Intelligent Path Panning Towards Collision-free Cooperating

Industrial Robots

L. Larsen

, J. Kim

and M. Kupke

German Aerospace Center (DLR), Center for Lightweight Production Technology (ZLP),

Am Technologiezentrum 4, Augsburg, Germany

University of Augsburg, Human Centered Multimedia, Universitaetsstr. 6a, Augsburg, Germany

1 INTRODUCTION

Due to raising costs of energy and the increasing

request for environment-friendly products, there is

a huge demand for lighter vehicles, helicopters or

airplanes made of carbon ﬁbre reinforced plastics

(CFRP). In the last years the structural weight in per-

centage of CFRPs in airplane production was clearly

rising. The AIRBUS A380 with about 20 % and

the AIRBUS A350 with about 50 % (see Figure1).

Carbon ﬁbres are very thin and have a high tensile

strength. Products which are assembled from CFRP

offer one of the highest strength-to-weight ratio com-

pared to any other material and also offer superior

thermal and conductive properties. These properties

makes the material very famous in car, aerospace, ma-

rine and sports industry.

To use CFRPs in a mass production it is impor-

tant to produce lightweight products cost-efﬁciently

in high quality and quantity. To achieve these goals,

a high degree of automation is necessary. For this

reason, the German Aerospace Center (DLR) devel-

oped a robotic portal system within the Center for

Lightweight-Production-Technology (ZLP). In this

system, large parts of CFRP for e.g. the aerospace

industry and the wind energy industry can be manu-

factured automatically. It supports three portal robots

and two industrial robots on a linear track, enabling

them to move along the middle beam. Its dimensions

are about 30 m in x-, 15 m in y- and 7 m in z-direction,

see Figure 2. All robots and portals in the robotic cell

share the same workspace so they can work together

which makes the programming one the one side very

difﬁcult and one the other side very interesting. To

achieve the goal of low-cost production, the DLR con-

centrates on dry ﬁbre placement of large carbon fab-

rics. This ensures higher lay up rates than tape laying

which reduces manufacturing costs of large compos-

ite structures. To handle these large and sensitive car-

bon ﬁbre fabrics, it is necessary to work with coop-

erating robots, which requires a sophisticated control

Figure 1: Rising overall structural CFRP weight in airplane

production (source: AIRBUS).

Figure 2: Multi-variant-production cell (MFZ) containing

two industrial robots on a linear track and three portal robots

in the same working space. The dimensions of the cell are

30 m in x-, 15 m in y- and 7 m in z-direction.

strategy.

2 RESEARCH PROBLEM

In conventional industrial robotic cells e.g. in an au-

tomotive assembly line the robots execute the same

task for a period of many month or years. The move-

ments of the robots are hard-coded for one process

with respect to no occurring collision. Normally a

high amount of time is needed to program the robots.

Larsen L., Kim J. and Kupke M..

Intelligent Path Panning Towards Collision-free Cooperating Industrial Robots.

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

A reconﬁguration of the cells implies a high resource

effort.

In the automotive industry the cycle time is mea-

sured in seconds and in the aerospace industry it is

measured in hours. One reason for that is that the as-

semblies e.g. a fuselage or a wing are much bigger.

Another reason is that the requirements on the accu-

racy are very high. Bringing to mind that transporting

a dry ﬁbre cut piece is like transporting a table dish it

becomes clear that the process speed is not the only

challenge rather it is the quality of the process and the

resulting product.

Like it is easier to set a table with two persons it

is also easier to transport a large four meter dry ﬁbre

cut piece with two robots. Figure 3 shows an exam-

ple process for the layup of carbon ﬁbre cut pieces

with cooperating robots. First the cut piece is grabbed

from a plane table. For lifting the cut pieces the robots

have special designed end-effector which can handle

the material very gentle. The material is very perme-

able to air. Therefore the end-effectors don’t work

with vacuum to grab the textile but with a high ﬂow

rate like a vacuum cleaner. After grabbing the ma-

terial it is transported on a linear axis to a negative

tooling of the assembly. In this case it is the lower

half of a fuselage with the dimension of an AIRBUS

A320. In the mould the textile is placed and ﬁxated.

