Towards a Tool-based Methodology for Developing Software for

Dynamic Robot Teams

Roland Gl

uck

, Alwin Hoffmann

, Ludwig N

agele

, Andreas Schierl

, Wolfgang Reif

and Heinz Voggenreiter

Institute for Software & Systems Engineering, University of Augsburg, 86159 Augsburg, Germany

German Aerospace Center, Institute of Structures and Design, Center for Lightweight Production Technology (ZLP),

86159 Augsburg, Germany

Keywords:

Software Methodology, Cooperating Robots, Cyber-Physical Production System, Task Planning, Optimiza-

tion.

Abstract:

Considering initiatives like Industry 4.0 or the Industrial Internet of Things, robots will play an important

role in intelligent factories, producing highly customized products with high variability and in small lot sizes.

In this setting, complexity of planning and programming such robotic applications grows due to the drastic

increase in ﬂexibility, performance and robustness required. In this paper, we propose a tool-supported met-

hodology for the development of control software for dynamically forming multi-functional robot teams. The

main challenges for achieving this overall goal are modeling of robot team skills, techniques for automatically

deriving process steps from the products’ construction plans, ﬁnding allocations of those steps to possible

robot teams with compatible skills and calculating collision-free execution schedules with a high degree of pa-

rallelization to improve cycle times. The proposed approach integrates process experts and automation experts

on all levels. Two case studies will serve as test beds to the developed approach: production of carbon-ﬁber

reinforced polymers and assembly of furniture.

1 INTRODUCTION

Nowadays, industrial robots play an important part in

the efﬁcient automation of production, with high ﬂex-

ibility, speed and precision as unique features com-

pared to other automation principles. However, they

are used mainly – for example in automotive indu-

stry – for the production of high lot sizes with relati-

vely low variability. For more than ten years, research

has been trying to make (industrial) robots efﬁciently

usable in small-lot production (Malec et al., 2007; Pi-

res, 2006). In this context, work has been performed

both on increasing the inherent ﬂexibility of robots,

e.g., by sensor systems and associated control algo-

rithms as described by Albu-Sch

affer et al. (2007)

and Finkemeyer et al. (2010), and on new methods

for more efﬁcient programming and for bringing ro-

bots into operation (Andersen et al., 2015; Neto et al.,

2010). However, for the Industrial Internet of Things

and Smart Factories (Kagermann et al., 2013), furt-

her steps in software development for robot systems

have to be taken to provide the ﬂexible adaptability of

robots required to produce custom-tailored products

with high variability in small lot sizes. In this con-

text, it is no longer sufﬁcient to consider an industrial

robot as a single isolated entity: in fact, the require-

ments of an intelligent factory can only be fulﬁlled

by dynamic and task-speciﬁc cooperation of several

robots. Building robot teams with appropriate, pos-

sibly exchangeable tools leads to multi-functional ro-

bot cells with drastically increased ﬂexibility, perfor-

mance and robustness (Angerer et al., 2015). Howe-

ver, this ﬂexibility comes with high demands on the

control software of these systems: every robot has to

be capable of several process steps, and must coordi-

nate itself and precisely interact with other robots and

machines. At the same time, the effort for a change

of the control software has to decrease considerably

to economically allow high variability and small lot-

sizes of the products. These goals require new ap-

proaches and methods for software development. By

building a detailed virtual model (called digital twin)

of the production line and its work pieces, resources

and processes, new approaches to handle the incre-

ased complexity arise. Different production teams

can be formed dynamically by suitably linking dif-

Glück, R., Hoffmann, A., Nägele, L., Schierl, A., Reif, W. and Voggenreiter, H.

Towards a Tool-based Methodology for Developing Software for Dynamic Robot Teams.

DOI: 10.5220/0006884806050612

In Proceedings of the 15th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2018) - Volume 2, pages 605-612

ISBN: 978-989-758-321-6

605

ferent robots and end-effectors, depending on tasks

and requirements. Each dynamic team of robots and

end-effectors offers new skills to the production plant,

yielding a highly ﬂexible but also highly complex

cyber-physical production system (Kagermann et al.,

2013). Composing and synchronizing suitable robot

teams for tasks is a scheduling and optimization pro-

blem, that should be automated as far as possible to

be applicable to small batch production.

