Annual Reviews in Control 31 (2007) 69–79

www.elsevier.com/locate/arcontrol

Present and future robot control development—An industrial perspective

Torgny Brogårdh
ABB Robotics, SE-721 68 Västerås, Sweden
Received 5 October 2006; accepted 14 January 2007
Available online 4 May 2007

Abstract
Robot control is a key competence for robot manufacturers and a lot of development is made to increase robot performance, reduce robot cost
and introduce new functionalities. Examples of development areas that get big attention today are multi robot control, safe control, force control,
3D vision, remote robot supervision and wireless communication. The application benefits from these developments are discussed as well as the
technical challenges that the robot manufacturers meet. Model-based control is now a key technology for the control of industrial robots and models
and control schemes are continuously refined to meet the requirements on higher performance even when the cost pressure leads to the design of
robot mechanics that is more difficult to control. Driving forces for the future development of robots can be found in, for example, new robot
applications in the automotive industry, especially for the final assembly, in small and medium size enterprises, in foundries, in food industry and in
the processing and assembly of large structures. Some scenarios on future robot control development are proposed. One scenario is that light-
weight robot concepts could have an impact on future car manufacturing and on future automation of small and medium size enterprises (SMEs).
Such a development could result in modular robots and in control schemes using sensors in the robot arm structure, sensors that could also be used
for the implementation of redundant safe control. Introducing highly modular robots will increase the need of robot installation support, making
Plug and Play functionality even more important. One possibility to obtain a highly modular robot program could be to use a recently developed
new type of parallel kinematic robot structure with large work space in relation to the robot foot print. For further efficient use of robots, the
scenario of adaptive robot performance is introduced. This means that the robot control is optimised with respect to the thermal and fatigue load on
the robot for the specific program that the robot performs. The main conclusion of the presentation is that industrial robot development is far away
from its limits and that a lot of research and development is needed to obtain a more widely use of robot automation in industry.
# 2007 Elsevier Ltd. All rights reserved.

Keywords: Industrial robots; Robot control; Control functions; Control applications

1. Introduction industrial robots of today (UNECE, 2004). This means that the
requirements emanating from this type of manufacturing
The development of industrial robots is characterized by a system drive much of the robot development. Thus, most robots
multidisciplinary fusion of a large spectrum of technologies. of today are well adapted to cost sensitive high volume flexible
Many of these technologies are not specific for robotics and can production in a very competitive environment. This has made it
be developed from solutions in other much larger product areas. necessary for the robot manufacturers to make very big efforts
However, robot control and then especially robot motion on the basic requirements on cost efficiency, high reliability and
control, is very specific to the robot product and constitutes one high productivity. Moreover, it has been necessary to adapt the
of the most important key competences for the development of robot control to the plant automation systems with respect to
industrial robotics. By applying and developing advanced application protocols, communication systems, I/O-interfaces,
control, it is possible to continuously improve the robot PLC-equipment, user interfaces, process equipment, etc., for
performance, which is necessary in order to increase optimal use of the robots and for short change-over times
performance and lower cost of industrial robot automation. between different products (RIA & NIST, 2000).
It should be emphasized that the automotive industries In the automotive industry, which will be the starting point
including their supply chains are the dominating customers for for this presentation, full robot automation is usually found for
car body assembly, press tending, painting and coating and to
some extent for engine and power train assembly (ABB-1,
E-mail address: [email protected]. 2003). These applications are well established and the robot

1367-5788/$ – see front matter # 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.arcontrol.2007.01.002
70 T. Brogårdh / Annual Reviews in Control 31 (2007) 69–79

