A Complete Solution to the Inverse Kinematics

Problem for 4-DOF Manipulator Robot

Yang Si, Qingxuan Jia, Gang Chen, Hanxu Sun
School of Automation
Beijing University of Posts and Telecommunications, BUPT
Beijing, China
[email protected]

Abstract—A complete closed-form solution to the inverse problem for redundant and non-redundant manipulators. Sung-
kinematics problem for a 4-DOF manipulator robot is proposed. Woo Kim [3] derived the inverse kinematics solution using
With the discussion of the existence of the offset and the distance the fuzzy logic. A. Rodriguez-Angeles [4] presents a novel on-
between the axis of 1st joint and 2nd joint, all the possible solutions line optimal control for tracking tasks on redundant robot
and singular configurations are presented. “Task attitude” is defined manipulators for which inverse kinematics is not required.
to describe the orientation of the end effector, which is convenient to Luis G. Herrera-Bendezu [5] developed a software package to
solve the inverse kinematics problems in engineering. Finally, solve the direct and inverse kinematics of an n degree-of-
simulation of path-planning is performed, and the efficiency of the freedom manipulator symbolically. It can either be applied to
proposed method is verified through Mathematica.
industrial robots, or to explore more complicated
Keywords—4-DOF manipulator;inverse kinematics;task attitude;
configurations.
path-planning Closed-form solution to inverse kinematics is proposed
main for non-redundant manipulators. I-Ming Chen [6]
I. Introduction developed a closed-form inverse kinematics solver for non-
redundant reconfiguration robots based on the Product-of
The desired trajectories and task point of robot
Exponentials formula. This solver can cope with all robots
manipulators are usually given in the Cartesian space, but
with DOFs of 4 or less. Jose Gabriel and John Q. Gan
controllers are designed in the joint space. So it is important to
proposed a complete closed-form solution to the inverse
establish the mapping relationships between the two spaces.
kinematics problem for a 5-DOF manipulator robot[7-8]. But
One of the main problems in robotics field is the inverse
this method did not discuss the possible influence by the offset
kinematics problem, which contributes to the conversion from
of the manipulator structure.
Cartesian space to the joint space. The inverse kinematics
problem can be described as follows: compute the joint angles In this paper, A complete closed-form solution to the
related to a given position and orientation of the end-effector. inverse kinematics problem for the 4-DOF manipulator robot
is proposed. With the discussion of the existence of the offset
According to the mechanism, a definite position and
and the distance between the axis of 1st joint and 2nd joint, all
orientation of the end-effector for a certain serial manipulator
the possible solutions and singular configurations are
can result in variable configurations, even in infinite
presented.
configurations. And the equations to solve the inverse
kinematics problem are generally nonlinear. Thus it is not This paper is organized as follows: Section Ċ establishes
always possible to find a closed-form solution. the kinematic model. The complete closed-form inverse
The different solutions can be classified into two kinematic solution for the 4-DOF manipulator is proposed in
approaches: closed-form analytical solution and numerical Section ċ. Section Č presents the application of the proposed
solution. Though closed-form solution mainly relies on the solution. Section č takes out simulation to verify the method.
kinematical structure of the manipulator, it is faster than the Conclusions are given in Section 9,.
numerical solution and all the possible solutions of inverse
kinematics can be given out. Numerical solution is a more II. Kinematic Model
general iterative method, but it is slower and can only find one
solution for one set of initial values. Furthermore, when the The 4-DOF manipulator discussed in this paper has four
numerical method fails to converge, the solution of the inverse rotary joints: the first one stands vertically, while the other
kinematics problem cannot be determined even if it exists. three are horizontal and parallel. Just as Fig.1 shows,
O0 − x0 y0 z0 is the global coordinate system, the centers
Among the numerical solutions to the inverse kinematics,
Farbod Fahimi [1] presented a kinematic position and velocity O1 − O4 are placed on the rotation axis of the joints, and OT is
solution scheme for spatial hyper-redundant manipulator by the origin of the tool coordinate. θi represents the variable of
introducing backbone curve concepts and new mode shapes. each joint, which equals to zero when the manipulator is in the
A.Ramdane-Cherif [2] proposed an iterative method using shown configuration. ai is the length of each link, and all the
neural optimization network to solve the inverse kinematic
This project is supported by the National Natural Science Foundation of
China (61175080), the National Key Basic Research Program of China
(2013CB733005) and Specialized Research Fund for the Doctoral Program
of Higher Education (20120005120004).