The position were the cut pieces must be placed

in the 3D mould is deﬁned by the designer in an com-

puter aided design (CAD) tool like CATIA Compos-

ites Design (CPD). With that tool different carbon ﬁ-

bre cut pieces with different ﬁbre orientations are put

on each other. Depending on the demands of the as-

sembly this can be ﬁve up to more than 20 layers with

thousands of cut pieces for a whole fuselage. As a re-

sult of the design phase the designer exports the exact

lay down position of every cut piece. This lay up deﬁ-

nition is called plybook. Figure 4 shows the ﬁrst layer

from a plybook for the example process in 3D and un-

wind to 2D. The coloured lines show the outlines of

the cut pieces. The gripping position also comes from

the CAD software.

3 STATE OF THE ART

This chapter shows the state of the art of collision de-

tection and path planning both in science and in in-

dustry.

Figure 3: A typical dry ﬁbre placement process for the pro-

duction of the lower half of an airplane fuselage with the

size of an AIRBUS A320 a.) grabbing a 4-6 m cut piece

from a ﬂat plane b.) transport from grabbing position to lay

up position c.) lay up of cut piece in 3 dimensional mould.

Figure 4: First layer of the plybook of the lower half of a

fuselage in 3D and unwind to 2D.

ICINCO2014-DoctoralConsortium

3.1 State of the Art in Science

3.1.1 Collision Detection

Industrial manipulators have high masses which are

moved. That makes it very important to detect colli-

sions of the robot with other robots or the environment

reliable.

Most collision detection algorithms distinguish

between a broad- and a near-phase (Kockara et al.,

2007). In the broad-phase objects are detected which

are inspected more accurately in the near-phase.

The most common methods to check the broad-

phase are exhaustive search, sweep and prune (Baraff,

2001) (Lin and Manocha, 1993) and hierarchical hash

tables (Mirtich, 1996). In industrial robot applications

the position of the robot is known at every time with

an accuracy of about ±3 mm. This makes the detec-

tion of the broad-phase very easy.

The algorithms to calculate the near-phase are

mostly divided into four groups: feature-based,

simplex-based, image-space and Spatial Data Struc-

tures (Mirtich, 1996). Feature-based algorithms di-

rectly work on the primitive form of the object.

Famous representatives are polygonal intersection

(Moore and Wilhelms, 1988), Lin-Canny (Lin and

Canny, 1991), V-Clip (Mirtich, 1998) and SWIFT

(Ehmann and Lin, 2000).

Simplex-based algorithms are based on a geo-

metrical approach which describes a n-dimensional

polygon which is the convex hull of n+1 vertices.

With increasing dimensionality this results in the fol-

lowing objects: point, line, triangle, triangular pyra-

mid. The Gilbert-Johnson-Keerthi (GJK) distance al-

gorithm is the most famous representative (Gilbert,

1988) (Cameron, 1997) (van den Bergen, 1999)

(van den Bergen, 2001) (Gilbert and Foo, 1990).

The family of Image-Space algorithms (ISB) de-

tects collisions by the calculation of overlappings.

For the detection of collisions in dynamic environ-

ments the ISBs are very efﬁcient, because the cal-

culation can be parallelized very easy and put on an

graphical processing unit (GPU). Famous examples

are (Jang, 2006) (Stewart, 2008) (Heidelberger, 2004)

(Jang et al., 2007) (Myszkowski et al., 1995) (Heidel-

berger et al., 2003).

Bounding Volume Hierarchies (BVH) belong to

the family of Spatial Data Structures. In (Figueiredo

et al., 2006) an overview over the different types can

be found. One kind are Bounding Spheres (Bradshaw

and O’Sullivan, 2004). The use of Bounding Spheres

is very famous for collision detection because the cal-

culation cost are very simple and fast. Another group

are the Bounding Boxes (BB) which describe objects

better than spheres. They are often used in applica-

tions like ray-tracing. Special forms of the bounding

boxes are Axis-Aligned Bounding Boxes (AABB),

which are aligned on the coordinate axes. They are

normally deﬁned by two points which deﬁne the ver-

tices on the cross diagonal. Another form are arbitrar-

ily oriented boxes called Oriented Bounding Boxes

(OBB). Furthermore objects can be deﬁned by poly-

hedrons. These bounding boxes are called k-Discrete

Oriented Polytopes (k-DOP). In contrast to OBB, k-

DOPs allow more bounding surfaces wherefore they

can describe objects better.

Due to the fact that modern measuring systems,

like a laser-scanner or a depth camera often create

point clouds as representation of the environment an-

other possibility of collision detection is the direct

calculation on these clouds (Klein and Zachmann,

2004) (Pan et al., 2011) (Yakut, 2010).