The main goal and contribution of this paper is

to raise awareness for the challenges of using multi-

functional robot cells for ﬂexible production, and to

propose a new methodology of software development

for these cells. Before a multi-functional robot cell

can be used efﬁciently, three main challenges have to

be solved:

Challenge 1: While skills of single robots and their

tools are frequently investigated (Huckaby et al.,

2013; Bj

orkelund et al., 2012; Pfrommer et al., 2015;

Michniewicz and Reinhart, 2014; Stenmark and Ma-

lec, 2015), there is little research about robot teams

with multi-functional or changing tools and sensor sy-

stems. To fully support robot teams, their skills have

to be derived from the single robot skills of the par-

ticipants, and properly described to form the founda-

tion for ﬂexible planning, scheduling and execution

of tasks with robot teams.

Challenge 2: As a next step, the state space of all

possible assignments of tasks to robots and teams as

well as their temporal schedules with parallelism and

collision-free synchronization is tremendously large,

demanding suitable automated techniques to derive

task assignment and schedules from the product or

its structural design. Integrating appropriate planning

and optimization algorithms based on a simulation of

the multi-functional robot cell will be required to of-

fer good resource utilization and cycle times for the

resulting manufacturing plan.

Challenge 3: For execution, dynamic formation of

robot teams and skills poses problems for current ro-

bot controllers. Here, the existing solutions for robot

teams have to be extended to support dynamic teams,

while still offering scalability (from a single robot up

to the size of the multi-functional robot cell) and pre-

cision (with real-time guarantees for coordinated mo-

tions). For seamless transition from simulation to re-

ality, ways of adapting the model of the robot cell to

real world positions are required.

To meet these challenges, the overall goal of our

work is to investigate and to apply a tool-based deve-

lopment methodology for the control software of dy-

namically forming multi-functional robot teams. To

provide a highly automated methodological approach

that fulﬁlls this research goal, we aim to build a mo-

dular planning and simulation tool that supports pro-

duction planning as well as robot team programming.

This paper is structured as follows: In Section 2,

related approaches for planning and ofﬂine program-

ming of robot tasks are introduced. Section 3 outlines

the two application domains that will be used for eva-

luation, followed by a description of the four phases

of the proposed approach in Section 4. Finally, a con-

clusion and outlook is given.

2 RELATED WORK

In classic production planning and control, robots are

seen as machines with ﬁxed, process speciﬁc capa-

bilities encoded as robot programs, and controlled

through manufacturing execution systems (MES) and

programmable logic controllers (PLC). Coordination

between multiple robot tasks is usually performed

through a PLC in the cell, while synchronized acti-

ons such as simultaneous processing of work pieces

with multiple robots is performed on the level of the

robot controllers through vendor-speciﬁc extensions

such as KUKA RoboTeam or ABB MultiMove (Gan

et al., 2013). Robots and PLCs communicate by using

binary or numeric signals and usually do not exchange

any processing details such as the status of work pie-

ces.

Many research approaches try to make this sy-

stem and software structure more ﬂexible. In the PPR

approach (product, process resource) by Bj

orkelund

et al. (2012) and its extensions from the SkillPro

project by Pfrommer et al. (2013), products, proces-

ses, resources and skills are modeled, including com-

plex parameters such as work piece sizes and ma-

terials. Additionally, service oriented architectures

(SOA) have been suggested to structure the software

of MES (Pfrommer et al., 2015), encapsulating ca-

pabilities in vendor speciﬁc skill execution engines.

An extension by Feldmann et al. (2013) adds seman-

tic aspects to the model of functions and parameters,

and uses SysML to model the behavior of devices and

processes. During operation, appropriate devices are

selected based on the abstract model, and PLC code in

a language deﬁned in IEC 61131-3 is generated. Ho-

wever, these approaches do not cover the cooperation

of multiple robots and the forming of dynamic robot

teams.

Other approaches concentrate on real time criti-

cal execution of robot actions and the cooperation of

multiple robots, such as the OROCOS project (Smits

et al., 2009), aRD(x) (Hammer and B

auml, 2014) and

Finroc (Reichardt et al., 2013). The iTaSC appro-

ach (Vanthienen et al., 2013) allows modeling robot

ICINCO 2018 - 15th International Conference on Informatics in Control, Automation and Robotics

606

motions through constraints on the kinematic chain,

which is also applicable to robot teams. These appro-

aches however do not explicitly model robot capabi-

lities that can be used on the MES level to plan with

capabilities and constraints of robot teams.

