )1 sin 1 2 Lz sin 1 2 Lz1 cos 1 2 xx 2( y a ) y 2 zz 0
L ( 3 x y ) cos 2 L ( 3( x b) y c )2 sin 2 2 Lz sin 2 2 Lz2 cos 2 2( x b) x 2( y c ) y 2 zz 0
L( 3 x y ) cos 3 L( 3( x b) y c)3 sin 3 2 Lz sin 3 2 Lz3 cos 3 2( x b) x 2( y c ) y 2 zz 0
Re-written:
xx ( y a ) y Ly cos 1 zz Lz sin 1 L( y a )1 sin 1 Lz1 cos 1
2( x b) x 2( y c) y L( 3 x y ) cos 2 2 zz 2 Lz sin 2 L( 3( x b) y c)2 sin 2 2 Lz2 cos 2
2( x b) x 2( y c ) y L( 3 x y ) cos 2 zz 2 Lz sin L( 3( x b) y c) sin 2 Lz cos
3 3 3 3 3 3
A X = B Θ
where:
The moving platform control point {P} traces an XY circle of center 0 0 1 m and radius 0.5
T
m. At the same time, the Z displacement goes through 2 complete sine wave motions centered on Z 1
m with a 0.2 m amplitude.
This IPK trajectory, at the end of motion, is pictured along with the simulated Delta Parallel
Robot, in the MATLAB graphics below.
P 0.108 0.180 1.244 , the IPK solution calculated Θ 10 20 30 . When given
B
P
T T
Θ 20.5 20.5 20.5 , the FPK solution calculated P 0 0 0.9 and when given
T
B T
P
Θ 47.5 11.6 21.4 , the FPK solution calculated P 0.3 0.5 1.1 .
T
B T
P
20
The IPK solution is more useful for Delta Robot control, to specify where the tool should be in
XYZ. This FPK trajectory example is just for demonstration purposes, not yielding any useful robot
motion. Simultaneously the three revolute joint angles are commanded as follows:
1 (t ) max sin(t )
2 (t ) max sin(2t )
3 (t ) max sin(3t )
where max 45 and t proceeds from 0 to 2 in 100 steps. This closed FPK trajectory, at the end of
motion, is pictured along with the simulated Delta Parallel Robot, in the MATLAB graphics below
FPK Trajectory
21
As shown below, the 3-dof prismatic-input Delta Robot is composed of three identical PUU legs
in parallel between the top fixed base and the bottom moving end-effector platform. The control
variables are Li , i 1, 2,3 . In this model a positive change in Li is downward, in the ZB direction.
The three-dof Delta Robot is again capable of XYZ translational control of its moving platform within its
workspace. The joint variables are L L1 L2 L3 , and the Cartesian variables are B PP x y z .
T T
This robot is also known as the Linear Delta Robot or the Linear-Rail Delta Robot. The constant
geometry (base points, platform vertices, etc.) presented earlier for the revolute-input Delta robot still
apply to the prismatic-input Delta Robot. Also, the Mobility (dof) calculations for the prismatic-input
case are identical to that presented for the revolute-input case, yielding M = 3. The Prismatic-Input
Delta Robot fixed base and platform geometric details are shown on the next page.
23
The fixed-base prismatic joint points Bi are constant in the base frame {B} and the platform-
fixed U-joint connection points Pi are constant in the base frame {P}. Both lie on the vertices of
equilateral triangles, of sides SB and sP, respectively.
SB SB
2 2 0
P
B1 WB P
B2 WB P
B 3 U B
0 0 0
sP sP
2 2 0
P
P1 wP P
P2 wP P
P3 u P
0 0
0
The vertices of the fixed-based equilateral triangle are not directly used in the kinematics
equations; they are used in MATLAB graphics for the support frame:
sb sb
2 2 0
B
b1 wb B
b2 wb B
b3 ub
0 0
0
where:
3 3 3 3
WB SB UB SB wb sb ub sb
6 3 6 3
3 3
wP sP uP sP
6 3
25
The model values above are for a specific commercial prismatic-input delta robot 3D printer, the
Delta Maker (deltamaker.com). Though Delta Robot symmetry is assumed, the following methods may
be adapted to the general case for the prismatic-input delta robot.
