Professional Documents
Culture Documents
Abstract
In this paper, a particular emphasis is put on the workspace mapping with deficient-DOF space between the PUMA 560 robot and its
exoskeleton-arm master-type manipulator, which is proved to be the key step for different structure-based master–slave manipulation.
The deficient-DOF space of PUMA 560, made up with the singularity and joint limitation, affects the maneuverability and stability of the
master–slave manipulation system. In this work, the deficient-DOF space in the mapped workspace with master–slave control is
investigated as a main factor of the workspace mapping. Meanwhile, the orthogonal experiment design method is introduced and two
rounds of orthogonal experiments are carried out for this mapping problem, which is simultaneously characterized by many other
variables At last the simulation and experiment results demonstrate that the scheme of the mapping is feasible and the orthogonal
experiment design method is effective. It is a novel application and exploration of the orthogonal experiment design method in the
mechanical or robot optimal design.
r 2006 Elsevier Ltd. All rights reserved.
Keywords: Exoskeleton arm; PUMA 560; Master–slave manipulator; Workspace mapping with deficient-DOF-space; Orthogonal experiment design
method
0736-5845/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.rcim.2006.05.007
ARTICLE IN PRESS
Y. Chen et al. / Robotics and Computer-Integrated Manufacturing 23 (2007) 478–487 479
2 3
where R3 3 are the 3 3 matrices of the end effector pose, C 1 S23 C 1 C 23 S4 S1 C 4 C 1 C 4 C 23 S 5 S 1 S4 S 5 þ C 1 S 23 C 5
6 S1 C 4 C 23 S5 þ C 1 S4 S5 þ S 1 S23 C 5 7
[Px Py Pz]T is the position of the end effector and n1An J 22 ¼ 4 S 1 S23 S1 C 23 S 4 þ C 1 C 4 5.
are the D–H homogeneous transformation matrices. C 23 S 23 S 4 C 23 C 5 S23 C 4 S 5
Fig. 5 shows the original workspace mapping between (6)
ZJUESA and PUMA 560. The red zone represents the
Here, Si, Ci, Sij and Cij represent sin(yi), cos(yi), sin(yi+yj)
workspace of ZJUESA and the blue volume that of
and cos(yi+yj), respectively, and a2, a3 and d3 can be found
PUMA. There are large differences in both shape and size.
in the D–H parameters of the PUMA 560. Thus, the
The workspace of the ZJUESA is much smaller than that
singularities can be identified by checking the determinants
of PUMA and it mostly locates at the central margin of the
of the two 3 3 matrices J11 and J22 as follows:
PUMA workspace casually. It is not an ideal condition.
But for further research, it is found that even with a good detðJ 11 Þ ¼ a2 ðd 4 C 3 a3 S 3 Þðd 4 S23 þ a2 C 2 þ a3 C 23 Þ. (7)
workspace mapping with point-to-point mapped method There are two conditions for forearm singularities. One is
(Eq. (1)), yet the deficient-DOF points, the black points, in the boundary singularity with
the mapped workspace will cause trouble in the manipula-
tion. This symptom cannot be diagnosed by the common gb ¼ d 4 C 3 a3 S 3 ¼ 0, (8)
workspace mapping only and the deficient-DOF space and the other is the interior singularity with
should be considered in the mapping process.
As defined before, the deficient-DOF space includes gi ¼ d 4 S23 þ a2 C 2 þ a3 C 23 ¼ 0. (9)
singularities and joint limitations. For PUMA 560, by Besides, the wrist singularity happens when
defining the control point to be at the wrist, the Jacobian
detðJ 22 Þ ¼ gw ffi S5 ¼ 0. (10)
matrix is a square 6 6 matrix, which can be partitioned into
" # For simplicity, we just discuss the workspace of the
J 11 033 master–slave manipulator by omitting the orientation of the
J¼ , (3) end effector. So in our work the wrist singularity is taken out
J 21 J 22
of account.
where J11, J21 and J22 are defined in Eqs. (4)–(6). In addition, the joint limitation points of the PUMA 560
also restrict the movement, which resource from the joints
2 3
d 4 S 1 S 23 a2 S 1 C 2 a3 S 1 C 23 d 2 C 1 d 4 C 1 C 23 a2 C 1 S 2 a3 C 1 S 23 ðd 4 C 23 a3 S23 ÞC 1
6 ðd 4 C 23 a3 S23 ÞS 1 7
J 11 ¼ 4 d 4 C 1 S23 þ a2 C 1 C 2 þ a3 C 1 C 23 d 2 S 1 d 4 S1 C 23 a2 S 1 S 2 a3 S 1 S 23 5, (4)
0 d 4 S 23 a2 C 2 a3 C 23 d 4 S23 a3 C 23
Fig. 5. The original workspace mapping between ZJUESA and PUMA 560 with its deficient-DOF space.