In the ﬁeld of software architectures and program-

ming methods for industrial robots, the Robotics API

(Angerer et al., 2013) allows implementing real-time

critical control software for industrial robots using

modern programming languages and tools. Toget-

her with a real-time robot control core, it allows si-

mulating and directly executing synchronized opera-

tion of multiple industrial robots by different ven-

dors (e.g. KUKA, Staubli, UR). Additionally, it puts

special emphasis on consistent design principles for

heterogeneous robot systems, and at run time works

with an extensive model of work pieces and the ro-

bot cell. While not addressing planning or optimiza-

tion problems, the concepts of modeling and execu-

ting multi-robot applications seem highly appropriate

for our planned approach.

A further approach is the project ROS-Industrial

While the Robot Operating System (ROS)

is an

open-source middleware and framework for robotics,

ROS-Industrial aims to enable ROS for industrial ro-

botic applications previously technically infeasible. It

deﬁnes standardized interfaces for controlling indus-

trial robots and developing hardware-agnostic soft-

ware for industrial applications. Many robot manu-

factures are involved such as ABB, Adept, Comau,

Fanuc, Universal Robot and Motoman. Although,

the distributed nature of ROS seems appropriate for

multi-robot applications, it is not the focus of ROS-

Industrial so far.

During off-line programming, robot programs are

created in a simulation environment, avoiding occu-

pying the robot cell during programming. Howe-

ver, the simulation has to be similar to reality so that

the developed software can be transferred to the real

cell. According to Gan et al. (2013), typical off-line

programming environments include Tecnomatix Ro-

boCad by Siemens, DELMIA Robotics Ofﬂine Pro-

gramming by Dassault Syst

emes and CimStation Ro-

botics by AC&E. They concentrate on typical robot

tasks such as spot or arc welding, cutting, milling or

painting, and typically do not support to program dy-

namic or heterogeneous robot teams. Other off-line

programming environments are offered by robot ma-

nufacturers, such as KUKA.Sim by KUKA or Robot

Studio by ABB. However support for robot coopera-

tion is still limited. All these off-line programming

environments transform robot paths (e.g. from CAD

www.rosindustrial.org

www.ros.org

data) into vendor speciﬁc robot code, losing informa-

tion about the work pieces so that errors found du-

ring testing in the real robot cell cannot be transferred

back to the model. These environments also are limi-

ted to the capabilities of the underlying robot control-

lers. In the academic context, simulations such as V-

Rep (Rohmer et al., 2013) or Gazebo (Koenig and Ho-

ward, 2004) are typically used. However, these only

support simulation, but no trajectory planning or ro-

bot programming, and thus do not offer an approach

to integrate task and motion planning for robot teams.

Looking at the ﬁeld of process planning and sche-

duling, some approaches concentrate on discrete plan-

ning algorithms for task allocation to robots. Here,

planning domain deﬁnition language (PDDL) intro-

duced by McDermott et al. (1998) and based on the

STRIPS approach (Fikes and Nilsson, 1971) is of-

ten used, e.g. in the SkillPro project. Further ex-

tensions such as the one introduced by Benton et al.

(2012) support soft and hard deadlines and non-linear

cost functions, as well as qualitative aspects such as

special treatment of fragile parts. Other approaches

include the use of SAT solvers (Imeson and Smith,

2014). However, all these works are limited to a sin-

gle robot that will execute the planning results, and do

not support cooperative execution by a team of robots.

In recent years, the integration of task and mo-

tion planning has received increased attentions. In the

work of Kaelbling and Lozano-P

erez (2011) and Le-

vihn et al. (2013), planning and task decomposition is

only performed to a limited depth, then executing the

results and continuing planning based on the resulting

situation to make the planning complexity managea-

ble. In contrast, Bhatia et al. (2011) propose to in-

tegrate the two planning layers (discrete for tasks and

sampling-based for motions), and plan both layers be-

fore execution. To describe assembly tasks, assembly

graphs (Stenmark and Malec, 2015) offer an approach

and a framework to plan, schedule and execute tasks

using an industrial tool chain. Further extensions such

as contact state graphs (Xiao and Ji, 2001) allow force

controlled manipulation. Macho et al. (2016) intro-

duced a hierarchical approach for modular planning

and automatic programming of robot tasks in the ﬁeld

of carbon ﬁber production, incorporating expert kno-

wledge through process speciﬁc modules and asking

the user about geometric degrees of freedom that are

not constrained by the problem deﬁnition. Still, these

approaches do not address multi-robot planning and

cooperative execution.

The ﬁeld of manufacturing with multiple robots

has also been covered by research approaches: Ter-

cio (Gombolay et al., 2013) is an algorithm to as-

sign tasks to multiple similar robot systems that work

Towards a Tool-based Methodology for Developing Software for Dynamic Robot Teams

607

on the same part, avoiding collisions by reducing the

overlap of workspaces. Another approach by Yun

and Rus (2014) uses two different robot types to plan

the assembly of components from many parts by ﬁrst

choosing an optimal placement of assembly robots,

and then planning which assembly robot each de-

livery robot should transport the different parts to.

Knepper et al. (2013) describe the manufacturing of

furniture using a team of mobile robots using an ob-

ject oriented extension of PDDL, but limited to a

ﬁxed set of possible robot cooperations required for

the task and not covering geometric motion planning.

In contrast, an approach for the manufacturing of

large structures by Bourne et al. (2015) connects plan-

ning and scheduling of robot teams, using a dynamic

scheduler based on assembly graphs generated during

planning, and focusing on reaching the required preci-

sion through a team of mobile robots. However, these

approaches do not cover the capabilities of dynamic

robot teams.

To summarize, the full potential of multi functio-

nal robot cells in intelligent factories can only be used

by forming dynamic, task speciﬁc robot teams. Ho-

wever, this ﬁeld has not been covered, neither from

a production nor software point of view. Existing ap-

proaches are not up to the resulting complexity of mo-

deling robot teams, planning and scheduling tasks and

simulating and executing the results with teams of ro-

bots, and especially to the integration of all levels re-

quired for a consistent methodological approach.

3 CASE STUDIES

To evaluate and demonstrate the methodology, two

case studies – one in the domain of carbon ﬁber-

reinforced polymer production and another in the

domain of furniture assembly – will serve as test beds.

3.1 Production of CFRP Parts

In aerospace industry, carbon ﬁber-reinforced poly-

mers (CFRP) are a widely used material due to their

strength-to-weight ratio. One possibility to produce

CFRP parts is to drape multiple layers of carbon ﬁber

cut pieces into a mold and infuse the ﬁnished com-

position with a suitable resin. After hardening in an

oven or autoclave, CFRP parts in the desired form are

obtained. Figure 1 shows a cross section through a

mold and two layers of cut pieces.

Experts use specialized CAD programs to con-

struct CFRP parts considering static and structural

goal properties. This design process yields a so-called

Figure 1: Example layer construction in a mold: ﬁve cut

pieces in two stacked layers.

Figure 2: Two robots carrying a cut piece.

construction part design (CPD) which is used as con-

struction plan. For automated production, one or two

robots use speciﬁc end-effectors to pick cut pieces

provided on a desk and place them into a form. Fi-

gure 2 shows two robots executing this task.

However, manually creating a program for robot

teams based on a CPD is very tedious and can take up

to hours for one single cut piece (N

agele et al., 2015).

For the production of a full-scale part, hundreds of

cut pieces have to be processed. To make this process

economically feasible, automated or computer aided

generation of programs is required.

3.2 Assembly

Geisberger and Broy (2015) describe how a cyber-

physical system can handle orders of custom-tailored

kitchens. Because the assembly of speciﬁc articles of

furniture is a suitable task for robot teams, we also

plan to apply our methodology to this domain. Here,

the number of basic parts is less than in the previous

case, but the process steps are of a much higher vari-

ability. Depending on the parts, assembling, screwing

or gluing steps have to be executed, each of them de-

manding own tools and algorithms for execution and

planning. Greater and heavier parts like table pla-

tes have to be handled by more than one robot. In

a multi-functional cell, many steps can be performed

in parallel, e.g. the simultaneous bolting of table legs

or the insertion of shelves as depicted in Figure 3. In

contrast to CFRP production, typically no machine-

ICINCO 2018 - 15th International Conference on Informatics in Control, Automation and Robotics

608

Figure 3: Example for assembly steps of a table.

readable construction plan exists for these assembly

tasks. However, existing assembly instructions offer

a good basis for it.