features with respect to installation, programming, integration,

maintenance, performance and functionalities are continuously
refined. From a control point of view, this means increasing
requirements on robustness, stability and accuracy. At the same
time, the cost pressure implies the need of development of less
rigid mechanical structures with more complex mechanical
robot vibration modes and larger variations of the dynamics of
the individual robots, which must be handled by the control
system.
Looking further into the automation of car manufacturing,
only few robots are used today for the final assembly. Here, new
robot technology and new flexible automation solutions are
needed to handle the complex assembly tasks and the variability
of the product geometries. One big challenge for industrial
robotics in the future is to obtain economically feasible
Fig. 1. Multi robot control requires high performance model-based control
solutions to this kind of applications, where robot control needs concepts and efficient methods for robot programming.
to deal with tolerances in geometries and processes, to be more
intuitive and to be more interactive. A break through in this there is also the problem of how to dynamically split tasks
direction could give a new wave of robotics for a large spectrum between robots and between clusters of robots for optimal
of industrial applications, where robots are not realistic to use productivity. Simultaneously, it should be noted that collabor-
today. However, before discussing the future of robot control ating robots are more difficult to program than single robot
technology, some examples of the ongoing major industrial installations and therefore off line programming using robot
robot control development will be given. Then the driving and cell models is even more motivated. One problem then is
forces for the robot development will be discussed and based on the accuracy of the collaborating robot installations. Since
this some scenarios about the future will be outlined. serially connected kinematic chains have to be controlled,
errors in the servo loops and in the models of robot kinematics
2. Present industrial robot control development and dynamics will give bigger pose deviations between the tool
and the work object than in single robot installations. Therefore,
There is a tendency of technology fashion in the the development of multi robot control will motivate further
development of industrial robots. This usually originates from improvements of the accuracy of the kinematic and dynamic
new production concepts brought forward in the automotive robot models as well as of the robot servo performance.
industry. One example of a present popular development is Related to the need of multi robot control are the
about multi robot control and different solutions are now requirements to develop robots with very high load capacity
presented by several robot manufacturers, though a couple of (ABB-2, 2001). Robots handling loads up to 500 kg have been
manufacturers already have offered this as a product for some developed to handle, for example, parts of car bodies. Then one
years (Bredin, 2005). The main reason for adopting multi robot heavy load handling robot holds the fixture with the car body
control in industry is the possibility to reduce production cost components while other robots make, for example, spot- or arc
by having robots working in parallel, especially for low speed welding. This kind of automation solution is motivated by the
processes as arc welding. Other benefits are that several robots higher flexibility obtained when replacing single purpose
can be controlled from one controller, floor space is saved, transport systems with robots. Beside the multi robot control
collision avoidance performance can be improved and cycle aspect on this, the development towards higher load capabilities
times can be reduced. In arc welding, it will moreover be result in lower mechanical robot eigenfrequencies, increasing
possible to obtain a symmetrical heat distribution by welding the difficulties to accurately model and control the robot
on the same object from different directions simultaneously. A structure.
common set up is to use two or more robots welding on the Another popular development direction is towards new
same work object, which is rotated by a manipulator of one or safety arrangements in robot installations (ESALAN Systems,
two degrees of freedom. For higher flexibility, the work object 2006). One short term motivation for this is the possibility to
can also be hold by a robot (Fig. 1) and still another robot can replace electrical and mechanical working range limits with
hold a component to be welded on the work object. Automotive safe software limits, which makes it cheaper and faster to
industry is also interested in reducing the cycle time of spot configure a robot cell. Moreover, the safety fences of a robot
welding robots by improved coordination of groups of robots cell can be more efficiently adapted to the work space
working on a common car body. Examples of difficult control limitations, which will save floor space in the workshop. There
tasks, when developing multi robot control for the manufactur- are also initiatives to develop new concepts for safe direct
ing industry, are exact timing of dynamically optimised servo collaboration between human and robot (Schraft, Meyer,
references, smooth transitions between coordinated and not Parlitz, & Helms, 2005). Examples of applications for such
coordinated robot movements, exception handling and failure collaboration are material handling, machine tending, compo-
recovery. When groups of robots work in a large production line nent transfer and assembly. In order to increase the safety level
 T. Brogårdh / Annual Reviews in Control 31 (2007) 69–79 71