Note that each individual prismatic length limits are 67 Li 479 mm; however, for all three
prismatic lengths equal, Li 358 mm is the maximum extent to avoid motion through the 3D printing
surface.
26
forums.reprap.org
deltamaker.com
27
B
B i B L i B l i B PP BP R P Pi B PP P Pi i 1,2,3
The three applicable constraints state that the lower leg lengths must have the correct, constant length l
(the virtual length through the center of each parallelogram):
li l
B
i
B
PP P Pi B B i B Li i 1,2,3
It will be more convenient to square both sides of the constraint equations above to avoid the square-root
in the Euclidean norms:
l i 1,2,3
2
li2 B
i lix2 liy2 liz2
Again, the Cartesian variables are PP x y z . The constant vector values for points Pi
B T
L L1 L2 L3 ; these formulas are much simpler than those for the rotational-input case presented
T
earlier:
0
B
Li 0 i 1,2,3
L
i
Substituting all above values into the vector-loop closure equations yields:
xa xa x
B
l1 y b B
l2 y b B
l3 y c
z L z L
1 z L2 3
S B sP
a
2 2
where: b WB wP
c uP U B
And the three constraint equations yield the kinematics equations for the prismatic-input Delta Robot:
28
The three absolute vector knee points are found using Ai Bi Li , i 1,2,3 :
B B B
SB SB
2 2 0
B
A1 WB B
A2 WB B
A3 U B
L L L
1 2 3
The 3-dof prismatic-input Delta Robot inverse position kinematics (IPK) problem is stated:
Given the Cartesian position of the moving platform control point (the origin of {P}), PP x y z ,
B T
calculate the three required actuated prismatic joint angles L L1 L2 L3 . The IPK solution for the
T
prismatic-input Delta Robot is much simpler than that for the revolute-input Delta robot presented
earlier and is easily found analytically. Referring to the prismatic-input Delta Robot kinematic diagram
above, the IPK problem can be solved independently for each of the three PUU legs. Geometrically,
each leg IPK solution is the intersection between a vertical line of unknown length Li (passing through
base point Bi ) and a known sphere (radius l, centered on the moving platform vertex Pi ).
B P
where:
C1 x 2 y 2 z 2 a 2 b 2 2ax 2by l 2
C2 x 2 y 2 z 2 a 2 b 2 2ax 2by l 2
C3 x 2 y 2 z 2 c 2 2cy l 2
So we simply have three independent quadratic equations to solve for the prismatic-length inputs Li, for
each leg independently, where Ai = 1, Bi = 2z, and the Ci are given above. The IPK solution simplifies
quite nicely:
Li z z 2 Ci i 1,2,3
29
Two Li solutions result from the in the quadratic formula. These solutions can be referred to as knee
up and knee down for each leg. This yields two IPK branch solutions for each leg of the prismatic-input
Delta Robot, for a total of 8 possible solutions. Generally the one overall solution with all knees up will
be chosen.
When z Ci , the solution for Li is imaginary. This case should never occur in theory since the
2
prismatic joint can extend as far as needed to maintain a real solution for each leg. However, in practice,
there are of course prismatic joint limits. When z Ci , the two solution branches (knee up and knee
2
The IPK input xyz is for the moving platform geometric center. When the desired control point
is offset from the center (as in the case of many 3D printers), an initial transformation is required prior to
implementing the IPK solution:
1 0 0 OX 1 0 0 OX
0 1 0 OY 0 1 0 OY
where NT
P and NT
P 1
0 0 1 0 0 0 1 0
0 0 0 1 0 0 0 1
This is a very simple transformation since the prismatic-input Delta Robot allows only translational
motion, with P R N R I3 . To save a lot of computations with 1s and 0s, simply subtract OX and
B B
OY from the x and y components, respectively, of the given (xyz)N to obtain the IPK moving-platform-
center input xyz. N stands for nozzle in a 3D printer. The z component is unchanged in this
transformation.