ARTICLE IN PRESS
482 Y. Chen et al. / Robotics and Computer-Integrated Manufacturing 23 (2007) 478–487
factor, the levels of factors are evenly set over the full Table 2
applicable range, e.g. factor A with AA[0, 360], and can be The factors and their levels of the orthogonal design
set arbitrarily. And the levels of each factor, equivalent to A (1) B C D E (mm) F (mm)
the step length for each variable in other optimal method,
could be chosen roughly in First round orthogonal 1 0 1 1 1 0 0
experiment design. The levels of all factors studied are 2 72 1.5 1.5 1.5 75 25
3 144 2 2 2 150 50
summarized in Table 2. The experimental plan is designed
4 216 2.5 2.5 2.5 225 75
according to an L25(56)orthogonal array. 5 288 3 3 3 300 100
With different combination of the factors levels, the
workspace of the ZJUESA and PUMA manipulator can be
calculated through the kinematical expression, Eq. (2) and
mapping Eq. (1). The volume of these two workspaces can Table 3
be regarded as the composition of many unit volumes Results Q and its analysis (%)
(2 2 2 mm). By comparing the number of these unit A B C D E F
volumes in these two workspace, Em, Pm and SL can
be determined. With Eqs. (11)–(13), we get the result. In Ij1 10.48 3.83 4.85 5.64 5.87 6.80
Table 3, the values resulting from the analysis in Ij2 6.99 5.32 5.40 6.57 7.31 6.86
Ij3 6.22 8.83 8.18 6.37 8.27 7.43
accordance with the orthogonal array are given, where
Ij4 5.57 9.40 9.65 8.65 7.89 7.91
2 3 2 3 Ij5 5.70 7.58 6.87 7.73 5.62 5.96
I A1 1 1 1 1 1 0 0 0 0 0 0
6 7 6 7 Kj 4.89 5.57 4.80 3.01 2.65 1.95
6 I A2 7 60 0 0 0 0 1 1 0 0 0 07
6 7 6 7
6 7 6 7
6 I A3 7 60 0 0 0 0 0 0 0 0 0 07
6 7 6 7
6 7 6 7
6 I A4 7 60 0 0 0 0 0 0 0 0 0 07
6 7 6 7
6 7
6 I A5 7 1660 0 0 0 0 0 0 1 1 1 17
7
6 7¼ 6 7
6 7 56 . . . . . . . 7
6 .. 7 6. . . . . . . .. .. .. .. 7
6 . 7 6 . . . . . . . . . . . 7
6 7 6 7
6 7 6 7
6 I F1 7 61 0 0 0 0 0 1 0 0 0 17
6 7 6 7
6 7 6 7
6 I F2 7 60 1 0 0 0 0 0 1 0 0 07
4 5 4 5
I F3 0 0 1 0 0 0 0 0 1 0 0
2 3
Q1
6 7
6 Q2 7
6 7
6 7
6 Q3 7
6 7
6 7
6 Q4 7
6 7
6 7
6 5 7
6 Q
7, ð12Þ
6 7
6 7
6 7
6 7
6 Q22 7 Fig. 9. The relations between the levels of the factors and Q.
6 7
6 7
6 Q23 7
4 5
Q25 factor level is when the Q gets the largest value, namely A1,
B4, C4, D4, E3 and F4.
It is interesting to know how good the results derived
K j ¼ maxfI j1 ; I j2 ; . . . ; I j5 g minfI j1 ; I j2 ; . . . ; I j5 g, (13)
from the above 25 trials are, when compared to all other
for j ¼ A, B, C, y, E, F. possible combinations. Because of its mutual balance of
According to Table 3, the relations between the levels of orthogonal arrays, this performance ratio can be guaran-
each factors and Q can be obtained intuitively, as shown in teed by the following theorem in nonparametric statistics
Fig. 9. From this graph, we can easily determine that the [20].