4 APPROACH

To make the complexity of dynamic robot teams ma-

nageable, we propose a planning approach with four

phases as shown in Figure 4. First, an existing blu-

eprint is analyzed and decomposed into atomic sub-

tasks (goals) that can be solved using the existing

robot team capabilities of the multi-functional cell

(Section 4.1). Appropriate planning modules, which

use robot team capabilities to create processes, are

assigned to these goals (Section 4.2). In a schedu-

ling and planning phase (Section 4.3), these planning

modules are used to derive a consistent overall pro-

cess that can be simulated and executed using a real

robot cell (Section 4.4). During important steps of

the entire process, expert knowledge is integrated, in-

cluding process experts contributing to the top-down

view of the product centered manufacturing problem,

and automation experts providing a bottom-up view

based on robot team capabilities. To ﬁnd appropri-

ate process parameters, the approach allows iterative

parameter tuning while keeping previous results, and

evaluate the effects in simulation.

4.1 Analysis

The methodology is based on a detailed blueprint des-

cribing the layout and structure of the desired product,

including details about different production steps as

well as temporal dependencies in the production pro-

cess. An extensible set of analysis modules trans-

forms the blueprint into a set of intermediate goals,

which are analyzed further until atomic goals are re-

ached. In the case of CFRP production, a ply-book

as part of the CPD is ﬁrst separated into the different

layers, and then into separate cut pieces that have to

be placed. Finally, for each cut piece atomic goals to

pick up or drape are created. For these resulting goals,

usually temporal dependencies have to be met: In the

CFRP example, a cut piece may only be draped after

the underlying layer has been placed. These depen-

dencies are expressed through constraints concerning

the goals. As analysis modules depend on the struc-

ture of the blueprint, they are mostly domain speciﬁc.

Thus, this phase heavily depends on the domain kno-

wledge of experts, and the analysis modules can vary

in number and structure. Additionally, different ana-

lysis modules can exist for the same problem (e.g. to

place the cut pieces from the center to the border, or

from one side to the other), leading to alternatives that

can be considered in the following phases.

As a result, the analysis provides a set of hierar-

chically structured, partially alternative goals, as well

as constraints between the different goals.

4.2 Assignment

In the second phase, an extensible set of planning mo-

dules is used to ﬁnd solutions for each of the goals

identiﬁed in phase 1. Each planning module knows

which types of goals it can plan for, and uses the capa-

bilities of the robot cell to compute a speciﬁc solution.

In the example of CFRP production, a planning mo-

dule uses the robot capabilities ’deform’, ’position’

and ’ﬁxate’ to plan for the speciﬁc goal ’place cut

piece’. On this layer, the main problems belong to

the automation domain, so automation experts have

to contribute to the creation and development of plan-

ning modules based on robot team capabilities. As a

result of this phase, each goal receives a set of poten-

tially suitable planning modules that ﬁnd their requi-

red capabilities in the robot cell, as well as constraints

concerning the capabilties derived by the planning

modules (e.g. grasping objects of a certain size).

4.3 Scheduling and Planning

The following phase attempts to create speciﬁc and

integrated robot programs based on the results of the

previous phases. Thus, for each speciﬁc goal a pro-

gram solution has to be calculated that is executa-

ble in the robot cell. For each goal, a planning mo-

dule is chosen, and for each of its required capabi-

lities a suitable robot or team is selected. Using the

selected robot or team, the module plans and crea-

tes a speciﬁc solution that occupies the assigned ro-

bots or teams. Additionally, further constraints arise

during planning, e.g. through occupied areas in the

workspace. These constraints have to be considered

by planning modules for following goals. If the re-

quired degrees of freedom do not allow planning the

following goals, parts of the previously created pro-

Towards a Tool-based Methodology for Developing Software for Dynamic Robot Teams

609

Phase 1:

Analysis

Phase 2:

Assignment

Phase 3:

Scheduling

& Planning

Phase 4:

Simulation

& Execution

Blueprint

Planning Modules

Analysis Modules

Reconfiguration Modules,

Robots and Teams

Product

Goals (circles) und

Constraints between Goals

(arrows)

Robot Programs for Goals and

Reconfigurations in given

Temporal Order

Goals with suitable

Planning Modules

Simulated or Real

Robot Cell

Figure 4: Four phases of the proposed approach and intermediate artifacts.

gram have to be replaced. This can be done by the

same planning module by computing another solution

with different constraints (e.g. using other trajectories

or other robots), by choosing another planning mo-

dule, or by using alternative goals found in the previ-

ous phases.

Apart from the actual task execution, each pro-

gram solution for a goal contains implicit assumpti-

ons about the start and end conﬁguration of the robot

cell, in the form of an expected position of all robots

before the task, as well as a resulting position after

execution. To consistently match subsequent program

parts, intermediate programs have to be planned that

perform necessary reconﬁgurations in the robot cells,

such as transfer motions, tool changes or dynamically

creating robot teams. In the example of Figure 2, both

robots can either work independently, or form a dyn-

amic team with cooperative end-effector movements.