In the industrial environment there are just a few

approaches which deal with collision avoidance based

on environment data without external sensors. In

(Fawaz et al., 2009) a virtual simulator is introduced

which allows an on-line collision monitoring. (Pe-

drocchi et al., 2009) picks an algorithm based on po-

tential ﬁeld method as central theme which can de-

tect and avoid collisions between industrial robots.

In (Cheng, 1995) a four dimensional (three spa-

tial dimensions and time) real-time collision detec-

tion technique for the UPS arm, a ten degree-of-

freedom hybrid serial-and-parallel redundant robotic

arm is shown. For the collision detection the geo-

metrical information is separated from the time by s-

formulations. After the separation the occurring col-

lisions can be calculated by bisection method in re-

altime. In (Hermann, 2013) a real-time collision de-

tection system, which is optimized for the calculation

on CUDA Graphical Processing Unit (GPU), is pre-

sented. The system is based on two voxelmaps which

are put on the GPU.

3.1.2 Path Planning of Industrial Robots

The main goal of motion or path-planning is to

ﬁnd a continuous motion that connects a starting-

and destination-point, while avoiding collisions with

known obstacles. For the calculation the robot and the

obstacles must be deﬁned in a 2D or 3D workspace.

Normally the coordinates that deﬁne the position and

the orientation of a coordinate frame that is attached

to a rigid body in three-dimensional space deﬁne its

conﬁguration space. In robotics the conﬁguration

space deﬁnes all positions of an end-effector which is

attached to a robot in three-dimensional space (Craig,

2005).

Another space which describes a position of a

IntelligentPathPanningTowardsCollision-freeCooperatingIndustrialRobots

robot is the joint space. The set of joint positions for

each link of the robot is called joint space. The for-

ward and backward kinematic of a robot manipulator

is the mapping between the conﬁguration space and

the joint space. The forward kinematic maps the joint

positions to the coordinate positions and the backward

from coordinate to joint positions. Finding a path for

a robot also means to ﬁnd the path in the conﬁguration

space and afterwards map it back to joint space.

For the determination of two- or tree-dimensional

path planning problems exist different grid based

methods, where a grid is put over the working space

of the robot. Famous examples are depth-ﬁrst search,

breadth-ﬁrst search, Dijkstra and A*. A very good

explanation of these algorithms can be found here

(LaValle, 2006).

Another group are the geometrical algorithms, in-

cluding Visibility Graphs (Scott and Vuillemin, 1986)

and Cell Decomposition (LaValle, 2006). A Visibil-

ity Graph reﬂects the free view between different ver-

tices in a scene. By adding a start and an endpoint

the shortest path can be found with the Dijkstra algo-

rithm. The Cell Decomposition algorithm divides the

scene in smaller regions (cells). After that the short-

est and easiest connection between different cells is

calculated with a tree data structure.

Path planning for robotic manipulators in scien-

tiﬁc of industrial applications is very interesting and

challenging problem especially if the environment of

the robot is not static e.g. when there is human-

machine interaction or multiple robots are in the same

workspace.

For high dimensional path detection problems

the Potential Field Method is very practical. The

basic concept of that method is that obstacles and the

robot are seen as electriﬁed particles with opposite

sign and the goal has the same sign as the robot.

The algorithm calculates a path where the distance

between the robot and the obstacles is big enough

and the length between the start- and the endpoint is

as small as possible (Barraquand et al., 1992) (Daily

and Bevly, 2008) (Kitamura et al., 1995) (Tang et al.,

2010).

Single Robot Path Planning. There are some exam-

ples for the path planning of a single robot manip-

ulator both off-line and on-line using computational

intelligence methods.

In (Saravanan et al., 2007) an off-line al-

gorithm based on an evolutionary algorithm is

presented which calculates an optimal trajectory

for a PUMA560 6-DOF manipulator. The aim of

the algorithm is to minimize the multi-criterion

cost function with actuator constraints, joint limits

and payload constraints by considering dynamic

constraints by motion. In (Ting et al., 2002) a

collision free off-line path planning algorithm is

introduced which is based on the assigned marked

number of the passable region via wave expansion

method. In (Klanke et al., 2006) an on-line path

planner for a redundant Mitsubishi PA-10 arm with

7-DOF is introduced which can deal with stationary,

non stationary or unknown environment. The method

works with the grid based dynamic wave expansions

neural network (Lebedev et al., 2005). In (Huang

and Lian, 1997) a model-free hybrid fuzzy logic and

neural network algorithm was proposed to control a

4-DOF manipulator. A conventional fuzzy controller

was used for the rough adjustment of each joint.