30
The 3-dof prismatic-input Delta Robot forward position kinematics (FPK) problem is stated:
Given the three actuated joint angles L L1 L2 L3 , calculate the resulting Cartesian position of the
T
Thanks to the translation-only motion of the 3-dof Delta Robot, there is a straightforward
analytical solution for which the correct solution set is easily chosen. Since L L1 L2 L3 are
T
given, we calculate the three absolute vector knee points using Ai Bi Li , i 1,2,3 . Referring to
B B B
the prismatic-input Delta Robot FPK diagram below, since we know that the moving platform
orientation is constant, always horizontal with P R I3 , we define three virtual sphere centers
B
B
Aiv B Ai P Pi , i 1,2,3 :
SB sP S B sP
2 2 2 2 0
B
A1v WB wP B
A2v WB wP B
A3v U B uP
L L L
1
2
3
and then the prismatic-input Delta Robot FPK solution is the intersection point of three known spheres.
Let a sphere be referred as a vector center point {c} and scalar radius r, ({c}, r). Therefore, the FPK
unknown point P is the intersection of the three known spheres:
B
P
( B
A1v , l) ( B
A2v , l) ( B
A3v , l).
31
Appendix A presents an analytical solution for the intersection point of the three given spheres,
from Williams et al.4 This solution also requires the solving of coupled transcendental equations. The
appendix presents the equations and analytical solution methods, and then discusses imaginary solutions,
singularities, and multiple solutions that can plague the algorithm, but all turn out to be no problem in
this design.
In particular, with this existing three-spheres-intersection algorithm, if all three given sphere
centers B
Aiv have the same Z height (a common case for the Delta Robot), there will be an
algorithmic singularity preventing a successful solution (dividing by zero). One way to fix this problem
is to simple rotate coordinates so all B
Aiv Z values are no longer the same, taking care to reverse this
coordinate transformation after the solution is accomplished. However, we present another solution
4
R.L. Williams II, J.S. Albus, and R.V. Bostelman, 2004, “3D Cable-Based Cartesian Metrology
System”, Journal of Robotic Systems, 21(5): 237-257.
32
(Appendix B) for the intersection of three spheres assuming that all three sphere Z heights are identical,
to be used in place of the primary solution when necessary.
Another applicable problem to be addressed is that the intersection of three spheres yields two
solutions in general (only one solution if the spheres meet tangentially, and zero solutions if the center
distance is too great for the given sphere radii l – in this latter case the solution is imaginary and the
input data is not consistent with prismatic-input Delta Robot assembly). The spheres-intersection
algorithm calculates both solution sets and it is possible to automatically make the computer choose the
correct solution by ensuring it is below the base triangle rather than above it.
The FPK solution xyz is for the moving platform geometric center. When the desired control
point is offset from the center (as in the case of many 3D printers), a further transformation is required
after to implementing the FPK solution:
where NT was given previously with the IPK solution. This is a very simple transformation since the
P
prismatic-input Delta Robot allows only translational motion, with P R N R I3 . To save a lot of
B B
computations with 1s and 0s, simply add OX and OY to the x and y components, respectively, of the FPK
moving-platform-center solution xyz, to obtain the desired FPK solution (xyz)N. The z component is
unchanged in this transformation.
33
This three-spheres-intersection approach to the FPK for the prismatic-input Delta Robot yields
results identical to solving the three kinematics equations for PP x y z given L L1 L2 L3 .
B T T
Since the three constraint equations for the prismatic-input Delta Robot are much simpler than those for
the revolute-input Delta Robot, we now present an alternative FPK analytical solution. The three
constraint equations are repeated below:
Subtracting the second equation from the first equation yields a linear equation, expressing x as a
function of z only:
x f ( z ) dz e
where:
L2 L1 L22 L12
d e
2a 4a
Further, subtracting the third equation from the first equation and substituting x f ( z ) from above
yields another linear equation, expressing y as a function of z only:
y g( z ) Dz E
where:
L L ad c 2 a 2 b 2 2ae L23 L12
D 3 1 E
bc 2(b c)
Substituting both x f ( z ) and y g ( z ) into the third equation yields a single equation in one
unknown z, a quadratic polynomial:
Az 2 Bz C 0
where:
A d 2 D2 1
B 2(de DE cD L3 )
C e2 E 2 c 2 2cE L23 l 2
B B 2 4 AC
z1,2
2A
There are two possible solution sets ( x1 , y1 , z1 ) and ( x2 , y2 , z2 ) , due to the in the quadratic formula.