These kinds of program completions mainly consist

of actions in the robot cell and are provided by auto-

mation experts as reconﬁguration modules.

The complete process can be considered as a com-

plex and dynamic planning and optimization problem,

providing a rich set of solution tools. For goals, the

relevant constraints can be transformed into a for-

mal constraint satisfaction problem that can be sol-

ved through techniques of constraint programming

and optimization. The central task is to select alter-

native goals and speciﬁc planning modules, to assign

robot teams and to calculate the temporal sequence of

goals and reconﬁgurations. Some decisions – such

as selecting alternative goals or supplying missing

information – can interactively be taken by experts

through semi-automatic programming, allowing auto-

mation and process experts to inﬂuence the kind and

quality of results. As a result of this phase, a consis-

tent and integrated control program has been created

that reaches the given goals using robots and teams in

a speciﬁc temporal order.

4.4 Simulation and Execution

Based on the created control program, process experts

can use a simulation environment to test or adapt the

generated program in a virtual environment, and to

visualize different alternatives for comparison, while

seeing meta information about process steps and for-

med teams. Additionally, the simulation environment

allows automation experts to use advanced functions

such as collision detection or geometric relationships

to implement planning and reconﬁguration modules,

directly using the model of the robot cell for planning.

To execute a control program on a real robot cell, the

required robot team capabilities have to be transfor-

med into equivalent robot and tool actions. Here,

the real-time critical implementation of team coope-

ration, as well as challenges of distribution and de-

centralization have to be respected.

ICINCO 2018 - 15th International Conference on Informatics in Control, Automation and Robotics

610

After a program has been tested in simulation, se-

amless transition to real multi-functional robot cells is

required. To facilitate this, adaptations of the program

need to be possible to ﬁx observed robot misbehavior

or production inaccuracies. The approach iteratively

allows for adjusting and calibrating process parame-

ters and robot cell settings such as positions or dimen-

sions that can be transformed into a corrected control

program through re-executing phase 3.

5 CONCLUSIONS

In this paper, we motivated the need for multi-

functional robot cells to handle small lot sizes and

mass customization. To cope with the great adap-

tability and ﬂexibility of such cells and to make

production economically feasible, effective planning

tools are required. As a solution, we proposed a tool-

supported methodology for the development of cont-

rol software for dynamically forming multi-functional

robot teams. The approach addresses the challenges

of modeling robot team skills, automatically deriving

process steps from the products’ construction plans,

ﬁnding allocations of those steps to possible robot te-

ams with compatible skills and calculating appropri-

ate execution schedules with good resource utilization

and cycle times. For challenging domains, the appro-

ach allows experts from both process and automation

to contribute their knowledge in all steps of process

deﬁnition.

We plan to implement the proposed approach and

to evaluate it in simulation as well as in reality. Ho-

wever, especially in the area of composing dynamic

robot teams and ﬁnding optimal task schedules for

them, further research is required to successfully ap-

ply the approach to challenging domains. Neverthe-

less, the proposed approach is a promising solution

for automatic production planning in multi-functional

robot cells involving expert knowledge.

ACKNOWLEDGEMENTS

This work is partly funded by the German Research

Foundation (DFG) under the TeamBotS grant.

REFERENCES

Albu-Sch

affer, A., Ott, C., and Hirzinger, G. (2007). A uni-

ﬁed passivity-based control framework for position,

torque and impedance control of ﬂexible joint robots.

Intl. J. Rob. Research, 26(1):23–39.

Andersen, R. H., Dalgaard, L., Beck, A. B., and Hallam, J.

(2015). An architecture for efﬁcient reuse in ﬂexible

production scenarios. In Proc. 2015 IEEE Conf. on

Autom. Science and Engineering (CASE), pages 151–

157.

Angerer, A., Hoffmann, A., Schierl, A., Vistein, M., and

Reif, W. (2013). Robotics API: Object-oriented soft-

ware development for industrial robots. Softw. Eng.

for Robotics, 4(1):1–22.

Angerer, A., Vistein, M., Hoffmann, A., Reif, W., Krebs, F.,

and Sch

onheits, M. (2015). Towards multi-functional

robot-based automation systems. In Filipe, J., Ma-

dani, K., Gusikhin, O. Y., and Sasiadek, J. Z., editors,

Proc. 12

Intl. Conf. on Inform. in Control, Autom. &

Robot. (ICINCO 2015), pages 438–443. SciTePress.