for human–robot cooperation, the redundancy of the robot parameters need to be available and low cost force sensors
supervision of hard- and software is increased, for example, by should be developed. Fast growing applications may be found
two channels measurement systems and failsafe bus- and I/O in the automotive industry and then for drive line component
systems. The redundancy needed can of course be obtained in assembly, for example, for shaft insertion, spline matching and
many ways but simultaneously it is important not to increase torque converter mounting (Fig. 2). Since force- and torque
the cost of the redundancy functionality more than it can be sensing will be necessary in future human–robot collaboration
accepted by the applications. One control aspect of new safety the ongoing force/torque sensor based control development
concepts is how to make use of the robot models already could be important for the understanding of how to integrate
running in real time in robot controllers to obtain a fault this technology into collaborative robotic systems.
detection that is sensitive enough without giving too many false Just as for force control, robot vision has been used for a long
alarms. It is also important to be able to supervise the safety time without finding any large volume robot applications in the
functionality so it really works when an emergency situation manufacturing industries. One of the reasons for this is the lack
takes place. For example, the brakes and the robot supervision of robustness of a 2D vision system in a typical workshop
functionality must be cyclic tested. This ongoing development environment and therefore vision systems to control robots are
of safety technology could be a first step against the safe mainly used where the camera scenes are well defined and the
robotics needed for future human robot collaboration in less light conditions can be controlled, for example, for objects to be
structured environment than in the robot installations of today. picked and placed on conveyors (ABB-3, 2004). However,
Still another ongoing development direction can be found in using the 3D vision products now available on the market it is
using wireless communication in robot systems (Katzel, 2004). possible to improve the robustness of robot vision and system
The biggest interest for wireless is with respect to the solutions can be developed to increase the flexibility in, for
communication between the teach pendant and the controller, example, material handling, machine tending and assembly
but there is also an interest in wireless communication between (Braintech-ref, 2006). Even bin picking can now be made
the robot controller and sensors and process equipment (Frey, robust enough for industrial use (Watanabe, Kazunori,
Endresen, Kreitz, & Scheibe, 2005). Experiences from test
installations have shown that the communication itself is not a
major problem in a workshop environment but the big issues are
about safety, for example, to find concepts for safe wireless
emergency handling and for safe selection of robots at the log in
of a teach pendant. It should also be pointed out that since a
large spectrum of devices with wireless transmission is
produced for the consumer market there are possibilities to
make use of such devices as low cost user interfaces to robot
controllers (d’Angelo & Corke, 2001). In a robot control
development perspective, new safety mechanisms for wireless
transmission will be important both with respect to wireless
programming devices and wireless sensors and actuators.
A sensor type that has got increasing industrial attention the
latest years is the six degrees of freedom (DOF) force/torque
sensor (ATI et al., 2006). Even though robot manufacturers
have been able to deliver robots with force control (Siciliano &
Villani, 2000) for a long time, there has not yet been any wider
use of this functionality. Examples of applications where six
DOF force/torque sensors are used for the control of industrial
robots are grinding, deflashing, deburring, milling, polishing,
testing and assembly. In the material removal applications, the
advantages with force sensor based control are higher process
quality, easier calibration and relaxed requirements on the
accuracy of fixtures and grippers (Pires, Ramming, Rauch, &
Araújo, 2002) The advantages in assembly are in addition
shorter cycle times, reduced impact forces and less risk of
jamming, wedging and galling (Zhang, Zhongxue, Brogårdh,
Wang, & Isaksson, 2004). In the applications for force sensor
based control, manual work with bad working conditions is
often found and health risk is therefore also a reason for the
introduction of this technology. In order to increase the use of
force sensor controlled robots, programming and tuning Fig. 2. Assembly of automobile drive line components, as the torque converter
methods need to be further developed, standard sets of process in the figure, is one example where robot force control has a big potential.
72 T. Brogårdh / Annual Reviews in Control 31 (2007) 69–79

Warashina, & Kumiya, 2005) and there is also the possibility to controllers and remote automatic data acquisition of robot
make use of 3D vision techniques for the calibration of tools, production data for optimisation, monitoring, preventive
work objects, fixtures and other robot cell components. The maintenance and fault isolation is further developed. The fault
main development with respect to 3D vision is of course related isolation can be based on data from the sensors in the servo
to feature extraction and other computer vision problems and loops, special supervision sensors, current and voltage levels in
the robot control comes in when designing high performance the drive system and virtual sensors using the observer concept.
vision interfaces (Lippiello, Siciliano, & Villani, 2005). This development will certainly proceed and it is expected that
Related to 3D vision is the laser tracking sensor technology dynamic robot models run in real time will be even more used
used mainly for the control of arc welding paths when the for fault detection, fault isolation and diagnosis, based on
position of the work object cannot be guaranteed (Servorobot, residual generation and system identification methods (Mattone
2006). These sensor types have been in use for a long time but & De Luca, 2003; Ostring, 2002).
there is a trend towards full 3D measurements, for example, by Model-based control (Sciavicco & Siciliano, 2000) has been
the introduction of more than one scanning line in the found to be very important in robotics in order to fulfil the
triangulation-based tracking sensors. The same development contradictory requirements on performance improvements and
trend is also found in optical measurement systems carried by cost reductions (Fig. 4). The ongoing development is directed
robots for inspection of, for example, car bodies and car sub towards more complex kinematic- and dynamic models, more
assemblies. For tracking sensors, it is important to optimise the complex multiple-input multiple-output (MIMO) control
use and coordination of the look-ahead information in relation schemes, bigger variations of static and dynamic model
to the robot program, to be able to handle situations with parameters, increasing noise- and disturbance levels, larger
lacking measurement data and to be able to perform efficient number of low mechanical eigenfrequencies and enlarged non-
recoveries from process failures. In a longer perspective, the 3D linearities. Even if a lot of academic research has been made on
vision concepts for handling, assembly, tracking, inspection, all of these aspects, a lot of applied research is needed in order
calibration, etc., can be further integrated into the robot to further improve the model-based robust control of industrial
controller and from performance point of view it could then for robots. Hand in hand with the further development of model-
some applications also be motivated to use 3D vision in the based control there is an important development of model-
robot servo loop. based design using virtual prototyping to improve the
Other ongoing improvements of the robot technology are performance/cost ratio, reduce the development cost and to
related to the user- and application interfaces of the robot be able to shorten the product cycles (Pettersson, 2005). Very
controller with the purpose to make robot programming, important is then the mechatronic design approach with
operation and maintenance simpler even though the complexity development teams consisting of both mechanical design and
of the robot systems is continuously increasing. Wizard-like robot control specialists.
step-by-step concepts based on graphical representations of
robot movements and process actions are realized also for the 3. Driving forces for the future development of
use on teach pendants (Fig. 3; ABB-4, 2006), the realistic robot industrial robot control
simulation interface is implemented on the robot programming
level (RRS, 2006), tools for process modelling are developed to As for all other industrial products, the main robot
simplify robot programming (Skarin, Claesson, & Bergling, development is governed by the need of its users. However,
2004), PLC functionality for logic control and advanced robots are quite unique with respect to its versatility, which
process- and equipment control is integrated into the robot means that robot technology is developed towards a large