Generally only one solution will be used, the one where xyz is below the top-mounted base triangle.
35
The prismatic-input Delta Robot velocity kinematics equations come from the first time
derivative of the three position constraint equations presented earlier:
( x a ) x ( y b) y ( z L1 ) z ( z L1 ) L1
( x a ) x ( y b) y ( z L ) z ( z L ) L
2 2 2
xx ( y c ) y ( z L3 ) z ( z L3 ) L3
A X = B L
x a yb z L1 x z L1 0 0 L1
x a yb z L2 y 0
z L2 0 L2
x yc z L3 z 0 0 z L3 L3
X = A B L
1
L B A X
1
xa yb
1
L1 z L1 z L1 x
xa yb
L2 1 y
L z L2 z L2 z
3 x yc
1
z L3 z L3
36
0.4
0 0.2
-0.2
Z
0
Y
-0.4
-0.6
-0.2
0.2 0.2
0 0
-0.2 -0.2 -0.4
Y -0.4 -0.4 X -0.4 -0.2 0 0.2
X
0 0
-0.2 -0.2
Z
-0.4 -0.4
-0.6 -0.6
0.2 0 -0.2 -0.4 -0.4 -0.2 0 0.2
Y X
General Inverse Position Snapshot Example
37
The moving platform control point {P} traces an XY circle of center 0 0 0.5 m and radius
T
0.1 m. At the same time, the Z displacement goes through 2 complete sine wave motions centered on
Z 1 m with a 0.05 m amplitude.
This IPK trajectory, at the end of motion, is pictured along with the simulated Delta Parallel
Robot, in the MATLAB graphics below.
0.4
0 0.2
-0.2
Z
Y
-0.4
-0.6
-0.2
0.2 0.2
0 0
-0.2 -0.2 -0.4
Y -0.4 -0.4 X -0.4 -0.2 0 0.2
X
0 0
-0.2 -0.2
Z
-0.4 -0.4
-0.6 -0.6
0.2 0 -0.2 -0.4 -0.4 -0.2 0 0.2
Y X
XY Circular Trajectory with Z sine wave
38
0.1
-0.1
X (m)
-0.2 x
P
y
P
z
-0.3 P
-0.4
-0.5
0 10 20 30 40 50 60 70
Steps
Commanded IPK Cartesian Positions
0.36
0.34
L1
0.32
L2
0.3 L3
Joint Lengths (m)
0.28
0.26
0.24
0.22
0.2
0.18
0 10 20 30 40 50 60 70
Time Steps
Calculated IPK Joint Angles
39
General Position. Given L 0.14 0.15 0.16 m, the calculated FPK results are (the admissible
T
0.4
0 0.2
-0.2
Z
Y
-0.4
-0.6
-0.2
0.2
0
-0.2 0 0.2
-0.2 -0.4
Y -0.4 -0.4 X -0.4 -0.2 0 0.2
X
0 0
-0.2 -0.2
Z
-0.4 -0.4
-0.6 -0.6
0.2 0 -0.2 -0.4 -0.4 -0.2 0 0.2
Y X
General Forward Position Snapshot Example
given P 0.0215
B
P 0.0363 0.4012 , the IPK solution calculated L 0.14 0.15 0.16 .
T T
when given L 0.1664 0.1516 0.1384 , the FPK solution calculated P 0.03 0.05 0.40 .
T B T
P
40
Acknowledgements
This work was initiated during the last week of the author’s sabbatical from Ohio University at
the University of Puerto Rico (UPR), Mayaguez, during Fall Semester 2014. The author gratefully
acknowledges financial support from Ricky Valentin, UPR ME chair, and the Russ College of
Engineering & Technology at Ohio University.