978-1-4673-6322-8/13/$31.00 2013
c IEEE 1880
links are parallel. The Denavit-Hartenberg parameters are
listed in Table 1.
ω x = − DS1 + AC1 (4)

ω y = DC1 + AS1 (5)

ω z = d1 − a2 S2 − a3 S23 (6)

where D = d 2 + d3 ᧨ A = a1 + a2C2 + a3C2+3 .

Fig. 1. the initial configuration of 4-dof manipulator Parameter D stands for the offset among the parallel
links. Without the offset, all the axes of the links are coplanar,
TABLE I DH PARAMETERS OF THE 4-DOF MANIPULATOR and their projection on the x0Oy0 plane of the coordinate
LINK θi (°) d i ( MM) ai −1 ( MM) α i −1 (°) system O0 − x0 y0 z0 appears to be a line(Fig.2(a)). However,
the existence of the offset will result in a displacement of the
1 θ1 (0) d1 0 0 point O4 ' (the projection of O4 on the x0Oy0 plane), as
2 θ 2 (0) d2 a1 -90 Fig.2(b) shows, which is no longer located on the x0 axis
3 θ3 (0) d3 a2 0 when the manipulator is in the initial configuration. The offset
4 θ 4 (0) 0 a3 0
makes a difference in the solution to the inverse kinematic
problem.

With the Denavit-Hartenberg representation, we obtain the

O4′
transformation matrices as follows:
O4′
0
1T = Rot ( z1 , θ1 ) ⋅ Trans ( z1 , d1 )
1
2T = Rot ( x1 , α1 ) ⋅ Trans (x1 , a1 ) ⋅ Rot (z 2 , θ 2 ) ⋅ Trans (z 2 , d 2 ) (a) without offset (b) with offset
2
3T = Trans ( x 2 , a2 ) ⋅ Rot (z 3 , θ 3 ) ⋅ Trans ( z 3 , d 3 ) (1) Fig. 2. Variation of the optimal solution and average population
3
4T = Trans (x 3 , a3 ) ⋅ Rot ( z 4 , θ 4 ) ⋅ Trans (z 4 , d 4 )
4
TT = Trans ( x 4 , a4 ) ⋅ Rot (z 4 , −90D ) ⋅ Rot (x 4 , −90D ) A. the solution of θ1
So, the transformation matrices from the global coordinate The solution of θ1 can be calculated in two different
to the tool coordinate is given by:
situations:
§ 0R 0
P· 1). If D = 0 , ωx and ω y can be described as follows:
T = 01T ⋅ 12T ⋅ 23T ⋅ 34T ⋅ T4T = ¨ T
0
T
T
¸ (2)
© 0 1 ¹
ω x = A ⋅ C1 (7)
The matrix and the vector describe the forward kinematics
model, given by the following equations: ω y = A ⋅ S1 (8)

§ C1C234 −C1 S234 − S1 · a. If A = 0 , then ω x = 0 and ω y = 0 , the point O4 is

¨ ¸
T
0
R = ¨ C234 S1 − S1 S234 C1 ¸ exactly located on the Z 0 axis. The manipulator loses the
¨ −S −C234 0 ¸¹ velocity in normal direction of the plane, which contains all
© 234
(3) the links, and the manipulator is in “waist-singular”
ªC1 (a1 + a2 C2 + a3C23 ) − (d 2 + d3 ) S1 º configuration (Fig. 3). In this case, any value of θ1 is a
0
P = « (d + d )C + (a + a C + a C ) S »
T « 2 3 1 1 2 2 3 23 1» solution and there are infinite solutions for θ1 .
«¬ d1 − a2 S 2 − a3 S 23 »¼
b. If A ≠ 0 , whether a1 (the distance between the axis of
Where Ci = cos θ i , Si = sin θ i , Cij = cos(θi + θ j ) and 1 joint and 2nd joint) is existent should be taken into account.
st

Sij = sin(θ i + θ j ) . If a1 = 0 , the manipulator always has a symmetrical

configuration about the x1Oy1 plane. Thus two different
III. Inverse Kinematic Model solutions for θ1 can be computed, just as the (8) and (9); But
The point O4 , which is the origin of the coordinate system if a1 ≠ 0 , two circles on the x0Oy0 plane can be obtained:
O4 − x4 y4 z 4 , is located on the wrist, as shown in Fig.1. The one’s radius is a2 + a3 − a1 ,the other’s is a2 + a3 + a1 . When
position of O4 is obtained by the following equations: the O4 ' locates inside the small circle, two solutions for θ1

2013 IEEE 8th Conference on Industrial Electronics and Applications (ICIEA) 1881
can be obtained, as the (8) and (9); otherwise, only one B. the solution of θ 2 and θ 3
solution for θ1 is computed, as (10) shows.
For any value of θ1 , Z 0 Rω is the vertical plane in which
When a1 = 0 or ω + ω ≤ (a2 + a3 − a1 ) ᧨ two solutions
2
x
2
y
2
z0 and the a1 lie. Projecting all the links to the plane Z 0 Rω ,
for θ1 ᧶ l (the length of the projection of O2O4 ), and O4′′ (ωr , ω z ) ( the
projection of O4 ) are obtained.
­°θ1a = a tan 2(ω x , ω y )
® (9)
°̄θ1b = a tan 2(ωx , ω y ) + π z0
a1
O2"

Δβ θ2
β
Z0
d1 O3"
Y0
θ3
0
Rω

X0 ωz
ωr
O4"

Fig. 3. “waist-singular” configuration Fig. 4. the projection on the Z 0 Rω plane

When (a2 + a3 − a1 )2 < ωx2 + ω y2 < (a2 + a3 + a1 ) 2 and a1 ≠ 0 The offset is taken into consideration, and the expressions
can be obtained as follows:
᧨only one solution for θ1 ᧶

θ1 = a tan 2(ω x , ω y ) (10) ωr = ω x2 + ω y2 − D 2 (13)

2) If D ≠ 0 , l = ωr2 + (d1 − ω z ) 2 (14)

When (a2 + a3 − a1 ) 2 + D 2 < ωx2 + ω y2 < (a2 + a3 + a1 ) 2 + D 2 ,
In the triangle ΔO2′′O3′′O4′′ ,
only one solution for θ1 :
a22 + a32 − 2a2 a3 cos(π − θ3 ) = lO2 ′′O ′′ (15)
θ1 = a tan 2(ω x , ω y ) −+θ1i (11) 2 4

So there are two solutions for θ3 :

Where +θ1i is the angle between OO4′ and x0 , when θ1
is at the initial configuration. If A = 0 , +θ1i = 90D , and if ­ ω x2 + ω y2 − D 2 + (d1 − ωz ) 2 − a22 − a32
A ≠ 0 , +θ1i = a tan 2(ω x , D ) . °θ 3a = a cos( )
° 2a2 a3
® (16)
When ωx2 + ω y2 ≤ (a2 + a3 − a1 ) 2 + D 2 , two solution for θ1 : ° ω x2 + ω y2 − D 2 + (d1 − ω z ) 2 − a22 − a32
°θ 3b = −a cos( 2a2 a3
)
¯
­°θ1a = a tan 2(ωx , ω y ) −+θ1i
® (12) When θ3 a = θ 3b = 0D , l = a3 − a2 ; when θ3 a = θ 3b = 180D ,
°̄θ1b = a tan 2(ωx , ω y ) −+θ1i ++θ1t
l = a3 + a2 . In the two conditions, θ 3 has only one solution.
Where +θ1t is the deviation of the two solutions of the 1st
The point O4′′ is exactly located on the circle centered on O2′′
joint, which relate to the same position of O4′ . and radius l . The manipulator loses the velocity along the

( )
direction of the second link and singularity happens.
If a1 ≠ 0 ᧨ +θ1t = 90D − a sin D ωx2 + ω y2 ᧷
as shown in Fig. 4, β is the angle between O2′′O4′′ and
else if a1 = 0 ᧨ +θ1t = 2 a cos ( D ω 2
x +ω ) .
2
y
Rω , and its value is obtained as follows:

β = a tan 2(ωr − a1 , d1 − ω z ) (17)

In the triangle ΔO2′′O3′′O4′′ ,

1882 2013 IEEE 8th Conference on Industrial Electronics and Applications (ICIEA)
 a22 + lO22 O4 − 2a2 lO2O4 cos Δβ = a32 (18) Begain
k =1

From the Fig. 4, θ 2 = β ±+β Given Pk ,α k

Compute θ1, k , a ,θ1,k ,b ,

That is: θ 2,k , a , θ 2, k ,b , θ 3, k , a ,θ 3, k ,b
The former joints
θ1, k −1 , θ 2, k −1 , θ 3, k −1 , θ 4, k −1
­θ = a tan 2( ω x2 + ω y2 − D 2 − a1 , d1 − ω z ) θ1,k,a −θ1,k −1 > θ13 k , a − θ3, k −1 >
° 2a Yes θ1,k,b −θ1,k −1 ? No Yes θ3, k ,b − θ3,k −1 ? No
° a22 − a32 + ωr2 + (d1 − ω z ) 2 θ1,k = θ1,k ,b θ1,k = θ1,k ,a θ3,k = θ3,k,b θ3,k = θ3,k,a
° + a sin
°° 2a2 ωr2 + (d1 − ω z ) 2
® (19)
θ3, k > 0 ?
°θ 2 b = a tan 2( ωx2 + ω y2 − D 2 − a1 , d1 − ω z ) Yes No
° θ 2, k = θ 2, k ,b θ2,k = θ2,k ,a
° a − a + ω + ( d1 − ω z )
2 2 2 2

° − a sin 2 3 r

°̄ 2a2 ωr2 + ( d1 − ω z ) 2 θ 4 ,k = α k
−θ 2, k − θ 3, k

According to the relationship between the two links: The solution

{θ 1, k , θ 2 , k , θ 3, k , θ 4 , k }
If θ3 > 0 , θ 2 = θ 2b ;
k = k +1 No
k = N?
If θ3 < 0 , θ 2 = θ 2 a ; Yes
End

If θ3 = 0 or θ3 = 180D , + β = 0 ᧨
Fig. 5. the procedure of solution chosen

θ 2 = β = a tan 2( ω x2 + ω y2 − D 2 − a1 , d1 − ωz ) . Where k = 1,", N , θ1, k , a and θ1, k ,b are the two values of
the first joint of step k .
C. the solution of θ 4
Only one group of solution for each given point is left after
“Task attitude”, which is defined as the angle between zT the solution chosen procedure, which is indispensable to
and plane x0 Oy0 , is assumed as α ( −90D < α < 90D ) . θ 4 can be trajectory planning with application of the inverse kinematics
approach proposed in the paper.
described as:
θ 4 = α − θ 2 − θ3 (20) V. Simulation Verification
As a result, a complete closed-form solution to the inverse In order to verify the correctness of the method proposed
kinematics problem of a 4-DOF manipulator robot can be in the paper, simulation of line path planning is designed:
calculated when the position and attitude is definite. For the the initial joint values: {−62D , −15D ,73D , −87D }
given 4-DOF manipulator, the number of solutions for the
joint variable θ1 can be one or two, even infinite when the initial position: {−599.7mm, −3413.9mm, −842.2mm}
singularity happens; the number of solutions for the pair the initial task attitude: α ini = −29D
(θ 2 ,θ 3 ) can be one or two related to each θ1 ; and the unique the final position: {−499.7mm, −3213.9mm, −1042.2mm}
θ 4 can be determined by given (θ1 ,θ 2 ,θ 3 ) . In conclusion, the final task attitude: α end = −29D
when the singularity does not happen, the inverse kinematics the total time: t f = 20 s
problem has four different solutions at most.
the accelerating time: ta = 5s .
IV. Application in Path-Planning The result of simulation is as Fig. 6 and Fig. 7.
When the 4-DOF manipulator is required to trace a certain It can be seen in the Fig. 6 and 7 that the curves are
trajectory, which is often dispersed into a series of points, it is continuous and smooth without any jerk. So the efficiency of
highly applicable for the method on solving the inverse the solution is verified.
kinematics problem proposed in this paper. Since all the
possible values of each joint can be computed from the
VI. Conclusion
complete solution to the inverse kinematics problem, the most
suitable group of joint values should be chosen. So that the In this paper, the influence of the existence of the offset D
joint displacements are small enough to prevent the 4-DOF and a1 to the closed-form solution is discussed. When the
manipulator robot from violating the velocity limits. The offset D is not existence, the 4-DOF manipulator robot may be
solution chosen procedure is as Fig.5 shows: in “waist-singular” configuration. But the possibility of waist
singularity never happens only if the offset D does exist.
What’s more, the deviation between the two solutions of the