Fig. 3. Using PC software, touch-screen and efficient tools for application development, the teach-pendant is simultaneously a generic GUI-device and a specialized
robot programmer- and operator tool.
 T. Brogårdh / Annual Reviews in Control 31 (2007) 69–79 73

Fig. 4. Model-based control of a whole robot family supported by a common controller is obtained by configurable rigid body and elastodynamic models.

spectrum of requirements from different automation concepts is at present very eager to use robotics. Looking at the
and different applications (Robot Manufacturer links, 2006). customer-driven requirements, the base requirements are of
Starting with the application requirements, development is course to reduce the production cost and increase the quality of
needed to cope with requirements arising from new processes the end products and new automation concepts are tried from
together with requirements from improvements of traditional time to time to obtain a significant increase in productivity.
processes. New processes are defined here as processes that are Examples of such concepts are compact robots, backward
new in combination with industrial robots as, for example, bending robots, heavy load robots, multi robot control, accurate
friction stir welding, milling, high performance laser cutting, robots for off-line programming, flexible framing robots,
gearbox assembly, flat TV-screen handling and sheet metal sensor-controlled robots and safe robots. At the same time,
deforming. Examples of traditional processes that ask for there is the continuous need to improve the flexibility of the use
further improvement of robot performance are water jet cutting, of robots, improve reliability, improve working conditions,
laser cutting, laser welding, gluing, grinding, de-burring, decrease life cycle cost, to make installation, system
measurement and assembly. Sometimes robot automation gets integration, programming and maintenance easier and to
very popular for manufacturers working with a specific increase performance (Fig. 5). All these requirements drive
application or material, as an example, the plastics industry the robot manufacturers to make a continuous incremental

Fig. 5. In order to fulfil the increasing requirements on robot accuracy for off line programming, kinematic and elastostatic identification- and compensation methods
are continuously refined and today a volumetric accuracy of 0.5 mm can be obtained even for a big robot.
74 T. Brogårdh / Annual Reviews in Control 31 (2007) 69–79

development mixed with bigger projects to develop new force controlled interaction between the operator and the IAD
product generations and new product concepts. (Colgate, Peshkin, & Klostermeyer, 2003). However, this kind of
Simultaneously with the customer- and application pull device does not give a real automation solution and the concept
there is of course a technology push that forces the robot cannot be developed towards a complete robotized final car
manufacturers to take advantage of the latest technology assembly. Instead, it could be expected that the automotive
development in other much bigger product segments. Thus, the industry will push for a new type of assembly robot systems and a
development of both hardware and software in the PC-area new type of infra structure for flexible automation.
makes a big impact on the robot controller development and In order to find other industries that might drive future robot
efforts are made to use also technology coming from the development, one possibility is to look at manufacturing
telecommunication area. The cost of software development and systems lacking the infrastructure needed for the efficient use of
maintenance is continuously increasing and a long software life robots today. One opportunity here is the need of automation in
time is therefore important, which means that, for example, new small and medium size enterprises (SMErobot, 2005). What is
efficient software development environments and new concepts needed in this environment is low cost safe robot systems that
for scalable system architectures, open interfaces and com- are easy to install, configure, calibrate, program and maintain.
munication concepts are important drivers of the robot The application processes will be the same as those that robots
controller development. Even if the fastest development in are used for today and the biggest challenge will be to develop
robotics is found on the controller side, there is also some robot technology that gives a much lower robot life cycle cost.
technology push on the electromechanical side. