Cable Path Optimization by Fixing Multiple Guides

on One Link for Industrial Robot Arms

Shintaro Iwamura (  ninbas2000@hotmail.com )
Omron (Japan) https://orcid.org/0000-0003-0451-3383
Yoshiki Mizukami
Takahiro Endo
Fumitoshi Matsuno

Research Article

Keywords: industrial robot arm, cable path optimization, cable geometry simulation, Hermite interpolation,
rotation minimizing frame

Posted Date: May 18th, 2023

DOI: https://doi.org/10.21203/rs.3.rs-2906743/v1

License:   This work is licensed under a Creative Commons Attribution 4.0 International License.
Read Full License
Cable Path Optimization by Fixing Multiple Guides on One Link for
Industrial Robot Arms

Shintaro Iwamura1, Yoshiki Mizukami2*, Takahiro Endo3, Fumitoshi

Matsuno 4

1
Software Development Dept., Controller Div., Product Business Division H.Q.,
Industrial Automation Company, Omron Corporation, Shimogyo-ku, Kyoto 600-8530,
Japan

2
Graduate School of Science and Engineering, Yamaguchi University, Ube, Yamaguchi
755-8611, Japan

3
Department of Mechanical Engineering and Science, Kyoto University, Kyodai-
katsura, Nishikyo-ku, Kyoto 615-8540, Japan

4
Department of Electronics and Information Systems Engineering, Osaka Institute of
Technology, 5-16-1 Omiya, Asahi-ku, Osaka, 535-8585, Japan

Corresponding author: Yoshiki Mizukami (e-mail: mizu@yamaguchi-u.ac.jp).

Cable Path Optimization by Fixing Multiple Guides on One Link for
Industrial Robot Arms

Abstract: Several methods have been developed in recent years to automatically

design cable paths for a given robot motion. However, an optimal solution may not
be found when applied to complex motions such as inspection or assembly
processes. Observation of experts’ works reveals that multiple guides are
empirically fixed on one link to avoid cable collisions with peripheral devices and
to lessen cable bending fatigue. This study proposes a cable path optimization
method with multiple guides on one link to obtain the optimal solution for complex
motions. Further, it experimentally demonstrates its effectiveness. However, fixing
multiple guides on one link may increase computational burden because the
solution search space will remarkably expand. Therefore, a combination of
Hermite interpolation and interpolated coordinate system calculation using rotation
minimizing frames was introduced to reduce the significant burden of connected
geometry computation. The connected geometry is the geometry of the cable
segment which connects the guides at both ends and is used as an initial geometry
in the physical simulation considering gravity to determine the convergent
geometry of the cable segment.

Keywords: industrial robot arm; cable path optimization; cable geometry

simulation; Hermite interpolation; rotation minimizing frame.

1. Introduction

Robot arms are an essential part of the production equipment in the industrial sector: ~3.0
million units are reported to be in use worldwide in 2021, with ~380,000 new units
shipped [1]. The robot arm is outfitted with external cables for the purpose of mediating
actuators, video and sensor control, power supply, pneumatics, and so on. Cable path
design is carried out with the objective of not interfering with the motion, i.e., avoiding
damage to the cable caused by motion. The engineer repeats the cable path design and
implementation testing while observing the robot motion. While these series of trial-and-
error processes are mentally and physically demanding, they are also time-constrained in
the sense that work delays can delay the start of the production line.
In cable path optimization, it is necessary to physically simulate the geometric
deformation of the cable due to the joint angle transition of the robot arm. Cosserat theory
[2] is a fundamental theory of micropolar elastic bodies for describing cables, defining
independent degrees of freedom of translation, and rotation for each element of the cable.
Spillmann and Teschner [3] established a model that divides the cable into one-
dimensional (1D) discrete elements along the centerline. They defined the position and
orientation of each cable element and calculated the continuous deformation energy of
each element using the finite element method. Hergenröther and Dähne [4] divided the
cable into rigid cylindrical elements and connected them with ball joints. Meanwhile,
Grégoire and Schömer [5] represented bending and torsional energy separately and
employed a generalized mass-spring model. Loock and Schömer [6] presented cable
torsion using torsion springs. Lv et al. [7] performed physical simulations of cable
harnesses, using linear springs for stretch, bending springs for bending, and torsion
springs for geometric and material torsion. The position-based dynamics (PBD)
framework has been proposed to increase simulation speed and stabilize rigid-body
collision detection [8]. It has been implemented in physics engines such as Bullet [9],
Havok [10], and PhysX [11]. In this study, the PhysX spring model is employed for cable
geometry simulation.
Kressin [12] and Carlson et al. [13] applied the cable dress pack geometry
simulation to the robot path optimization problem to estimate the path with less stress on
the cable dress pack. A dress pack is a flexible tubular part attached to the robot arm to
protect the cable. Hermansson et al. [14] optimized motions with respect to quasi-static
motion and deformation by introducing inertia into the system. These approaches
optimize the robot trajectory by fixing the configuration of the dress pack, i.e., the cable
path. However, they do not discuss the optimization of the cable path itself.
Iwamura et al. [15] proposed a cable path optimization method for given Pick-
and-Place motions. The entire cable was decomposed into cable segments separated by
guide positions. A candidate set of parameter vectors (PVs) consisting of the cable
segment lengths and the configurations of the guides at both ends was prepared. Cable
stress was quantified by physically simulating the cable segment geometry during motion
for each PV. However, this tracking was considerably burdensome. Therefore, Iwamura
et al. [16] defined several static robot poses that are prone to cable stress during motion
as critical poses. The cable stresses for these critical poses were quantified with the
significantly reduced burden.
This paper discusses the design of the optimal cable path for a given robot motion.
Here, the motions of the robot arm are roughly classified into two types. The first is a
simple motion in which the end link accesses the object from above only, as in Pick-and-
Place operations. The other is a complex motion in which the end link accesses the object
from multiple directions, as in assembly or inspection operations. Compared to simple
motions, complex motions tend to make cable path design more difficult because of the
variety of robot arm working poses. Several attempts to observe or imitate the procedures
of experts to solve engineering problems have been reported [17, 18]. Therefore, we
observed the procedures of cable path design experts and found the following. In the
design of experts for simple motions, basically, one guide is attached to one link.
Meanwhile, for complex motions, multiple guides are often attached to one link. This can
be interpreted as an attempt to limit the movement of the cable by increasing the guides
for a given motion to avoid interference with the peripheral objects and lessen bending
cable fatigue at the connection to sensor devices and other equipment.
This study proposes a cable path optimization method that handles complex
motions by fixing multiple guides on one link. As the number of guides increases, the
computational burden increases due to the expansion of the solution search space. As a
result of profiling the computational burden of each process in the previous method [16],
the computation of connected geometry turned out to have a relatively heavy burden. The
connected geometry is the geometry of the cable segment that connects the guides at both
ends and is used as an initial geometry in the physical simulation considering gravity to
determine the convergent geometry of the cable segment. Fig. 1 shows the iterative
deformation of the cable segment in the previous method [16]. The simulation of the cable
segment geometry starts with the linear geometry, which connects one end to one guide
(Fig. 1(a)) and moves the other end to match the other guide kinematically (Fig. 1(b)–
(d)). However, the process of transforming the linear geometry into a cable segment
geometry connecting the guides, i.e., the connected geometry (Fig. 1(d)), was
computationally demanding. Therefore, in addition to the above-mentioned proposal, fast
connected geometry computation using Hermite interpolation [19] and rotation
minimizing frame (RMF) [20] is proposed.
 (a)1st step (b) 33rd step (c) 66th step (d) 100th step

Fig. 1 Deformation of a cable segment from the linear geometry (1st step) to the
connected geometry (100th step) in the previous study

The reason for employing a Hermite interpolation is that the geometry can be
computed simply by giving the positions and tangent vectors of the two ends. For an nth-
order spline interpolation [21], n+1 control points through which the cable segment passes
are required. For an nth-order Bezier curve [22], n+1 control points are required between
the start and end points. However, no control points other than the start and end points
are passed through. Spline and Bezier curves are unsuitable for computing cable geometry
between guides because these control points are not given. The Frenet–Serret formula is
widely used to calculate interpolating coordinate systems. However, continuity between
adjacent coordinate systems is not guaranteed [23]. Therefore, RMF is employed to solve
this problem.
The rest of this paper is organized as follows. Section 2 presents the modeling and
problem setting. Section 3 describes the proposed method. Section 4 discusses the
reduction of computational burden for connected geometry of cable segments. Sections 5
and 6 present the simulation and experimental results, respectively. Finally, Section 7
provides the conclusion.
2. Modeling and problem setting

2.1. Modeling of links, guides, and cable segments

Fig. 2 shows a robot arm consisting of N+1 links 𝐿𝑛 (𝑛 = 0, … , 𝑁) connected by N joints.

Link 𝐿0 is fixed to the platform. The nth joint angle representing the relative angle of the
link 𝐿𝑛 to link 𝐿𝑛−1 is expressed in terms of θ𝑛 (𝑡) as a function of time t. Let coordinate
system Σ𝐿0 be the absolute coordinate system with the origin at the base point of link 𝐿0 .
The coordinate system Σ𝐿𝑛 be the local coordinate system of 𝐿𝑛 with the origin at the base
point of 𝐿𝑛 (𝑛 = 1, … , 𝑁). Some links possess a mounting adapter, which is usually a pair
of screw holes for fixing optional devices and is indicated by black dots. The local
coordinate system specified by the mounting adapter 𝐴𝑛 on 𝐿𝑛 is represented by Σ𝐴𝑛 . The
coordinate system of 𝐿𝑛 , Σ𝐿𝑛 , at t is represented as a nonlinear function of the joint angle
Θ(𝑡) = [θ1 (𝑡), θ2 (𝑡), … , θ𝑁 (𝑡)]𝑇 .

Fig. 2 Modeling of links, guides, and cable segments

Cable segments may be connected between links that are more than one link apart because
guides cannot be fixed on links that do not have a mounting adapter. The cable guide
𝐺𝑚 (𝑚 = 1, … , 𝑀) is attached to 𝐿𝑛 (𝑛 = 1, … , 𝑁). Fig. 2 shows that 𝐺2 and 𝐺3 are fixed
to the same link 𝐿2 . 𝐶2 , which is connected to them, does not traverse any joint. In this
study, cable segments such as 𝐶2 are referred to as intra-link segments. Cable segments
traversing one or more joints such as 𝐶1 and 𝐶3 are referred to as inter-link segments. The
introduction of intra-link cables has made it possible to attach two or more guides per
link, which was not possible in previous studies [16]. In this study, the index 𝑚 of the
cable segment is numbered along the cable segments. The 𝑚 of the cable segment
coincides with the index of the guide at the root of the cable segment. The index 𝑛 of the
link is numbered along the links from the root link.
The configuration of 𝐺𝑚 in Σ𝐿𝑛 is given by the relative configuration 𝑃𝑚 with
respect to Σ𝐴𝑛 . Therefore, the configuration of 𝐺𝑚 in Σ𝐿0 is a function of joint angles
Θ(𝑡) = [θ1 (𝑡), θ2 (𝑡), … , θ𝑛 (𝑡)]T , Σ𝐴𝑛 and 𝑃𝑚 . The entire cable can be modeled as a set
of cable segments {𝐶𝑚 }(𝑚 = 1, … , 𝑀 − 1) [16]. One end of 𝐶𝑚 is attached to 𝐺𝑚 , and
the other end is attached to guide 𝐺𝑚+1 .
A cable segment is composed of multiple rigid sub-segments and virtual mass-
spring joints [7, 24]. The virtual joint with six degrees of freedom constrains two adjacent
sub-segments. As shown in Fig. 3, 𝐶𝑚 traversing two joints connects 𝐿𝑛 and 𝐿𝑛+2 via 𝐺𝑚
and 𝐺𝑚+1 , which are composed of 𝑆𝑚 sub-segments with a fixed-length 𝑙𝑠𝑠 and
𝑆𝑚 virtual joints.
Thus, the length of 𝐶𝑚 is expressed as 𝑙𝑐𝑚 = 𝑆𝑚 × 𝑙𝑠𝑠 . The first sub-segment of 𝐶𝑚
is connected to 𝐺𝑚 so that the configurations of the sub-segment and 𝐺𝑚 are the same.
The 𝑆𝑚 th sub-segment is connected to the first sub-segment of the next cable segment
𝐶𝑚+1 . The geometry of 𝐶𝑚 is acquired on the basis of the configurations of 𝑆𝑚 sub-
segments.
 Fig. 3 Sub-segments comprising a cable segment

2.2．Problem setting

As in the previous works [15, 16], the objective of this study is to design a cable path with
the shortest cable length that satisfies three stress constraints for a given robot motion.
The cable path is represented by (𝑆1 , … , 𝑆𝑀 , 𝑃1 , … , 𝑃𝑀+1 ), which includes the length of
the cable segment 𝑆𝑚 and the configuration 𝑃𝑚 of the guides supporting both ends of the
cable segment. This parameter vector is called the entire cable parameter vector (ECPV).
Then the cable path design can be formulated as an optimization problem with respect to
the ECPV, which is shown in the following equation.

Given Θ(𝑡) = [θ1 (𝑡), θ2 (𝑡), … , θ𝑛 (𝑡)]T , find

(𝑆1 , … , 𝑆𝑀 , 𝑃1 , … , 𝑃𝑀+1 ) = 𝑎𝑟𝑔 𝑚𝑖𝑛 ∑𝑀

𝑚=1 𝑆𝑚 ,

𝑠ubject to

𝐼𝑚𝑝𝑢𝑙𝑠𝑒𝑡ℎ ≥ 𝐼𝑚𝑝𝑢𝑙𝑠𝑒(𝑆̂𝑚 ,𝑃𝑚,𝑃𝑚 + 1) (Θ(𝑡)) ,

(1)
𝑆𝑡𝑟𝑒𝑡𝑐ℎ𝑡ℎ ≥ 𝑆𝑡𝑟𝑒𝑡𝑐ℎ(𝑆̂𝑚 ,𝑃𝑚 ,𝑃𝑚 + 1 ) (Θ(𝑡)) ,

𝐶𝑢𝑟𝑣𝑎𝑡𝑢𝑟𝑒𝑡ℎ ≤ 𝐶𝑢𝑟𝑣𝑎𝑡𝑢𝑟𝑒(𝑆̂𝑚 ,𝑃𝑚 ,𝑃𝑚 + 1) (Θ(𝑡)) ,

for 0 ≤ 𝑡 ≤ 𝑇, 𝑚 = 1,2, … , 𝑀,

∀𝑆̂𝑚 ∈ {𝑆̂𝑚 ∈ ℕ| 𝑆𝑚 − 𝑅 ≤ 𝑆̂𝑚 ≤ 𝑆𝑚 + 𝑅},

where 𝑆𝑡𝑟𝑒𝑡𝑐ℎ(𝑆𝑚 ,𝑃𝑚 ,𝑃𝑚 + 1) (Θ(𝑡)), 𝐶𝑢𝑟𝑣𝑎𝑡𝑢𝑟𝑒(𝑆𝑚 ,𝑃𝑚 ,𝑃𝑚 + 1) (Θ(𝑡)) are the stress
constraints of the impulse, stretch and curvature radius, respectively, described in Section
3. In order to cope with fluctuations in the physical properties of the cable such as the
variation of the cables’ materials, errors in cable installation, and cable degradation over
time, robustness with respect to the number of sub-segments 𝑆𝑚 is introduced so that even
if the number of sub-segments is increased or decreased within a predefined robustness
R [15], its PV will satisfy the stress constraints described above.