Another controller which used a back-propagation

(BP) neural network was designed to control the

coupling between the links. By the combination of

fuzzy and neural network the learning time of the

neural network could be dramatically reduced. In

(Zavlangas and Tzafestas, 2000) a fuzzy approach

is presented for the on-line local navigation and

obstacle avoidance for an industrial 3-DOF robotic

manipulator. The system is divided into separate

fuzzy units each of them controlling a robotic joint

separately. In (Cueva and Ramos, 2002) a method

based on genetic algorithm is presented to calculate

collision free paths in 2D for redundant or non redun-

dant manipulators. In (Kazem et al., 2008) a genetic

algorithm is proposed to optimize the point-to-point

trajectory of a redundant 3-link arm. The algorithm

can ﬁnd a collision free path with minimum travelling

time and space. In (Gosselinj, 1994) an approach

is presented using neural network and fuzzy logic

for the path planning of a 3-link manipulator in 2D.

The neural network is used to predict in real-time the

trajectory of a moving object ﬁlmed by a camera to

be caught by the manipulator. The fuzzy logic is used

to control the joints of the robot.

Cooperating Robots Path Planning. Path planning

for cooperating mobile robots is a popular research

area however in the area of cooperating industrial ma-

nipulators are just a few research results.

In (Juan C. Fraile and Dodds, 1999) a trajectory

planning system is introduced which calculates tra-

jectories with a minimum time performance for three

industrial manipulators each with ﬁve joints sharing

the same workspace and working on the same work-

piece. (Tzafestas et al., 1998) present a path planning

system, for a cooperating three-robot system transfer-

ring a large object from a start to a ﬁnal position. The

method is based on the master-and-two-slave mode

where one robot is the master pretending to do the

movements and the other robots are following. (Ali

et al., 2002) present a path planning algorithm using

ICINCO2014-DoctoralConsortium

coevolutionary genetic algorithm (CGA) for two co-

operative robots. A coevolutionary genetic search is

performed on the conﬁguration space to ﬁnd the mini-

mum and collision free path. Finally the CGA method

is compared against A* and genetic algorithms (GA).

In (Curkovic et al., 2013) a coevolutionary algorithm

for the collision free motion planning of two 6-DOF

industrial manipulators with overlapping workspaces

is presented. The planning is based on a hall of fame -

Pareto-based co-evolutionary algorithm which allows

the real-time calculation of the path. (Garg and Ku-

mar, 2002) present a strategy for the determination of

an optimal path for multiple 2-link robots in 2D re-

quiring the least amount of torque with genetic algo-

rithm. (Li et al., 2012) introduce a system using neu-

ral network for the control of multiple redundant ma-

nipulators moving the same payload together. Each

module in the neural network controls one manipula-

tor and all networks together solve the common task.

3.2 State of the Art in Industry

3.2.1 On-line Programming

A common on-line programming technology is the

manual teach-in of a robot. This means the process of

generating a sequence of robot poses or instructions

the robot has to do to accomplish the desired task.

To do that the programmer has to know the program-

ming language and the teach panel of the robot he

uses. Normally the teach panel has buttons for saving

and editing the program and buttons or a 3D mouse to

move the robot. The programmer can move the robot

in different coordinate systems which makes the po-

sitioning more intuitive. The robot offers a Cartesian

world coordinate system and a tool center point (TCP)

Cartesian coordinate system. Furthermore the robot

offers the possibility to deﬁne a base coordinate sys-

tem for a work piece. Another possibility is to move

the joints of the robot. Between the point the pro-

grammer can deﬁne in which motion proﬁle e.g. lin-

ear or point-to-point and with which acceleration and

velocity the robot should move.

The advantage of the teach-in process is that it

is very clear for the user because he has a concrete

reference to the process while programming. A task

in a working space without many obstacles is easy to

program. If the working space is very complex with

many obstacles and a multiple bent work pieces where

many interpolation points are needed the teach-in is

very difﬁcult. Another disadvantage of this method is

that it is a very time consuming process. After a raw

teach-in follows another very time consuming opti-

mization of the trajectory of the robot by optimizing

single points and adapting the velocity proﬁle for the

movements. To do that the programmer has to run the

created program on-line on the robotic cell. The result

of the teach-in is one solution for the trajectory of the

process which depends on the skills and experience of

the programmer.