2013 IEEE 8th Conference on Industrial Electronics and Applications (ICIEA) 1883
 T1+°/ T2+°/ T3+°/ T4+°/

 62 15
100  90
 63
20
 64 90  100
25
 65
80  110
30
 66

t+s / t+s/ t+s / t+s /

5 10 15 20 5 10 15 20 5 10 15 20 5 10 15 20

Fig. 6. Joint angle in the path-planning

v1+°ss / v2 +°ss / v3 +° / v4 +°/

t+s / t+s / t+s /

5 10 15 20 5 10 15 20 0.20 5 10 15 20
 0.01 0.05
0.05
0.15
0.10
 0.02 0.10
0.15
0.10 0.05
 0.03 0.20
t+s/
 0.04 5 10 15 20 0.25
0.15

Fig. 7. Joint velocity in the path-planning

1st joint is different, according to whether the offset D is

existent or not. In addition, the number of possible solutions to REFERENCES
the first joint is influenced by the existence of a1 . [1] Kinematic and velocity solution for spatial hyper-redundant robots,”
IEEE Trans. Robotics and Automation, vol. 18, NO.1, Feb. 2002.
The concept of “task attitude” is defined as the angle [2] Ramdane-Cherif, A., Meddah, D.Y., Perdereau, V., Drouin, M.,
between zT in tool coordinate of the 4-DOF manipulator and "Inverse kinematic solution based on Lyapunov function for redundant
and non-redundant robots," IEEE Trans, Computational Intelligence in
the plane x0 Oy0 in global coordinate, which describes the Robotics and Automation, pp.226-231, Jul 10-11 1997.
desired orientation of the end-effector. The efficiency of the [3] Sung-Woo Kim and Ju-Jang Lee, “Inverse Kinematics solution based on
proposed method is verified by the simulation of line path- fuzzy logic for redundant manipulators,” IEEE/RSJ Trans.,Intelligent
Robots and Systems, July 26-30, 1993.
planning.
[4] 5RGULJXH]$QJHOHV $ &UX]9LOODU &$ 0XUR0DOGRQDGR
The procedure to solve the inverse kinematics problem ' 2QOLQH IUHH LQYHUVH NLQHPDWLFV MRLQWFDUWHVLDQ
proposed in this paper is dependent of the kinematical RSWLPDO FRQWURO RI UHGXQGDQW PDQLSXODWRUV IEEE Trans
Mechatronics and Automation, pp.672-677, Aug 5-8 2008.
structure of the manipulator, thus the procedure should be
[5] +HUUHUD%HQGH]X /*  0X ( &DLQ -7  6\PEROLF
derived as long as the structure of the manipulator changes. So FRPSXWDWLRQ RI URERW PDQLSXODWRU NLQHPDWLFVIEEE Trans,
it may be inconvenient in engineering application and needs to Robotics and Automation, YROSS$SU.
be improved. [6] I-Ming Chen, Yan Gao, “Closed-Form Inverse Kinematics Solver for
Reconfigurable Robots,” IEEE Trans. Robotics and Auto mation, May
21-26, 2001.
Acknowledgment
[7] Ramírez-Torres, J.G.; Toscano-Pulido, G.; Ramírez-Saldívar, A.;
This project is supported by the National Natural Science Hernández-Ramírez, A., “A Complete Closed-Form Solution to the
Foundation of China (61175080), the National Key Basic Inverse Kinematics Problem for the P2Arm Manipulator Robot,” IEEE
Research Program of China (2013CB733005) and Specialized Trans. Robotics and Automotive, pp.372-377, Sept. 28-Oct. 1, 2010.
Research Fund for the Doctoral Program of Higher Education [8] John Q. Gan*, Eimei Oyama, Eric M. Rosales and Huosheng Hu,”A
(20120005120004). complete analytical solution to the inverse kinematics of the Pioneer 2
robotic arm,” Robotica, vol. 23, pp. 123-129, 2005

1884 2013 IEEE 8th Conference on Industrial Electronics and Applications (ICIEA)