3．Proposed method: Cable path optimization by fixing multiple guides on one

link

In this study, we propose a cable path optimization method for a complex motion as an
extension of the previous methods [15, 16]. As described in Section 1, the previous
1)
method [16] performs static cable geometry simulation for critical poses in the
attachment test to reduce the number of PV candidates and then applies dynamic cable
geometry simulation for robot motions in the motion test to find the ECPV with the
shortest cable length that satisfies the stress and robustness constraints. Note that the
previous method, which assumes only simple motion, fixed up to one guide to one link
so that the attachment test dealt with only the inter-link segments. On the other hand, the
proposed method fixes more than two guides to one link so that the attachment test deals
with both the inter-link and intra-link cables. This allows us to realize the cable path
optimization for the complex motion.

1)
Critical poses [16] are robot poses that are prone to applying stress to the cable with impulse,
stretching, bending, and other movements. Various critical poses are extracted for each cable segment
based on the robot motion. A critical pose is defined as a set of joint angles with a global extreme
value. In critical poses, satisfying the stress constraints in Eq. (1) is difficult. Herein, it is assumed that
if a cable path can satisfy the stress constraints in all critical poses, the cable path can also satisfy the
stress constraints in other motion poses.
 (b) Inter-link Attachment test module

(c) Intra-link attachment test module

(a) Entire procedure

Fig. 4 Optimization procedure in the robotic arm cable-path planning

 Table 1 Definition of symbols for PV candidates in the optimization procedure

𝐷0,𝑚 The PV candidate set of 𝐶𝑚

𝐷1,𝑚 The PV candidate set satisfying stress constraints in the attachment test

𝐷2,𝑚 The PV candidate set satisfying stress constraints and a robustness constraint
on the cable length in the attachment test

𝐷′0,𝑚 The PV candidate set satisfying the adjacency condition between 𝐷0,𝑚 and
𝐷2,𝑚−1 of 𝐶𝑚−1 in the attachment test

𝐷3 The ECPV set satisfying the adjacency condition between all cable segments

𝑑̂ The ECPV candidate with the shortest entire cable length in 𝐷3 in the motion test

𝑑∗ The optimal ECPV satisfying the stress and robustness constraints and the
adjacency condition with the shortest entire cable length

As described in Section 2, the entire cable is represented by combining two types

of cable segments: intra-link and inter-link segments. Fig. 4(a) shows the entire procedure
for cable path optimization, and Table 1 summarizes the definition of symbols for PV
candidates in the optimization procedure. Note that to deal with both inter-link and intra-
link cable segments, the attachment test module shown by the bold frame in Fig. 4(a) can
be switched to one of the modules shown in Fig. 4(b) and (c).
The PV of a 𝐶𝑚 comprises 𝑆𝑚 (number of sub-segments of 𝐶𝑚 ), 𝑃𝑚
(configuration of 𝐺𝑚 ), and 𝑃𝑚 + 1 (configuration of 𝐺𝑚 + 1 ). The PV candidate set 𝐷0,𝑚
of 𝐶𝑚 is generated based on the combination of all these parameters [16] (Appendix A).
In the attachment test, the cable segment geometry is simulated for static critical
poses. Here, a cascade structure for the attachment test (CSAT) was introduced in [16].
It cascades attachment test modules for each cable segment. The burden is reduced by
removing the PVs that are connected to the PVs rejected in the previous module. In the
final step of the attachment test, a set of entire cable parameter vectors (ECPVs) 𝐷3 is
generated by combining the PVs of the cable segments that satisfy the adjacency
condition.
 Then, the motion tests for the cable segments are performed on the ECPV 𝑑̂ ,
where the 𝑑̂ has the shortest cable length in 𝐷′3 that is initially copied from 𝐷3 . The stress
and robustness constraints through the motion are checked [15]. If all the cable segments
pass the motion test, 𝑑̂ is adopted as the ECPV 𝑑∗ representing the optimal cable path.
Otherwise, the next shorter 𝑑̂ is selected from 𝐷′3 , and the motion test is repeated.
Fig. 4(b) shows the attachment test module for an inter-link cable segment. Since
the inter-link cable segment 𝐶𝑚 traverses one or more joints between the two guides, the
critical pose set Θ𝑚 contains multiple critical poses. Attachment simulations are
performed in parallel for these critical poses to obtain the PV candidate set 𝐷1,𝑚 that
satisfy the stress constraints on impulse, stretch, and curvature radius from 𝐷0,𝑚 .
Furthermore, we obtain the PV candidate set 𝐷2,𝑚 satisfying the lengthwise robustness
constraints from 𝐷1,𝑚 [15].
Fig. 4(c) shows the attachment test module for the intra-link segment 𝐶𝑚 . The
intra-link segments do not contain joints between the guides. In other words, the
geometries of intra-link segments are motion independent. Thus, they do not have critical
poses to which the attachment simulation should be applied. Therefore, we run the
attachment simulation on 𝐷0,𝑚 for the set Θ′𝑚 that contains only the initial pose as one of
the arbitrary poses. This yields the PV candidate set 𝐷1,𝑚 that satisfies the stress
constraints (Appendix B).
The attachment test for intra-link segments can be performed even if the joint
angle time series Θ(t) representing the motion is not given. Therefore, the attachment test
module for each intra-link segment for the given motion can be omitted by referring to a
dictionary database that has accumulated the results of attachment tests for intra-link
segments conducted in the prior experiment.
The proposed method can avoid adding unnecessary intra-link cables and guides
for the given motion by setting 0 as one of the candidate values for the number of sub-
segments 𝑆𝑚 . If 𝑆𝑚 is 0, the configurations of 𝐺𝑚 and 𝐺𝑚+1 are identical, and cable 𝐶𝑚
is practically non-existent. In this case, 𝐶𝑚−1 and 𝐶𝑚+1 are directly connected cable paths
via 𝐺𝑚 (or 𝐺𝑚+1 ).
In the attachment test module for 𝐶𝑚 (2 ≤ m), the candidate PV set 𝐷′0,𝑚 is
generated from the PV set 𝐷2,𝑚−1 obtained in the previous module and initial PV
candidate set 𝐷0,𝑚 using the adjacency condition (Appendix C).
4. Reduction of computational burden for connected geometry of cable segments

Hermite interpolation and RMF are employed to reduce the computation burden for
generating the connected geometry of the cable sub-segments. At first, Hermite
interpolation is used to calculate the positions of the sub-segments. However, due to the
nature of Hermite interpolation [19], the interpolated points are not uniformly spaced.
Therefore, sampling and linear interpolation are used to calculate uniformly spaced sub-
segment positions. The orientations of the sub-segment positions are then obtained using
RMF [20] based on the sub-segment positions and the orientation of the start guide. Next,
the error of the torsion angle between the orientations of the end sub-segment and the end
guide is obtained. Finally, the orientation of each sub-segment is proportionally modified
based on the obtained error so that the end sub-segment and the end guide have the same
orientation and all the sub-segments have continuous orientations.
The connected geometry of a cable segment is given by the sub-segment positions
{𝑝𝑠 |1 ≤ s ≤ 𝑆} and orientations {𝑞𝑠 |1 ≤ s ≤ 𝑆}, where 𝑆 = 𝑆𝑚 + 1 since the geometry
is represented with 𝑆𝑚 sub-segments of 𝐶𝑚 and the first sub-segment of 𝐶𝑚+1 as
described in Fig. 3. Let us describe how to calculate {𝑝𝑠 } and {𝑞𝑠 } using the Hermite
interpolation and RMF.

4.1. Hermite interpolation

Given the positions of the endpoints (𝑃̂0 and 𝑃̂1 ) and tangent vectors (𝑉̂0 and 𝑉̂1 )
associated with 𝑃̂0 and 𝑃̂1 , the cubic Hermite interpolated points 𝑝(𝑡) are obtained by

𝑝(𝑡) = (2𝑡 3 − 3𝑡 2 + 1)𝑃̂0 + (𝑡 3 − 2𝑡 2 + 𝑡)𝑉̂0 + (−2𝑡 3 + 3𝑡 2 )𝑃̂1

+(𝑡 3 − 𝑡 2 )𝑉̂1 , (2)

where t is a parameter with a [0:1] variant. 𝑃̂0 and 𝑃̂1 are the guide positions obtained
from guide configurations 𝑃𝑚 and 𝑃𝑚+1 , respectively. 𝑉̂0 and 𝑉̂1 are expressed using the
length parameter α and unit vectors 𝑣0 and 𝑣1 of the cable tangential direction obtained
from the guide configuration 𝑃𝑚 and 𝑃𝑚+1 , respectively.

𝑉̂0 = 𝛼𝑣0 , (3)

 𝑉̂1 = 𝛼𝑣1 , (4)

where a larger α increases the total length of the Hermite curve and vice versa. The
interpolated points {𝑝′ 𝑠 |1 ≤ 𝑠 ≤ 𝑆 ′ , 𝑆 ≪ S′ } are given by using Eq. (2) with an interval
of 𝑡 as shown in Fig. 5(a). The Newton-Raphson method [25] is used to obtain the
appropriate value of α corresponding to the desired curve length, that is, the cable segment
length 𝑙𝑐𝑚 . The total length of the Hermite curve is obtained as the sum of the distances
′
between the adjacent interpolated points, ∑𝑆1 −1‖𝑝′ 𝑠+1 − 𝑝′ 𝑠 ‖ . Sampling and linear
2
′
interpolation are then performed on {𝑝 𝑠 } to generate uniformly spaced sub-segment
positions {𝑝𝑠 |1 ≤ 𝑠 ≤ 𝑆} so that the adjacency distances are equal to the sub-segment
fixed-length 𝑙𝑠𝑠 as shown in Fig. 5(b).

(a) Two control points (green) (b) Uniformly spaced points as

and interpolated points (blue) the sub-segment positions
on Hermite curve
Fig. 5 Sub-segment positions based on Hermite interpolation

4.2 RMF
The continuous orientations {𝑞′𝑠 |1 ≤ s ≤ 𝑆} at the sub-segment positions, as
shown in Fig. 6(a), are obtained by applying RMF to {𝑝𝑠 } and the orientation of the start
guide’s configuration 𝑃𝑚 . The orientation 𝑞1 at the first sub-segment of 𝐶𝑚 is equivalent
to the orientation of 𝑃𝑚 due to the characteristics of RMF, but the orientation 𝑞𝑠 at the
last sub-segment is not necessarily equivalent to that of 𝑃𝑚+1 . Therefore, the error ∆𝜓 of
the torsion angles between the orientation 𝑞′𝑆 at the 𝑆-th sub-segment and orientation of
the end guide’s configuration 𝑃𝑚+1 is obtained. Finally, by adding the proportionally-
allocated torsion angle modification ∆𝜓(𝑠 − 1)/(𝑆 − 1) to the continuous orientation at
each sub-segment, {𝑞′𝑠 }, the sub-segment orientations {𝑞𝑠 |1 ≤ 𝑠 ≤ 𝑆} are determined as
shown in Fig. 6(b).

(a) Continuous orientations {𝑞 ′ 𝑠 } (b) Sub-segment orientations {𝑞𝑠 }

given by RMF at the sub-
segment positions

Fig. 6 Sub-segment orientations based on RMF and torsion modification

5. Simulation

This study investigates the effectiveness of fast connected geometry computation using
Hermite interpolation and RMF, as well as the cable path design using intra-link
segments. In this section, a comparative study of the previous method and two proposed
methods shown below will be conducted.
Previous method: This is the method proposed in [16]. The connected geometry
of cable segments is generated by deforming a linear geometry using a physical
simulation. All cable segments are assumed to be inter-link segments, i.e., the maximum
number of guides per link is 1.
 Proposed Method 1: The fast connected geometry computation is employed to
reduce the computational burden of the cable segment connected geometry in the previous
method. Al l cable segments are assumed to be inter-link segments.
Proposed Method 2: In addition to the fast connected geometry computation, both
the previous inter-link and proposed intra-link segments are employed to design the cable
path for complex motion, i.e., the maximum number of guides per link is assumed to be
two.
For clarity, the employments of fast connected geometry computation and intra-
link segments for each method are summarized in Table 2.
The specifications of the computer used are as follows: the OS was Windows 10
Professional. The CPU was an Intel Core i9-7980XE 2.6GHz. The RAM was 16GB. The
GPU was an NVIDIA GeForce RTX 2080.
The parameterization of guides and cables is described in Section 5.1. Next,
comparison experiments will be conducted in Section 5.2. on the cable geometry and
computational time obtained by the fast connected geometry computation with those
obtained by the previous connected geometry computation [16]. Then, the cable path
solution and computational time of the three methods when applied to a simple motion,
namely pick-and-place motion, are described in Section 5.3. Finally, the cable path
solution and computational time when the three methods are applied to a complex motion,
namely inspection motion, are described in Section 5.4.

Table 2 Cable path optimization methods to be investigated

Fast connected Intra-link segment

geometry computation

Previous method - -

Proposed method 1 Employed -

Proposed method 2 Employed Employed

5.1. Parameterization of guides and cables

A 6-axis articulated robot (Omron Viper 850 [26]) was used in the simulation. In the pick-
and-place motion, a motorized gripper (SMC LEHF32EK2-32 [27]) was attached to the
end link. A cable (SMC LE-CP-3 [28]; diameter 8 mm; weight 260 g/m; hereinafter SMC
cable) was used to transmit power and control signals to a motorized gripper. A camera
(BASLER acA1300-30gc [29]) was attached to the termination link in the inspection
motion. A cable (GEV-C6HF-L05-ACE [30]; diameter 6 mm; weight 662 g/m; hereafter,
GEV cable) was used to transmit the camera power and imaging signals. These cables are
the subjects of the cable geometry simulation in respective motions.
For the physical simulation of the cable geometry, the mass-spring model of
PhysX SDK 4.0 [11] was employed with the rigid body segments connected by virtual
joints with six degrees of freedom. Here, the bending spring constant, torsion spring
constant, and damping coefficient are as shown in Table 3, based on the measurement
method described in a previous study [15].
The parameters included in the stress constraint for the cable geometry are
described. To avoid a collision with surrounding objects, 𝐼𝑚𝑝𝑢𝑙𝑠𝑒𝑡ℎ was set to 0.
𝑆𝑡𝑟𝑒𝑡𝑐ℎ𝑡ℎ was set at 102% to deal with the solution oscillations, although the natural
limit in terms of physical stress is 100%. 𝐶𝑢𝑟𝑣𝑎𝑡𝑢𝑟𝑒𝑡ℎ for SMC and GEV cables were
48 mm and 60 mm according to the respective specifications.

Table 3 Physical parameters of the cable

Target cable Bending stiffness Twist stiffness Dumping coefficient

SMC cable 5.1 × 10−2 Nm/rad 1.5 Nm/rad 1.0 × 102 Ns/m

GEV cable 6.7 × 10−2 Nm/rad 2.1 Nm/rad 1.0 × 102 Ns/m
(a) Without intra-link segments (b) With intra-link segments

(Previous method and proposed method 1) (Proposed method 2)

Fig. 7 Labeling of links, guides, and cables

Fig. 7(a) and (b) show the labels of the links, guides, and cables as well as the
absolute coordinate system for the cases without (previous method and proposed method
1) and with (proposed method 2) intra-link segments, respectively. The robot arm consists
of 𝐿0 to 𝐿6 . As mentioned earlier, the origin of the absolute coordinate system Σ𝐿0 is fixed
at the base of 𝐿0 . The origin of Σ𝐿n is fixed at the origin of rotation of 𝐿𝑛 (1 ≤ 𝑛 ≤ 6).
Guide 𝐺1 is attached to 𝐿3 . Guides 𝐺2 and 𝐺3 are attached to 𝐿4 . Guide 𝐺4 is attached to
𝐿6 . Guide 𝐺5 coincides with the connector of the gripper or camera installed on 𝐿6 in each
of the pick-and-place and inspection motions.
Fig. 7 shows the case without intra-link cables (previous and proposed method 1).
It corresponds to the case where 𝐺2 and 𝐺3 , as well as 𝐺4 and 𝐺5 , are identical, i.e., the
number of sub-segments 𝑆2 and 𝑆4 in 𝐶2 and 𝐶4 are 0.
Candidate sub-segment numbers 𝑆m of 𝐶m for pick-and-place and inspection
motions are shown in Table 4. In the previous method and proposed method 1, the
candidate for 𝑆m in 𝐶2 and 𝐶3 is only 0. Meanwhile, in the proposed method 2, they are
the sum set of 0 and arithmetic series. On the right side of the table, the number of 𝑆m
candidates is indicated. The unit sub-segment length 𝑙𝑠𝑠 was set to 10 mm. The unit sub-
segment consists of a cylinder that is 5 mm long and 8 mm in diameter with a 5 mm void
(virtual joint). The tolerance ∆𝑆 for the arithmetic series of cable segment lengths was 1.
 𝑙𝑅 20mm
The range parameter R for robustness is 2(= = ). Thus, the variation 𝑙𝑅 (=
∆𝑆×𝑙𝑠𝑠 1×10mm

20 mm) with respect to the length of the cable segment can be allowed.

Table 4 Candidates for segment number 𝑆m in cable path optimization

Segment Segment Number of sub-segments {𝑆𝑚 }

Number of
parameters
Index Type Pick-and-place motion Inspection motion

𝐶1 Inter-link {15, 16, ⋯ ,24, 25} {45, 46, ⋯ , 54,55} 11

Previous method and proposed method 1: {0} 1

𝐶2 Intra-link
Proposed method 2: {0} ∪ {10,11, ⋯ ,19,20} 12 (=1+11)

𝐶3 Inter-link {35,36, ⋯ ,49, 50} {30,31, ⋯ ,44, 45} 16

Previous method and proposed method 1: {0} 1

𝐶4 Intra-link Proposed method 2

Proposed method 2
{0} ∪ 27 (=1+16)
{0} ∪ {35,36, ⋯ ,49,50}
25,26, ⋯ ,39,40}

Next, we describe the candidate configurations 𝑃m of 𝐺m (1 ≤ 𝑚 ≤ 5). In Fig.

8(a), the motorized gripper used for pick-and-place motion is attached to the final link
(𝐿6 ). In Fig. 8(b), the camera used for the inspection motion is attached to the final link.
The candidates for the configuration 𝑃1 of 𝐺1 are shown in red in the figure, where z = 0
and (𝜑, 𝜃, 𝜓) = (0,0,0) are used for drawing convenience. 𝑃1 and 𝑃2 candidates are shown
in green. 𝑃4 candidates are shown in blue and yellow. The candidate for 𝑃5 is the only
one shown in yellow. Candidates for 𝑃m (1 ≤ 𝑚 ≤ 5) in the pick-and-place and
inspection motions are shown in Tables 5(a) and 4(b), respectively.
 (a) Pick-and-place motion (b) Inspection motion

Fig. 8 Candidate configurations of guides on 𝐿3 , 𝐿4 , and 𝐿6

Table 5 Candidates for guide configurations 𝑃𝑚 in cable path optimization

(a) Pick-and-place motion

- Position Orientation Num. of

comb.
- 𝑥 𝑦 𝑧 𝜑 𝜃 𝜓

𝑃1 {0} {−40,0,40} {25} {0} {0} {−90, −45,0,45,90} 30

𝑃2 , 𝑃3 {0, 50} {−60,0,60} {25} {0} {0} {−90, −45,0,45,90} 60

𝑃4 (𝑥, 𝑦, 𝑧, 𝜑, 𝜃, 𝜓) ∈ {(150,70,40,0,0,0), 4

(150, −70,40,0,0,0), (190,0,40,0,0,90), (75,30, −200,0,0,0)}}

𝑃5 75 30 −200 0 0 0 1
 (b) inspection motion

- Position Orientation Num. of

comb.
- 𝑥 𝑦 𝑧 𝜑 𝜃 𝜓

𝑃1 {0} {−40,0,40} {25} {0} {0} {−90, −45,0,45,90} 30

𝑃2 , 𝑃3 {0, 50} {−60,0,60} {25} {0} {0} {−90, −45,0,45,90} 60

𝑃4 (𝑥, 𝑦, 𝑧, 𝜑, 𝜃, 𝜓) ∈ {(30, −50,0,0,0,0), 4

(30,50,0,0,0,0), (30,0, −55,0,0,90), (−65,0,120,0,0,0)}}

𝑃5 −65 0 120 0 0 0 1

5.2. Fast connected geometry computation

The cable geometries and computational times obtained by the previous and fast
connected geometry computations are compared in this section. Here, the connected
geometries of cable segments 𝐶3 with two lengths are computed. Fig. 9(a) and (b) show
the attachment of previous methods based on the deformation from linear geometry for
cable segments of 250 and 500 mm, respectively. In the initial state shown on the left side
of these figures, the cable segments were arranged as a linear geometry towards the x
direction. Then, as shown in the middle, connected geometries were obtained when the
physical simulation connected the ends of the cable segments to their respective guides.
By continuing the physical simulation up to the steady state, the convergent geometries
shown on the right side were obtained. Note that, in the previous connected geometry
computation, physical simulation was used for obtaining both the connected and
convergence geometries. For a cable segment of 250 mm (Fig. 9(a)), the connected
geometry smoothly connected the guides. The convergence geometry was approximately
identical to the connected geometry. Meanwhile, for a cable segment length of 500 mm
(Fig. 9(b)), the connected geometry was wiggly. However, the convergence geometry
was smooth.
 Linear geometry Connected geometry Convergent geometry

(a) Cable segment of 250 mm (𝑆m = 25)

Linear geometry Connected geometry Convergent geometry

(b) Cable segment of 500 mm (𝑆m = 50)

Fig. 9 Cable segment geometries by the previous method using deformation from linear
geometry
 Connected geometry Convergent geometry

(a) Cable segment of 250 mm (𝑆m = 25)

Connected geometry Convergent geometry

(b) Cable segment of 500 mm (𝑆m = 50)

Fig. 10 Cable segment geometries by the proposed method using Hermite interpolation
and RMF

Fig. 10(a) and (b) show the attachment of 250 and 500 mm cable segments, respectively,
using fast connected geometry computation. The Hermite interpolation and RMF were
used to obtain the connected geometries where both end-cable segments were connected
to the respective guides, followed by the physical simulation to obtain the convergence
geometries.
 Table 6 Computation time comparison for connected and convergent geometries of
cable segments

Cable 250 mm 500 mm

length

Connected Convergent Connected Convergent

Sum Sum
geometry geometry geometry geometry

Previous 0.75 s 0.09 s 0.84 s 1.21 s 0.84 s 2.05 s

method

Proposed 0.01 s 0.10 s 0.11 s 0.01 s 0.51 s 0.52 s

method

Next, Table 6 shows the time required to obtain the connected and convergence
geometries for the previous and proposed methods and sum of these times by the previous
and proposed methods. In terms of the computational time for the connected geometry,
the proposed method is approximately 80 and 120 times faster than the previous method
for the cable segment lengths of 250 and 500 mm, respectively. Meanwhile, there is no
significant difference between the two methods in terms of the computational time for
convergent geometries. In terms of the total time, the proposed method is approximately
8 and 4 times faster than the previous method for cable segment lengths of 250 and 500
mm, respectively. The results described here indicated that the proposed method gives
convergent geometries comparable to those of the previous methods. It requires less total
computational time to obtain convergent geometries.