For the programming of cooperating robots the

robot manufacturer KUKA offers an application

package called RoboTeam which makes it possible

that up to 15 robots can work in a team. RoboTeam

makes it possible that robots can do load sharing or

workpieces can be processed during a transfer from

different robots. The robots keep their standard con-

troller and are connected and synchronized with a lo-

cal network. The application package offers two func-

tionality types. The motion cooperation and the pro-

gram cooperation.

In this way, all tasks that directly affect the robot

group are also carried out autonomously by the group.

By setting shared synchronization markers, it is pos-

sible to synchronize the program sequences of several

robots e.g. the synchronized grabbing of a workpiece.

The distributed sequence control of complete manu-

facturing programs is carried out decentralised within

the networked robot group. Each robot in the group

can start a manufacturing program on another robot

or wait for the end of a manufacturing program. This

means that it is possible to dispense with an external

PLC in many cases, leading to signiﬁcant cost reduc-

tions for the production cell. The program operation

allows the synchronization and monitoring of shared

workspaces of the robots.

The motion cooperation allows to geometrically

connect the TCPs of the robots. As a result it is pos-

sible to couple the geometric path of the robots. This

allows a ﬂexible solution for all processes where very

heavy or very large workpieces have to be transported

by more than one robot.

The teach-in of cooperation robots has additional

disadvantages adverse to the teach-in of one robot.

One is that it is very difﬁcult to have an overview off

the teaching process. To avoid collisions while the

teach process there should be minimum one more per-

son. Another disadvantage is that the programming

of two robots is not very user friendly. The program-

mer has to switch between the teach panels of the two

robots all the time.

3.2.2 Off-Line Programming (OLP)

Off-line programming is a simulation based program-

ming of a robot. The robot cell is modelled in the

computer. This can be done by containing obstacles,

tools, jigs and the robots. For the modelling the geom-

etry and the size of all things which are in the cell and

IntelligentPathPanningTowardsCollision-freeCooperatingIndustrialRobots

necessary for the process must be exactly known. The

quality of the OLP program signiﬁcantly rises with

the quality of the model. After the cell is modelled a

collision free path can be manually generated by mov-

ing the robot in the virtual world. Afterwards the path

can be optimized for a fast execution. After the op-

timization the path must be exported to a real robot

program in the appropriate robot language.

The OLP programming has a lot of advantages. It

is possible to test different scenarios of the process be-

fore the cell is build. After that an optimization of the

installation of the process stations can be done before

the cell is built. Another advantage is that the cell can

be taken into operating state directly after completion.

But OLP has also some disadvantages which are

inappropriate for the use in CFRP production. The

biggest drawback is that cooperating robots are not

supported. Another one is that the resulting program

is static and it is not possible to put correction values

to the production process. So the whole quality of the

resulting assembly mainly depends on the accuracy of

the virtual model of the cell in the OLP tool.

4 METHODOLOGY

The goal of this thesis is to accurately examine fuzzy

logic, artiﬁcial neural network and evolutionary com-

puting methods on the applicability for path planning

for cooperating industrial manipulators in a CFRP

production environment. First each of these methods

will be examined stand-alone without optimization.

In that stage the advantages and disadvantages of ev-

ery method will be identiﬁed and compared.

After the analysis the strength of the methods will

be combined for the path planning of cooperating

robots. The process which will be considered is the

pick, transport, drape and lay up of dry ﬁbre carbon

material. While the path planning it will always be

important to focus on the gentle handling of the ma-

terial. The raw material is very sensitive and its char-

acteristic are a little bit similar to cloth. In a later

step more robots could be involved in the path plan-

ning. Two robots could be responsible for the lay up

and a third robot could do a quality check e.g. with a

camera or a laser in the same working space. In con-

ventional robotic cells the other robots would stop so

that the measuring robot can do his job. To use the

robotic cell shown in Figure 2 economically well the

processes should not be mutual exclusive but rather

different jobs should run parallel.

Additionally the fuzzy logic, artiﬁcial neural net-

work and evolutionary computing methods will be

compared with classical path planning methods like

e.g. Dijkstra, A* and potential ﬁeld method. A spe-

cial focus will be set on the reaction of the different

methods on local minima. The potential ﬁeld method

e.g. is very famous for getting stuck in local minima

(Stroulia et al., 1997).

Another point which will be examined is the com-

plexity of calculating a path for a scenario. Normally

exact motion planning with complex constraints in a

high-dimensional system is computationally very ex-

pensive. According to (Eberhart and Shi, 2007) com-

putational intelligence methods need very less com-

putation power and nevertheless it is possible to solve

problems which are otherwise impossible or imprac-

tical. According to the author computationally in-

telligence tools do offer solutions to some problems

which are not able to solve with any other method.