Benton, J., Coles, A., and Coles, A. (2012). Temporal plan-

ning with preferences and time-dependent continuous

costs. In Proc. 22

Intl. Conf. on Autom. Planning &

Scheduling, ICAPS.

Bhatia, A., Maly, M. R., Kavraki, L. E., and Vardi, M. Y.

(2011). Motion planning with complex goals. IEEE

Robotics Automation Magazine, 18(3):55–64.

orkelund, A., Bruyninckx, H., Malec, J., Nilsson, K., and

Nugues, P. (2012). Knowledge for intelligent indus-

trial robots. In AAAI Spring Symposium: Designing

Intelligent Robots.

Bourne, D., Choset, H., Hu, H., Kantor, G., Niessl, C., Ru-

binstein, Z., Simmons, R., and Smith, S. (2015). Mo-

bile manufacturing of large structures. In Proc. 2015

IEEE Intl. Conf on Robotics & Automation (ICRA),

pages 1565–1572.

Feldmann, S., Loskyll, M., Rosch, S., Schlick, J., Zuhlke,

D., and Vogel-Heuser, B. (2013). Increasing agility in

engineering and runtime of automated manufacturing

systems. In Proc. 2013 IEEE Intl. Conf. on Industrial

Technology (ICIT), pages 1303–1308.

Fikes, R. and Nilsson, N. J. (1971). STRIPS: A new appro-

ach to the application of theorem proving to problem

solving. Artif. Intell., 2(3/4):189–208.

Finkemeyer, B., Kr

oger, T., and Wahl, F. (2010). The adap-

tive selection matrix - a key component for sensor-

based control of robotic manipulators. In Proc. 2010

IEEE Intl. Conf. on Robotics and Automation (ICRA),

pages 3855–3862.

Gan, Y., Dai, X., and Li, D. (2013). Off-line programming

techniques for multirobot cooperation system. In Intl.

J. of Advanced Robotic Systems. InTech Europe.

Geisberger, E. and Broy, M. (2015). Living in a networked

world. Integrated research agenda Cyber-Physical Sy-

stems (agendaCPS) (acatech STUDY). Herbert Utz

Verlag.

Gombolay, M., Wilcox, R., and Shah, J. (2013). Fast sche-

duling of multi-robot teams with temporospatial con-

straints. In Proc. of Robotics: Science and Systems

(RSS), Berlin, Germany.

Hammer, T. and B

auml, B. (2014). The communication

layer of the ardx software framework: Highly perfor-

mant and realtime deterministic. Intelligent & Robotic

Systems, 77(1):171–185.

Towards a Tool-based Methodology for Developing Software for Dynamic Robot Teams

611

Huckaby, J., Vassos, S., and Christensen, H. I. (2013). Plan-

ning with a task modeling framework in manufactu-

ring robotics. In Proc. 2013 IEEE/RSJ Intl. Conf. on

Intelligent Robots and Systems (IROS), pages 5787–

5794.

Imeson, F. and Smith, S. L. (2014). A language for robot

path planning in discrete environments: The tsp with

boolean satisﬁability constraints. In Proc. 2014 IEEE

Intl. Conf. on Robotics and Automation (ICRA), pages

5772–5777.

Kaelbling, L. P. and Lozano-P

erez, T. (2011). Hierarchical

task and motion planning in the now. In Proc. 2011

IEEE Intl. Conf. on Robotics and Automation (ICRA),

pages 1470–1477.

Kagermann, H., Wahlster, W., and Helbig, J., editors

(2013). Umsetzungsempfehlungen f

ur das Zukunfts-

projekt Industrie 4.0 – Abschlussbericht des Ar-

beitskreises Industrie 4.0. acatech – Deutsche Aka-

demie der Technikwissenschaften e.V.

Knepper, R. A., Layton, T., Romanishin, J., and Rus, D.

(2013). Ikeabot: An autonomous multi-robot coordi-

nated furniture assembly system. In Proc. 2013 IEEE

Intl. Conf. on Robotics and Automation (ICRA), pages

855–862.

Koenig, N. and Howard, A. (2004). Design and use para-

digms for gazebo, an open-source multi-robot simula-

tor. In Proc. 2004 IEEE/RSJ Intl. Conf. on Intelligent

Robots and Systems (IROS), pages 2149–2154.