superiority and the degree of the influence (sensitivity) of
each design factor. In this example, it can be concluded Theorem 1. Suppose random variable X is subject to a
that the sensitivity of the factors A–C are high and factors probabilistically continuous distribution F(X), and x1, x2,
E and F have weak influence, since KA, KB and KC are y, xn are simple samples (or random observation values) of
much bigger than KE and KF. The best combination of each X. If x1, x2, y, xn are sorted in ascending order, denoted as
ARTICLE IN PRESS
Y. Chen et al. / Robotics and Computer-Integrated Manufacturing 23 (2007) 478–487 485
x1ox2o?o xn, then the performance ratio for xi Formula (15) means that the best experimental result in
(i ¼ 1, 2, y, n) is these simple samples is probabilistically better than
(n/n+1)% of all possible results defined in the whole discrete
i
E½F ðxi Þ ¼ . (14) solution space. In this case, the best result by L25(56) is better
nþ1 than 96.15% (¼ 25/26) results of all 15625 trials.
In particular, Then, for more accurate mapping, we carry out the
second round orthogonal experiment design based on the
n
E½F ðxi Þ ¼ . (15) result of the first round. There are also 6 factors with 5
nþ1 levels, which are allocated near the first round result as A1,
B4, C4, D4, E3 and F4. Table 4 shows the factors and levels
Table 4 of the second round orthogonal design. In this round, the
Factors and levels of the second round orthogonal experiment design difference between the levels of each factor is much smaller
than that in the previous round. This brings small step
A (1) B C D E (mm) F (mm) length of each variable in the calculation. Table 5 shows the
1 0 2.3 2.3 2.3 100 60 results of the second round orthogonal experiment design.
2 10 2.4 2.4 2.4 125 65 In each column, we choose the level corresponding to the
3 20 2.5 2.5 2.5 150 70 maximum value of Q, namely A4, B5, C1, D3, E2 and F1.
4 30 2.6 2.6 2.6 175 75 These levels are the best mapping parameters and listed in
5 40 2.7 2.7 2.7 200 80
Table 6.
By the kinematics calculation, we can get the final
effect of the workspace mapping. Fig. 10 shows the result
Table 5 of the workspace mapping with deficient-DOF space.
Results of the second round orthogonal experiment design (%) Fig. 11 gives the comparison between the workspace of
A B C D E F the ZJUESA and the deficient-DOF space of the
PUMA 560. The figures show that the workspace of the
Ij1 17.75 16.71 20.32 18.22 17.52 19.22 ZJUESA and the PUMA 560 matches each other
Ij2 17.93 17.69 20.19 18.82 18.63 17.69
much better than that in Fig. 5. Only 0.2% points in the
Ij3 17.81 17.83 18.22 19.13 18.48 18.36
Ij4 18.62 19.15 16.33 17.76 18.35 17.90 mapping workspace are deficient-DOF points of PUMA
Ij5 18.55 19.27 15.60 16.73 17.67 17.48 560 robot, which influence the master–slave manipulator
Kj 0.8 2.56 4.72 2.60 1.11 1.74 slightly. For these unavoidable deficient-DOF points, the
Jacobian transpose, pseudoinverse, or selective damped
least-squares method can be easily investigated to solve the
Table 6
problem.
The best parameters for the workspace mapping
6. Conclusion
A (1) B C D E (mm) F (mm)
30 2.7 2.3 2.5 125 60 In this paper our work is focused on the workspace
mapping with deficient-DOF space of the PUMA 560
Fig. 10. The result of the workspace mapping with deficient-DOF space of PUMA 560 and ZJUESA.
ARTICLE IN PRESS
486 Y. Chen et al. / Robotics and Computer-Integrated Manufacturing 23 (2007) 478–487
Fig. 11. The comparison between the workspace of ZJUESA and the deficient-DOF space of PUMA 560.
robot and its exoskeleton-arm manipulator, ZJUESA. It is tional conference on robotics and automation, May 2002, Washing-
the extension of research of our previous work on the ton, DC, p. 1269–74.
workspace mapping of the different structure and kine- [4] Wang H, Low KH, Gong F, Wang MY. A virtual circle method
for kinematic mapping from human hand to a nonanthropomorphic
matic master–slave manipulators. It is from the viewpoint robot hand. The Eighth International Conference on Control,
of the mapping optimization to minimize the negative Automation, Robotics and Vision, Kunming, China, December
influence on the master–slave manipulation due to the 2004.
singularities and joint limitation points of the PUMA 560 [5] Weston BG, Ryan PF, Michael LT, Mark RC. Calibration and
or any other slave-type manipulators. Mapping of a Human Hand for Dexterous Telemanipulation. Haptic
interfaces for virtual environments and teleoperator systems sympo-
In the mapping optimization, the main factor, the sium, ASME IMECE 2000 conference, 2000.
deficient-DOF space made up with the singularity and [6] Bouzit M. Design, Implementation and Testing of a Data Glove with
joint limitation, and any other factors are considered. For Force Feedback for Virtual and Real Objects Telemanipulation. PhD
such multi-factor optimal problem, the orthogonal experi- thesis, Laboratoire de Robotique de Paris, University of Pierre Et
ment design method is utilized for setting the mapping Marie Curie, France, 1996.