5.3. Comparative simulation for simple motion

Cable path optimization for a pick-and-place motion as an example of simple motion was
studied. The initial pose of the robot arm is shown in Fig. 11(a). First, the robot arm picks
an object in the front (Fig. 11(b)). Next, the robot arm places the object on the right side
(Fig. 11(c)). The joint angle time series Θ(t) was generated based on a script describing
the way points and velocities, and so on, using a robot programming environment, the
Omron Sysmac Studio [31].

(a) Initial pose (b) Pick pose (c) Place pose

Fig. 11 Robot arm poses in the simple motion

𝜃4 [deg]
𝜃5 [deg]
𝜃6 [deg]

Fig. 12 Joint angle time series Θ(t) with the critical poses for the simple motion; the
large black dots represent the global extrema of the joint angles; the small black dots
represent the extrema of the joint angles during the period in which the other joint angle
has the global extremum
 The relationship between the critical poses extracted from the simple motion and
Θ(t) is shown in Fig. 12 [16]. The horizontal axis represents the elapsed time in simple
motion. The vertical axis represents the three joint angles. Since the inter-link segment
𝐶1 traversed only the fourth joint, the two critical poses of 𝐶1 were determined based on
the global extrema of 𝜃4 . Since the inter-link segment 𝐶3 traversed the fifth and sixth
joints, the four critical poses of 𝐶3 were determined based on the combination of the
extreme values of 𝜃5 and 𝜃6 .

Table 7 Number of solutions in the attachment test and the acquisition of the optimal
solution in motion test for simple motion

- Attachment test Motion

test
- 1st segment: 2nd segment: 3rd segment: 4th segment:
𝐶4 𝐶3 𝐶2 𝐶1

Previous method - 25 - 21 Solved

Proposed - 25 - 21 Solved
method 1

Proposed 16 285 179 77 Solved

method 2

Table 7 shows the number of solutions obtained in the attachment test and whether
the optimal solution was acquired in the motion test by "Solved" or “Unsolved.” The
previous and two proposed methods are used for the cable path optimization. The number
of solutions and solution sets obtained by the previous method and proposed method 1
were identical. Since the previous method and the proposed method 1 don’t use the 𝐶4
and 𝐶2 intra-link segments, numbers of their solutions are described as “-.” For the
proposed method 2, more solutions were obtained than the others in the attachment test.
The reason is that more PVs in 𝐷0,3 of 𝐶3 satisfy the adjacency condition for the
preceding 𝐶4 intra-link segment due to the increased configurations of 𝐺4 in 𝐷2,4 . The
same remark can be applied to the influence of the 𝐶2 intra-link segment on 𝐶1 . The
optimal solution is obtained in the motion test by all the three methods. The visual
representation of the optimal solution, i.e., ECPV d∗ , will be presented in Section 6.

Table 8 Computation times in the attachment and motion tests for simple motion

- Attachment test MotionTest Total

(min) Time
st nd rd th
- 1 seg. 2 seg. 3 seg. 4 seg. Total
(min)
𝐶4 𝐶3 (min) 𝐶2 (min) 𝐶1 (min) (min)
(min)

Previous - 8.1 - 32.3 40.4 1.3 41.7

method

Proposed - 2.1 - 1.9 4.0 1.3 5.3

method 1

Proposed 0.1 11.8 2.7 6.7 21.3 1.3 22.6

method 2

Table 8 shows the computational time required for the attachment and motion
tests for simple motion. In the attachment test, the total computational time for the inter-
link segments 𝐶3 and 𝐶1 is reduced by a factor of 10 for the proposed method 1. This
occurs due to the effect of the fast connected geometry computation as opposed to the
previous method. Meanwhile, the computational time for the proposed method 2, which
introduces the intra-link segments 𝐶4 and 𝐶2 , is approximately one-half. Note that
compared to the proposed method 1, although the increase in computational time for the
intra-link segment of the proposed method 2 is negligible, the computation time for the
inter-link segment is significantly increased. The is because more PVs in 𝐷0,3 of 𝐶3
satisfy the adjacency condition for the preceding 𝐶4 intra-link segment, as already
mentioned. Note that the computational times for the motion test with all the methods
were equal. The reason is that the ECPV d̂'s, which give the shortest entire cable length
and were obtained from the attachment tests by these methods, were consistent. This d̂
was adopted as d∗ through the motion test. The obtained results show that the fast
connected geometry computation is meaningful for load reduction. Applying intra-link
segments for simple motions is not recommended in terms of the computational burden.

5.4. Comparative simulation for the complex motion

The cable path optimization for an inspection motion as an example of complex motion
was studied. The initial pose of the robot arm is shown in Fig. 13(a). The camera at the
tip of the robot arm captures the left, top-left, top-right, and right sides in turn (Fig. 13(b)–
(c)).

(a) Initial pose (b) Left inspection pose

(c) Top-left inspection pose (d) Top-right inspection pose (e) Right inspection pose

Fig. 13 Robot poses in the complex motion

 𝜃4 [deg]
𝜃5 [deg]
𝜃6 [deg]

Fig. 14 Joint angle time series Θ(t) with the critical poses for the complex motion

The relationship between the critical poses extracted from the complex motion
and Θ(t) is shown in Fig. 14 [16]. Note that the variation ranges of 𝜃4 , 𝜃5 , and 𝜃6 in the
simple motion shown in Fig. 12 were less than approximately 50º, while those in the
complex motion were approximately 150º or more. Two and four critical poses were
extracted for 𝐶1 and 𝐶3 , respectively.

Table 9 Number of solutions in the attachment test and the acquisition of the optimal
solution in motion test for complex motion

- Attachment test Motion

Test
- 1st segment: 2nd segment: 3rd segment: 4th segment:
𝐶4 𝐶3 𝐶2 𝐶1

Previous method - 0 - 0 None

Proposed - 0 - 0 None
method 1

Proposed 19 30 246 199 Solved

method 2
 Table 9 shows the number of solutions obtained in the attachment test using the
three methods and acquisition of the optimal solution in the motion test with “Solved” or
“None.” The previous method and proposed method 1 could not obtain solutions at 𝐶3 in
the attachment test, and therefore could not obtain an optimal solution for the entire cable.
Meanwhile, the optimal solution was obtained with the proposed method 2. The results
show that the method with intra-link cables may provide an optimal solution for the
complex motions for which the cable optimization method without intra-link cables does
not provide an optimal solution. The visual representation of the optimal solution, i.e.,
ECPV d∗ , is presented in Section 6.

Table 10 Computation times in the attachment and motion tests for complex motion

- Attachment test Motion Total

Test Time
- 1st seg. 2nd seg. 3rd seg. 4th seg. Total
(min) (min)
𝐶4 𝐶3 (min) 𝐶2 𝐶1 (min)
(min) (min) (min)

Previous - 15.1 - - 15.1 - -

method

Proposed - 4.3 - - 4.3 - -

method 1

Proposed 0.2 10.5 2.8 14.0 27.5 1.7 29.2

method 2

Table 10 shows the computational time required for the attachment and motion
test for the complex motion. The attachment test for 𝐶3 took approximately 15 and 4
minutes of computational time for the previous and proposed method 1, respectively.
However, further computations were suspended for the reasons already mentioned. The
proposed method took 27.5 and 1.7 min for the attachment and motion tests, respectively,
resulting in a total of 29.2 min.
As mentioned in Section 3, the geometries of the intra-link segments 𝐶𝑚 are
motion independent. It implies that 𝐷1,𝑚 and 𝐷2,𝑚 can be dictionaried through a prior
processing without assuming any given motion. According to Table 8, the total
computational time of the proposed method 2 for the simple motion was 22.6 min. It can
be reduced to 19.8 min (88%) using this dictionary. Meanwhile, according to Table 10,
the total computational time for complex motions was 29.2 min, which can be reduced to
26.2 min (90%).