A very important point on which will be focussed

while the examination of the path planning is the col-

lision detection. Typical path planning application

have a collisions checker which tries to avoid colli-

sions between the robot and static obstacles in the

path. The path planning in this thesis should also

consider possible collisions between different robots

which are moving objects. Another important consid-

eration in that thesis are intended collisions between

the robot tool and the jigs which are necessary for the

lay up process. The robot tools have silicon vacuum

cups which are needed to press the material on the

jigs. These volitional collisions should be considered

by the path planning.

All test for this thesis should ﬁrstly be simulated.

After successful testing the methods should be tested

in a real production scenario in an industrial scale in

the multifunctional robotic cell in Figure 2. For the

testing of the algorithms a simulation environment

has been implemented in C#, which allows the visu-

alisation of 6-DOF industrial robots. The environ-

ment is called CoCo-Framework. CoCo stands for

Collisions-free Cooperation which reﬂects the main

aim of that thesis. In these framework it is possible

to import obstacles for the scene from CAD data or

from external sensors like depth cameras. It also al-

lows to attach end-effectors to the robots which are

used for the production process. For the movement

of the robots a forward- and inverse kinematic are

implemented. Furthermore typical motion proﬁles

like linear, point-to-point and circ movement are im-

plemented. While moving the robots in the simula-

tion environment occurring collisions can be detected

by different bounding box based algorithms like e.g

GilbertJohnsonKeerthi distance algorithm (GJK).

Figure 5 shows a screenshot of the visualisation of

the CoCo environment. The scene shows a dry ﬁbre

placement process for cooperating robots. The robots

ICINCO2014-DoctoralConsortium

are equipped with special end-effectors to handle dry

ﬁbre textiles. The half cylinder shows the lower half

of a fuselage. The colourful outlines show the plac-

ing position for the ﬁbre cut pieces. Figure 6 shows

the path planning for a single robot process in the

CoCo simulation framework. On the left side an in-

dustrial robot can be seen. Each link is surrounded

by a bounding box for collision checking. The red

box in the middle represents an obstacle in the pro-

cess. The grey half cylinder represent the destination.

In this scene the shortest TCP path is calculated with

cell decomposition.

Figure 5: Cooperating robots in CoCo simulation frame-

work putting a dry ﬁbre cut piece in the lower half of a

fuselage.

Figure 6: Path planning for a single robot with the CoCo

simulation framework. The red box represents an obsta-

cle. The blue lines show the cell decomposition of the

workspace and the red line shows one possible path of the

TCP from the actual position of the robot to the lay up po-

sition.

5 EXPECTED OUTCOME

For a CFRP process the state-of-the art cooperating

teaching methods are not suitable. There is the need

for an economic possibility to generate robot pro-

grams to produce high quality CFRP parts. To achieve

that cooperating robots should work more material

based. The normal master-slave control mode which

is state of the art is not satisfactory. The objective

of that thesis is to develop a system which allows

the collisions free path planning for cooperating in-

dustrial robots in the CFRP airplane production. For

the system come into operation methods of computa-

tional intelligence to which belong fuzzy-logic, artiﬁ-

cial neural networks and evolutionary computing.

6 CONCLUSION

In industrial environment where many robots share

the same workspace it is essential to ﬁnd feasible, col-

lision free paths for the robots. As already shown by

other authors computational intelligence methods are

appropriate for the path planning of single robots. For

cooperating robots the collision free path planning is

even more complex. However here computational in-

telligence methods are very promising candidates for

the planning.

REFERENCES

Ali, M. S. A. D., Babu, N. R., Varghese, K., Engg, C., and

Madras, I. I. T. (2002). Ofﬂine path planning of co-

operative manipulators using co-evolutionary genetic

algorithms. ISARC.

Baraff, D. (2001). Physically Based Modeling: Rigid Body

Simulation. Technical report.

Barraquand, J., Langlas, B., and Latombe, J. C. (1992).

Numerical Potential Field Techniques for Robot Path

Planning. 22:224–241.

Bradshaw, G. and O’Sullivan, C. (2004). Adaptive Medial-

Axis Approximation for Sphere-Tree Construction.

ACM Transactions on Graphics, 23:1–26.

Cameron, S. (1997). Enhancing GJK: Computing Min-

imum and Penetration Distances between Convex

Polyhedra. In Proceedings of International Confer-

ence on Robotics and Automation, pages 3112–3117.

Cheng, H. H. (1995). Real-time four-dimensional Collision

Detection for an Industrial Robot Manipulator.

Craig, J. J. (2005). Introduction to Robotics.

Cueva, V. D. and Ramos, F. (2002). Adapting the Messy Ge-

netic Algorithm for Path Planning in Redundant and

Non-redundant Manipulators. (1):21–30.

Curkovic, P., Jerbic, B., and Stipancic, T. (2013). Co-

Evolutionary Algorithm for Motion Planning of Two

Industrial Robots with Overlapping Workspaces. In-

ternational Journal of Advanced Robotic Systems,

page 1.

Daily, R. and Bevly, D. M. (2008). Harmonic Potential

Field Path Planning for High Speed Vehicles. In

American Control Conference, pages 4609–4614.

Eberhart, R. C. and Shi, Y. (2007). Computational Intelli-

gence Concepts to Implementations.

Ehmann, S. A. and Lin, M. C. (2000). SWIFT: Accelerated

Proximity Queries Using Multi-Level Voronoi March-

ing.

IntelligentPathPanningTowardsCollision-freeCooperatingIndustrialRobots

Fawaz, K., Merzouki, R., and Ould-Bouamama, B. (2009).

Model Based Real Time Monitoring for Collision

Detection of an Industrial Robot. Mechatronics,

19(5):695 – 704.

Figueiredo, M., Oliveira, J., Araujo, B., and Pereira, J.

(2006). An Efﬁcient Collision Detection Algorithm

for Point Cloud Models. Technical report.

Garg, D. P. and Kumar, M. (2002). Optimization techniques

applied to multiple manipulators for path planning and

torque minimization. Engineering Applications of Ar-

tiﬁcial Intelligence, 15(3-4):241–252.

Gilbert, E. G. (1988). A Fast Procedure for Computing

the Distance Between Objects in Three-Dimensional

Space. Technical report.

Gilbert, E. G. and Foo, C. P. (1990). Computing the

Distance Between General Convex Objects in Three-

Dimensional Space. IEEE Transactions on Robotics

and Automation, 6(1):53–61.

Gosselinj, C. (1994). Robot Path Planning Using Neural

Networks and Fuzzy Logic. (418).

Heidelberger, B. (2004). Detection of Collisions and Self-

collisions Using Image-space Techniques. Journal of

WSCG, 12(3):145–152.

Heidelberger, B., Teschner, M., and Gross, M. (2003). Vol-

umetric Collision Detection for Deformable Objects.

Technical report.

Hermann, A. (2013). GPU-based real-time collision de-

tection for motion execution in mobile manipulation

planning. Advanced Robotics (ICAR), 2013 16th In-

ternational Conference on.

Huang, S.-j. and Lian, R.-j. (1997). A hybrid fuzzy logic

and neural network algorithm for robot motion con-

trol. IEEE Transactions on Industrial Electronics,

44(3):408–417.

Jang, H.-Y. (2006). GPU-based Image-space Approach to

Collision Detection among Closed Objects.

Jang, H.-Y., Jeong, T., and Han, J. (2007). Image-space

Collision Detection Through Alternate Surface Peel-

ing. In Proceedings of the 3rd international confer-

ence on Advances in visual computing - Volume Part I,

ISVC’07, pages 66–75, Berlin, Heidelberg. Springer-

Verlag.

Juan C. Fraile, M. M. and Dodds, G. I. (1999). Optimiza-

tion of Collision Free Trajectories in Multi-Robot Sys-

tems.

Kazem, B. I., Mahdi, A. I., and Oudah, A. T. (2008). Mo-

tion Planning for a Robot Arm by Using Genetic Al-

gorithm. 2(3):131–136.

Kitamura, Y., Tanaka, T., Kishino, F., and Yachida, M.

(1995). 3-D Path Planning in a Dynamic Environ-

ment Using an Octree and an Artiﬁcial Potential Field.

Technical report.

Klanke, S., Lebedev, D., Haschke, R., Steil, J., and Ritter,

H. (2006). Dynamic Path Planning for a 7-DOF Robot

Arm. 2006 IEEE/RSJ International Conference on In-

telligent Robots and Systems, (i):3879–3884.

Klein, J. and Zachmann, G. (2004). Point Cloud Collision

Detection. In Cani, M.-P. and Slater, M., editors, Com-

puter Graphics forum (Proc. EUROGRAPHICS), vol-

ume 23, pages 567–576, Grenoble, France.