41
Appendices
Appendix A. Three-Spheres Intersection Algorithm
We now derive the equations and solution for the intersection point of three given spheres. This
solution is required in the forward pose kinematics solution for many cable-suspended robots and other
parallel robots. Let us assume that the three given spheres are ( c1 ,r1), ( c 2 ,r2), and ( c 3 ,r3). That is,
center vectors c1 x1 y1 z1T , c 2 x2 y2 z2 T , c3 x3 y3 z3 T , and radii r1, r2, and r3 are
known (The three sphere center vectors must be expressed in the same frame, {0} in this appendix; the
answer will be in the same coordinate frame). The equations of the three spheres are:
x x1 2 y y1 2 z z1 2 r12
x x 2 2 y y 2 2 z z 2 2 r22 (A.1)
x x3 2 y y3 2 z z 3 2 r32
Equations (A.1) are three coupled nonlinear equations in the three unknowns x, y, and z. The
solution will yield the intersection point P x y zT . The solution approach is to expand equations
(A.1) and combine them in ways so that we obtain x f y and z f y ; we then substitute these
functions into one of the original sphere equations and obtain one quadratic equation in y only. This can
be readily solved, yielding two y solutions. Then we again use x f y and z f y to determine the
remaining unknowns x and z, one for each y solution. Let us now derive this solution.
First, expand equations (A.1) by squaring all left side terms. Then subtract the third from the
first and the third from the second equations, yielding (notice this eliminates the squares of the
unknowns):
where:
a11 2x3 x1 a21 2x3 x2
b1 r12 r32 x12 y12 z12 x32 y 32 z 32
a12 2 y3 y1 a22 2 y3 y 2
b2 r22 r32 x 22 y 22 z 22 x32 y 32 z 32
a13 2z3 z1 a23 2z3 z 2
b1 a11 a
z x 12 y (A.4)
a13 a13 a13
b2 a 21 a
z x 22 y (A.5)
a 23 a 23 a 23
42
x f y a4 y a5 (A.6)
where:
a2 a3 a11 a 21 a12 a 22 b2 b
a4 a5 a1 a2 a3 1
a1 a1 a13 a 23 a13 a 23 a 23 a13
z f y a6 y a7 (A.7)
where:
a 21 a 4 a 22 b2 a 21 a5
a6 a7
a 23 a 23
Now substitute (A.6) and (A.7) into the first equation in (A.1) to eliminate x and z and obtain a
single quadratic in y only:
ay 2 by c 0 (A.8)
where:
a a42 1 a62
b 2a 4 a5 x1 2 y1 2a6 a7 z1
c a5 a5 2 x1 a7 a7 2 z1 x12 y12 z12 r12
To complete the intersection of three spheres solution, substitute both y values y+ and y- from
(A.9) into (A.6) and (A.7):
x a4 y a5 (A.10)
z a6 y a7 (A.11)
In general there are two solutions, one corresponding to the positive and the second to the
negative in (A.9). Obviously, the + and – solutions cannot be switched:
Example
Let us now present an example to demonstrate the solutions in the intersection of three spheres
algorithm. Given three spheres (c,r):
0 0 0 , 2
T
3 0 0 , 5
T
1 3 1 , 3
T
(A.13)
The intersection of three spheres algorithm yields the following two valid solutions:
These two solutions may be verified by a 3D sketch. This completes the intersection of three
spheres algorithm. In the next subsections we present several important topics related to this three-
spheres intersection algorithm: imaginary solutions, singularities, and multiple solutions.
Imaginary Solutions
The three spheres intersection algorithm can yield imaginary solutions. This occurs when the
radicand b 2 4ac in (A.9) is less than zero; this yields imaginary solutions for y , which physically
means not all three spheres intersect. If this occurs in the hardware, there is either a joint angle sensing
error or a modeling error, since the hardware should assemble properly.
A special case occurs when the radicand b 2 4ac in (A.9) is equal to zero. In this case, both
solutions have degenerated to a single solution, i.e. two spheres meet tangentially in a single point, and
the third sphere also passes through this point.
Singularities
The three spheres intersection algorithm and hence the overall forward pose kinematics solution
is subject to singularities. These are all algorithmic singularities, i.e. there is division by zero in the
mathematics, but no problem exists in the hardware (no loss or gain in degrees of freedom). This
subsection derives and analyzes the algorithmic singularities for the three spheres intersection algorithm
presented above. Different possible three spheres intersection algorithms exist, by combining different
equations starting with (A.1) and eliminating and solving for different variables first. Each has a
different set of algorithmic singularities. We only analyze the algorithm presented above.