6. Experiment

The cable path was implemented on the actual robot based on the optimal solution d* for
simple motion obtained in Section 5.3. The implementation of the optimal cable path on
the initial, pick, and place poses in the simulation and real environments are shown in
Fig. 15(a)–(c) and Fig. 15(d)–(f), respectively.
The cable path was implemented on the actual robot based on the optimal solution
d* for the complex motion obtained in Section 5.4. The implementation of the optimal
cable path on the initial and inspection poses in the simulation and real environments are
shown in Fig. 16(a)–(c) and Fig. 16(d)–(f), respectively. As seen in Supplemental Videos
1 and 2, the robot is operated at approximately half the maximum speed from the
viewpoint of stability of the cable simulation and visual inspection of the implemented
cable geometry during the operation.
The simulated and experimental cable geometries are similar for both simple and
complex motions. It was also visually confirmed that the stress constraint on the cable
curvature radius were satisfied. There were no collisions between the cable and robot arm.
Although the possibility of cable collisions cannot be negligible when operating at
extremely high speeds, it is considered that the obtained optimal solution can be used as
a preliminary reference for practical cable path design.
 (a) Simulation: initial (b) Simulation: pick pose (c) Simulation: place
pose pose

(d) Experiment: initial (e) Experiment: pick pose (f) Experiment: place
pose pose
Fig. 15 Simulated and experimental results for simple motion (supplementary video 1),
where d∗ = (S1 , S2 , S3 , S4 , P1 , P2 , P3 , P4 , P5 ) = (20,0,0,47, (0,40,25,0,0, −90),
(0,60,25,0,0,45), (0,60,25,0,0,45), (75,30, −200,0,0,0), (75,30, −200,0,0,0))
 (a) Simulation: (b) Simulation: (c) (d) (e) Simulation:
initial pose left pose Simulation: Simulation: right pose
top left pose top right
pose

(d) (e) Experiment: (f) (g) (h) Experiment:

Experiment: left pose Experiment: Experiment: right pose
initial pose top left pose top right
pose
Fig. 16 Simulated and experimental results for complex motion (supplementary video
2), where d∗ = (S1 , S2 , S3 , S4 , P1 , P2 , P3 , P4 , P5 ) = (53,13,37,26,
(0, −40,25,0,0, −90), (0, −60,25,0,0, −90), (50, −60,25,0,0, −45),
(30, −50,0,0,0,0), (−65,0,120,0,0,0))

7. Conclusion

This paper discussed a method for optimizing the cable path for a given robot motion.
Here, inspired by the observation of experts' cable path design works, we proposed the
installation of multiple guides on one link to avoid cable collisions with peripheral
devices and reduce bending cable fatigue. In addition, the fast connected geometry
computation using the Hermite interpolation and RMF to reduce the burden and
dictionary of the optimal PVs for intra-link cables were discussed.
The simulations compared the previous and fast connected geometry
computations and showed that the latter is superior from the viewpoint of computational
burden. In addition, pick-and-place motion and inspection motion were referred to as
examples of simple and complex motion, respectively. The optimal solutions and
computational times of the previous and proposed methods for these motions were
compared. The obtained results show that the proposed method can derive solutions not
only for simple motions with a smaller computational burden but also for complex
motions for which no solution can be derived by previous methods. Experiments
confirmed that the cable paths obtained in the simulation could be applied to the actual
robot while satisfying stress constraints. In the case of complex motions for which an
optimal solution cannot be obtained, it is necessary to optimize the robot motion and the
cable path simultaneously, but this is an issue for the future.

Appendix
A. PV candidate set 𝑫𝟎,𝒎

𝐷0,𝑚 [16] is given by

𝐷0,𝑚 = {(𝑆𝑚,𝑖𝑚 , 𝑃𝑚,𝑗𝑚 , 𝑃𝑚 + 1,𝑗𝑚 + 1 )|1 ≤ 𝑖𝑚 ≤ 𝐼𝑚 , 1 ≤ 𝑗𝑚 ≤ 𝐽𝑚 ,

(A-1)
1 ≤ 𝑗𝑚 + 1 ≤ 𝐽𝑚 + 1 , 1 ≤ 𝑚 < 𝑀},

where {𝑆𝑚,𝑖𝑚 | 1 ≤ 𝑖𝑚 ≤ 𝐼𝑚 } denotes a candidate set of sub-segment numbers in 𝐶𝑚 that

connects 𝐺𝑚 and 𝐺𝑚+1 . 𝐼𝑚 denotes the candidate number of sub-segment numbers.
{ 𝑆𝑚,𝑖𝑚 } are prepared as an arithmetic series of constant interval ∆𝑆 . Furthermore,
{𝑃𝑚,𝑗𝑚 | 1 ≤ 𝑗𝑚 ≤ 𝐽𝑚 } denotes a candidate set of configurations of 𝐺𝑚 and 𝐽𝑚 represents
the candidate number of configurations of 𝐺𝑚 . { 𝑃𝑚 + 1,𝑗𝑚 + 1 | 1 ≤ 𝑗𝑚 + 1 ≤ 𝐽𝑚 + 1 }
indicates a candidate set of configurations of 𝐺𝑚 + 1 . 𝐽𝑚+1 represents the candidate
number of configurations of 𝐺𝑚+1 .

B. PV candidate set 𝑫𝟏,𝒎

𝐷1,𝑚 satisfying stress constraints in the attachment test is given by

 𝐷1,𝑚 = {(𝑆𝑚,𝑖𝑚 , 𝑃𝑚,𝑗𝑚 , 𝑃𝑚 + 1,𝑗𝑚 + 1 ) ∈ 𝐷0,𝑚 |

𝐼𝑚𝑝𝑢𝑙𝑠𝑒𝑡ℎ ≥ 𝐼𝑚𝑝𝑢𝑙𝑠𝑒(𝑆𝑚,𝑖𝑚 ,𝑃𝑚,𝑗𝑚 ,𝑃𝑚 + 1,𝑗𝑚 + 1 ) (𝜃),

𝑆𝑡𝑟𝑒𝑡𝑐ℎ𝑡ℎ ≥ 𝑆𝑡𝑟𝑒𝑡𝑐ℎ(𝑆𝑚,𝑖 (𝜃), (B-1)

𝑚 ,𝑃𝑚,𝑗𝑚 ,𝑃𝑚 + 1,𝑗𝑚 + 1 )

𝐶𝑢𝑟𝑣𝑎𝑡𝑢𝑟𝑒𝑡ℎ ≤ 𝐶𝑢𝑟𝑣𝑎𝑡𝑢𝑟𝑒(𝑆𝑚,𝑖 (𝜃),

𝑚 ,𝑃𝑚,𝑗𝑚 ,𝑃𝑚 + 1,𝑗𝑚 + 1 )

𝑓𝑜𝑟 ∀𝜃 ∈ Θ′𝑚 },

where 𝐼𝑚𝑝𝑢𝑙𝑠𝑒(𝑆𝑚,𝑖𝑚 ,𝑃𝑚,𝑗𝑚 ,𝑃𝑚 + 1,𝑗𝑚 + 1 ) (𝜃) represents the maximum impulse received by

each sub-segment of 𝐶𝑚 , when the joint angles are set to Θ′𝑚 that contains only the initial
pose as one of the arbitrary poses. Similarly, 𝑆𝑡𝑟𝑒𝑡𝑐ℎ(𝑆𝑚,𝑖𝑚 ,𝑃𝑚,𝑗𝑚 ,𝑃𝑚 + 1,𝑗𝑚 + 1 ) (𝜃) and

𝐶𝑢𝑟𝑣𝑎𝑡𝑢𝑟𝑒(𝑆𝑚,𝑖𝑚 ,𝑃𝑚,𝑗𝑚 ,𝑃𝑚 + 1,𝑗𝑚 + 1 ) (𝜃) are the maximum stretch and minimum curvature

radius of the sub-segments, respectively.