Kockara, S., Halic, T., K.Iqbal, Bayrak, C., and Rowe, R.

(2007). Collision detection: A survey. Technical re-

port.

LaValle, S. M. (2006). Planning Algorithms. Cambridge

University Press, Cambridge.

Lebedev, D. V., Steil, J. J., and Ritter, H. J. (2005). The

dynamic wave expansion neural network model for

robot motion planning in time-varying environments.

49(0):1–50.

Li, S., Chen, S., Liu, B., Li, Y., and Liang, Y. (2012). De-

centralized kinematic control of a class of collabora-

tive redundant manipulators via recurrent neural net-

works. Neurocomputing, 91:1–10.

Lin, M. C. and Canny, J. F. (1991). A Fast Algorithm for

Incremental Distance Calculation. In In IEEE Interna-

tional Conference on Robotics and Automation, pages

1008–1014.

Lin, M. C. and Manocha, D. (1993). Efﬁcient Contact De-

termination Between Geometric Models. Technical

report, International Journal of Computational Geom-

etry And Applications.

Mirtich, B. V. (1996). Impulse-based Dynamic Simulation

of Rigid Body Systems. PhD thesis. AAI9723116.

Mirtich, B. V. (1998). V-Clip: fast and robust polyhedral

collision detection. ACM Trans. Graph., 17(3):177–

208.

Moore, M. and Wilhelms, J. (1988). Collision Detection

and Response for Computer Animation. In Computer

Graphics, pages 289–298.

Myszkowski, K., Okunev, O. G., and Kunii, T. L. (1995).

Fast Collision Detection Between Complex Solids Us-

ing Rasterizing Graphics Hardware. The Visual Com-

puter, 11(9):497–511.

Pan, J., Chitta, S., and Manocha, D. (2011). Probabilistic

Collision Detection Between Noisy Point Clouds Us-

ing Robust Classiﬁcation. In International Symposium

on Robotics Research, Flagstaff, Arizona.

Pedrocchi, N., Malosio, M., and Tosatti, L. M. (2009). Safe

Obstacle Avoidance for Industrial Robot Working

Without Fences. In Proceedings of the 2009 IEEE/RSJ

international conference on Intelligent robots and sys-

tems, IROS’09, pages 3435–3440, Piscataway, NJ,

USA. IEEE Press.

Saravanan, R., Ramabalan, S., and Balamurugan, C. (2007).

Evolutionary optimal trajectory planning for indus-

trial robot with payload constraints. The Interna-

tional Journal of Advanced Manufacturing Technol-

ogy, 38(11-12):1213–1226.

Scott, D. S. and Vuillemin, J. (1986). Art Gallery Theorems

and Algorithms.

Stewart, N. T. (2008). An Image-Space Algorithm for

Hardware-based Rendering of Constructive Solid Ge-

ometry - (1).

Stroulia, E., Schneider, D., and Prassler, E. (1997). Pur-

poseful and Reactive Navigation Based on 9 Qualita-

tive Path Planning and Fuzzy Logic.

Tang, L., Dian, S., Gu, G., Zhou, K., Wang, S., and Feng,

X. (2010). A Novel Potential Field Method for Obsta-

cle Avoidance and Path Planning of Mobile Robots.

Technical report.

ICINCO2014-DoctoralConsortium

Ting, Y., Lei, W., and Jar, H. (2002). A path planning al-

gorithm for industrial robots. Computers & Industrial

Engineering, 42(2-4):299–308.

Tzafestas, C. S., Prokopiou, P. A., and Tzafestas, S. G.

(1998). Path Planning and Control of a Coopera-

tive Three-Robot System Manipulating Large Objects.

pages 99–116.

van den Bergen, G. (1999). A Fast and Robust GJK Imple-

mentation for Collision Detection of Convex Objects.

J. Graph. Tools, 4(2):7–25.

van den Bergen, G. (2001). Proximity Queries and Penetra-

tion Depth Computation on 3D Game Objects. In Pro-

ceedings of Game Developers Conference 2001, San

Jose, CA.

Yakut, I. D. (2010). Collision Detection of Multiple Moving

Objects on GPU. Technical report.

Zavlangas, P. G. and Tzafestas, S. G. (2000). Industrial

Robot Navigation and Obstacle Avoidance Employing

Fuzzy Logic. pages 85–97.

IntelligentPathPanningTowardsCollision-freeCooperatingIndustrialRobots