Inspecting the algorithm, represented in equations (A.1) – (A.12), we see there are four
singularity conditions, all involving division by zero.
Singularity Conditions
a13 0
a23 0
(A.15)
a1 0
a0
44
a13 2 z3 z1 0 (A.16)
a23 2 z3 z2 0 (A.17)
are satisfied when the centers of spheres 1 and 3 or spheres 2 and 3 have the same z coordinate, i.e.
z1 z3 or z2 z3 . Therefore, in the nominal case where all four virtual sphere centers have the same z
height, this three-spheres intersection algorithm is always singular. An alternate solution is presented in
Appendix B to overcome this problem.
Simplifies to:
x3 x1 x3 x 2
(A.19)
z 3 z1 z 3 z 2
For this condition to be satisfied, the centers of spheres 1, 2, and 3 must be collinear in the XZ
plane. In general, singularity condition 3 lies along the edge of the useful workspace and thus presents
no problem in hardware implementation if the system is properly designed regarding workspace
limitations.
Is satisfied when:
a42 a62 1 (A.21)
It is impossible to satisfy this condition as long as a4 and a6 from (A.6) and (A.7) are real
numbers, as is the case in hardware implementations. Thus, the fourth singularity condition is never a
problem.
Multiple Solutions
In general the three spheres intersection algorithm yields two distinct, correct solutions ( in
(A.9 – A.11)). Generally only one of these is the correct valid solution, determined by the admissible
Delta Robot assembly configurations.
45
( x x1 )2 ( y y1 )2 ( z zn )2 r12 (B.1)
( x x2 ) ( y
2
y2 ) ( z zn ) r22
2 2
(B.2)
( x x3 )2 ( y y3 )2 ( z zn )2 r32 (B.3)
Since all Z sphere-center heights are the same, we have z1 z2 z3 zn . The unknown three-
spheres intersection point is P x y zT . Expanding (1-3) yields:
All non-linear terms of the unknowns x, y cancelled out in the subtractions above. Also, all z-
related terms cancelled out in the above subtractions since all sphere-center z heights are identical.
Equations (7-8) are two linear equations in the two unknowns x, y, of the following form.
a b x c
(B.9)
d e y f
where:
a 2( x3 x1 )
b 2( y3 y1 )
c r12 r32 x12 y12 x32 y32
d 2( x3 x2 )
e 2( y3 y2 )
f r22 r32 x22 y22 x32 y32
ce bf
x
ae bd
(B.10)
af cd
y
ae bd
Az 2 Bz C 0 (B.11)
where:
A 1
B 2 zn
C zn2 r12 ( x x1 ) 2 ( y y1 )2
Knowing the unique values x, y, the two possible solutions for the unknown z are found from the
quadratic formula:
B B 2 4C
z p,m (B.12)
2
For the Delta Robot, ALWAYS choose the z height solution that is below the base triangle, i.e.
negative z, since that is the only physically-admissible solution.
This simplified three-spheres intersection algorithm solution for x, y, z fails in two cases:
i) When the determinant of the coefficient matrix in the x, y, linear solution (B.10) is zero.
This is an algorithmic singularity whose condition can be simplified as follows. (B.13) becomes:
( x3 x1 )( y3 y2 ) ( y3 y1 )( x3 x2 ) (B.14)
If (B.14) is satisfied there will be an algorithmic singularity. Note that the algorithmic
singularity condition (B.14) is only a function of constant terms. Therefore, this singularity can be
avoided by design, i.e. proper placement of the robot base locations in the XY plane. For a symmetric
Delta Robot, this particular algorithmic singularity is avoided by design.
ii) When the radicand in (B.12) is negative, the solution for z will be imaginary. The condition
B 2 4C 0 yields:
( x x1 )2 ( y y1 )2 r12 (B.15)
When this inequality is satisfied, the solution for z will be imaginary, which means that the robot
will not assemble for that configuration. Note that (B.15) is an inequality for a circle. This singularity
will NEVER occur if valid inputs are given for the FPK problem, i.e. the Delta Robot assembles.