C. PV candidate set 𝑫′𝟎,𝒎

𝐷′0,𝑚 satisfying the adjacency condition between 𝐷0,𝑚 and 𝐷2,𝑚−1 in the attachment test [16]
is given by

𝐷′0,𝑚 = {(𝑆𝑚,𝑖𝑚 , 𝑃𝑚,𝑗𝑚 , 𝑃𝑚 + 1,𝑗𝑚 + 1 )|

(𝑆𝑚−1,𝑖𝑚−1 , 𝑃𝑚−1,𝑗𝑚−1 , 𝑃𝑚,𝑗𝑚 ) ∈ 𝐷2,𝑚−1 , (C-1)

(𝑆𝑚,𝑖𝑚 , 𝑃𝑚,𝑗𝑚 , 𝑃𝑚 + 1,𝑗𝑚 + 1 ) ∈ 𝐷0,𝑚 },

Compared to 𝐷0,𝑚 , 𝐷′0,𝑚 includes fewer PVs. Thus, the burden can be reduced by
applying the adjacency condition in CSAT [16].
Statements & Declarations

Funding

The authors declare that no funds, grants, or other support were received during the preparation
of this manuscript.

Competing Interests

The authors have no relevant financial or non-financial interests to disclose.

References

[1] International Federation of Robotics (2021) Executive summary world robotics 2021

industrial robots.

https://ifr.org/img/worldrobotics/Executive_Summary_WR_Industrial_Robots_2021.pd

f. Accessed 01 April 2023.

[2] Antman SS (1995) Nonlinear problems of elasticity. Springer.

[3] Spillmann J, Teschner M (2007) CoRdE: Cosserat rod elements for the dynamic

simulation of one-dimensional elastic objects. In: Proceedings of the 2007 ACM

SIGGRAPH/Eurographics symposium on Computer animation, pp 63–72.

[4] Hergenröther E, Dähne P (2000) Real-time virtual cables based on kinematic

simulation. In: Proc WSCG, pp 402–409.

[5] Grégoire M, Schömer E (2007) Interactive simulation of one-dimensional flexible

parts. Computer-Aided Design 39:694–707. https://doi.org/10.1016/j.cad.2007.05.005

[6] Loock A, Schömer E (2001) A virtual environment for interactive assembly

simulation: From rigid bodies to deformable cables. In: 5th World Multiconf Syst
Cybern Informatics (SCI ’01) 2001, pp 325–332.

[7] Lv N, Liu J, Ding X, Liu J, Lin H, Ma J (2017) Physically based real-time

interactive assembly simulation of cable harness. J Manuf Syst 43:385–399.

https://doi.org/10.1016/j.jmsy.2017.02.001

[8] Müller M, Heidelberger B, Hennix M, Ratcliff J (2007) Position based dynamics. J

Vis Commun Image Represent 18:109–118. https://doi.org/10.1016/j.jvcir.2007.01.005.

[9] Bullet Real-Time Physics Simulation. https://pybullet.org/wordpress/. Accessed 1

May 2023.

[10] Havok physics. https://www.havok.com/havok-physics/. Accessed 1 May 2023.

[11] Phys X. SDK. https://developer.nvidia.com/physx-sdk. Accessed 1 May 2023.

[12] Kressin J (2013) Path optimization for multi-robot station minimizing dresspack

wear. Master’s thesis, Chalmers University of Technology, Department of Signals and

Systems.

[13] Carlson JS, Kressin J, Hermansson T, Bohlin R, Sundbäck M, Hansson H (2016)

Robot station optimization for minimizing dress pack problems. Procedia CIRP 44:389–

394. https://doi.org/10.1016/j.procir.2016.01.022

[14] Hermansson T, Carlson JS, Linn J, Kressin J (2021) Quasi-static path optimization

for industrial robots with dress packs. Robot Comput Integr Manuf 68:102055.

https://doi.org/10.1016/j.rcim.2020.102055

[15] Iwamura S, Mizukami Y, Endo T, Matsuno F (2022) Cable-path optimization

method for industrial robot arms. Robot Comput Integr Manuf 73:102245.
https://doi.org/10.1016/j.rcim.2021.10224

[16] Iwamura S, Mizukami Y, Endo T, Matsuno F (2022) Efficient cable path

optimization based on critical robot poses for industrial robot arms . IEEE Access

10:66028–66044. https://doi.org/10.1109/ACCESS.2022.3184421

[17] Loris R, Palaniappan V, Marco M, Giuseppe B, Arash A, Dario P (2023) A human-

centric framework for robotic task learning and optimization, J Manuf Syst 67:68–79.

https://doi.org/10.1016/j.jmsy.2023.01.003

[18] Roveda L, Maggioni B, Marescotti E, Shahid AA, Zanchettin AM, Bemporad A,

Piga D (2021) Pairwise preferences-based optimization of a path-based velocity planner

in robotic sealing tasks. In: 2021 IEEE/RSJ International Conference on Intelligent

Robots and Systems (IROS), Prague, Czech Republic, pp 1674–1681.

https://doi.org/10.1109/IROS51168.2021.9636184.

[19] Mon AA, Aye KK (1995) Hermite interpolation by Pythagorean hodograph curve.

World Acad Sci Eng Technol 38:35–39.

[20] Wang W, Jüttler B, Zheng D, Liu Y (2008) Computation of rotation minimizing

frames. ACM Trans Graph 27:1–18. https://doi.org/10.1145/1330511.1330513

[21] Hall CA, Meyer WW (1976) Optimal error bounds for cubic spline interpolation. J

Approx Theor 16:105–122. https://doi.org/10.1016/0021-9045(76)90040-X

[22] Mortenson ME (1999) Chapter 15. Mathematics for computer graphics

applications. In: The Bezier curve. Industrial Press Inc., pp 264–275.

[23] Bloomenthal J (1990) Calculation of reference frames along a space curve. Graph

Gems 1:567–571. https://doi.org/10.1016/B978-0-08-050753-8.50124-8

[24] Yuan B, Du H, Wang H, Xiong W (2015) The simulation of cable harness based on

mass-spring model. MATEC Web of Conferences 31:10002.

https://doi.org/10.1051/matecconf/20153110002

[25] Dedieu JP (2015) Newton-Raphson method. In: Engquist B (ed) Encyclopedia of

applied and computational mathematics. Springer, Berlin, Heidelberg, pp 1023–1028.

https://doi.org/10.1007/978-3-540-70529-1_374.

[26] Omron. Viper.

http://www.ia.omron.com/products/family/3520/download/catalog.html. Accessed 1

May 2023.

[27] SMC LEHF32EK2-32. http://ca01.smcworld.com/catalog/Electric/mpv/Electric-

LEH/data/Electric-LEH.pdf. Accessed 1 May 2023.

[28] SMC LE-CP-3. http://ca01.smcworld.com/catalog/Electric/mpv/Electric-LEC-

JXC/data/Electric-LEC-JXC.pdf. Accessed 1 May 2023.

[29] Basler Ace Series. https://www.baslerweb.com/en/products/cameras/area-scan-

cameras/ace/. Accessed 1 May 2023.

[30] Basler cable GigE. https://www.baslerweb.com/en/products/cable/basler-cable-

gige-cat-6-rj45-sl-hor-rj45-drc-p-5-m/. Accessed 1 May 2023.

[31] Omron. Sysmac studio.

https://automation.omron.com/en/ca/products/family/sysstdio. Accessed 1 May 2023.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.

SupplementalVideo1.mp4
SupplementalVideo2.mp4