Levihn, M., Kaelbling, L. P., Lozano-P

erez, T., and Stilman,

M. (2013). Foresight and reconsideration in hierarchi-

cal planning and execution. In Proc. 2013 IEEE/RSJ

Intl. Conf. on Intelligent Robots and Systems (IROS),

pages 224–231.

Macho, M., N

agele, L., Hoffmann, A., Angerer, A., and

Reif, W. (2016). A ﬂexible architecture for automa-

tically generating robot applications based on expert

knowledge. In Proc. 47

Intl. Symp. on Robotics,

ISR 2016, Munich, Germany. VDE Verlag.

Malec, J., Nilsson, A., Nilsson, K., and Nowaczyk, S.

(2007). Knowledge-based reconﬁguration of automa-

tion systems. In Proc. 2007 IEEE Intl. Conf. on Auto-

mation Science and Engineering (CASE), pages 170–

175.

McDermott, D., Ghallab, M., Howe, A., Knoblock, C.,

Ram, A., Veloso, M., Weld, D., and Wilkins, D.

(1998). PDDL — the planning domain deﬁnition lan-

guage. Technical Report TR 98 003/DCS TR 1165,

Yale Center for Computational Vision and Control.

Michniewicz, J. and Reinhart, G. (2014). Cyber-physical

robotics – automated analysis, programming and con-

ﬁguration of robot cells based on cyber-physical-

systems. Procedia Technology, 15:566–575.

agele, L., Macho, M., Angerer, A., Hoffmann, A., Vistein,

M., Sch

onheits, M., and Reif, W. (2015). A backward-

oriented approach for ofﬂine programming of com-

plex manufacturing tasks. In Proc. 6

Intl. Conf.

on Automation, Robotics and Applications (ICARA

2015), pages 124–130.

Neto, P., Norberto Pires, J., and Paulo Moreira, A. (2010).

High-level programming and control for industrial ro-

botics: using a hand-held accelerometer-based input

device for gesture and posture recognition. Industrial

Robot, 37(2):137–147.

Pfrommer, J., Schleipen, M., and Beyerer, J. (2013). PPRS:

production skills and their relation to product, process,

and resource. In Proc. IEEE 18

Conf. on Emerging

Technologies & Factory Automation (ETFA 2013).

Pfrommer, J., Stogl, D., Aleksandrov, K., Navarro, S. E.,

Hein, B., and Beyerer, J. (2015). Plug & produce by

modelling skills and service-oriented orchestration of

reconﬁgurable manufacturing systems. Automatisie-

rungstechnik, 63(10):790–800.

Pires, J. N. (2006). Robotics for small and medium enterpri-

ses: control and programming challenges. Industrial

Robot, 33(6).

Reichardt, M., F

ohst, T., and Berns, K. (2013). On soft-

ware quality-motivated design of a real-time frame-

work for complex robot control systems. In Proc. 7

Intl. Workshop on Software Quality and Maintainabi-

lity, Genova, Italy.

Rohmer, E., Singh, S. P., and Freese, M. (2013). V-rep: A

versatile and scalable robot simulation framework. In

Proc. 2013 IEEE/RSJ Intl. Conf. on Intelligent Robots

and Systems (IROS), pages 1321–1326.

Smits, R., Laet, T. D., Claes, K., Soetens, P., Schutter, J. D.,

and Bruyninckx, H. (2009). Orocos: A software fra-

mework for complex sensor-driven robot tasks. IEEE

Rob. & Autom. Mag.

Stenmark, M. and Malec, J. (2015). Knowledge-based in-

struction of manipulation tasks for industrial robo-

tics. Robotics and Computer-Integrated Manufactu-

ring, 33:56–67.

Vanthienen, D., Klotzbuucher, M., De Schutter, J., De Laet,

T., and Bruyninckx, H. (2013). Rapid application de-

velopment of constrained-based task modelling and

execution using domain speciﬁc languages. In Proc.

2013 IEEE/RSJ Intl. Conf. on Intelligent Robots and

Systems (IROS), pages 1860–1866.

Xiao, J. and Ji, X. (2001). Automatic generation of high-

level contact state space. Rob. Research, 20(7):584–

606.

Yun, S. and Rus, D. (2014). Adaptive coordinating

construction of truss structures using distributed

equal-mass partitioning. IEEE Trans. on Robotics,

30(1):188–202.

ICINCO 2018 - 15th International Conference on Informatics in Control, Automation and Robotics

612