[7] Liu J, Zhang YR. Dataglove-based grasp planning for multi-fingered
parameters, and then two rounds of orthogonal design robot hand. In: Proceedings of the 11th world congress on
experiments are implemented. Because of its convenience, mechanism and machine science, April 2004, Tianjin, China, vol. 4,
resistance to noise and validity over the entire region, good p. 1827–31.
mapping result is obtained. This verifies the feasibility of [8] Fischer M, van der Smagt P, Hirzinger G. Learning techniques
the mapping scheme and the effectiveness of the orthogo- in a dataglove-based telemanipulation system for the DLR hand.
In: IEEE proceedings of the international conference on robotics and
nal experiment design method. However, it is a novel
automation. May 1998, Leuven, Belgium, vol. 2, pp. 1603–8.
application and exploration of this method in the mechan- [9] Kyriakopoulos KJ, et al. Kinematic analysis and position/force
ical or robot optimal design. control of the Anthrobot dexterous hand. IEEE Trans Syst Man
Cybernet B 1997;27(1):95–104.
[10] Rohling RN, Hollerbach JM. Optimized fingertip mapping for
Acknowledgment teleoperation of dexterous robot hands. In: IEEE proceedings
international conference on robotics and automation. May 1993,
The presented work is supported by the Natural Science Atlanta, GA, USA, vol. 3, pp. 769–75.
[11] Nakamura Y. Advanced Robotics—Redundancy and Optimization.
Foundation of China (no. 50305035) for which the authors Reading, MA: Addison-Wesley; 1991.
express thanks. [12] Chiaverini S, Egeland O. A solution to the singularity problem for
six-joint manipulators. In: Proceedings of the 1990 IEEE Interna-
tional conference on robot and automation, May 1990, Cincinnati,
References OH, p. 644–9.
[13] Nakamura Y, Hanafusa H. Inverse kinematic solution with
[1] Sheridan TB. Telerobotics. Automatica 1989;25(4):487–507. singularity robustness for robot manipulator control. ASME J Dyn
[2] Rajiv VD, Everett SE, Pernalete N, Manocha KA. Teleoperation Syst Meas Control 1986;108:163–71.
Assistance Through Variable Velocity Mapping. IEEE Trans [14] Wampler CW. Manipulator inverse kinematic solutions based on
Robotics Autom 2001;17(5):761–6. vector formulations and damped least-squares methods. IEEE Trans
[3] Norali P, Wentao Y, Rajiv D, Wilfrido M. Development of a Robotic Syst Man Cyber 1986;16:93–101.
Haptic Interface to Assist the Performance of Vocational Tasks by [15] Jeong Y, Lee D, Kim K, Park JO. A wearable robotic arm with high
People with Disabilities. In: Proceedings of the 2002 IEEE interna- force-reflection capability. In: Proceedings of the 2000 IEEE
ARTICLE IN PRESS
Y. Chen et al. / Robotics and Computer-Integrated Manufacturing 23 (2007) 478–487 487
International workshop on robot and human interactive commu- [18] Amago T. Sizing optimization using response surface method
nication, September 2000, p. 27–9. in FOA. R&D Review of Toyota CRDL, vol 37(1), 2002,
[16] Yang CJ, Niu B, Zhang JF, Chen Y. Different structure-based p. 1–7.
control system of the PUMA 560 manipulator with an arm [19] Shanghai Science and Technology Station. The Method of the
exoskeleton. In: Proceedings of the 2004 IEEE conference on Orthogonal Experiment Design—multi-factors Experiment Method.
robotics, automation and Mechatronics, Singapore, p. 572–7. Shanghai: Shanghai People’s Press; 1975.
[17] Cheng FT, Hour TL, Sun YY, Chen TH. Study and resolution of [20] Manjunath S. The triplex design group of Chinese Association of
singularities for a 6-DOF PUMA 560 manipulator. IEEE Trans Syst Statistics: orthogonal method and triplex design. Beijing: Science
Man Cybern 1997;27(2):332–43. Press; 1987.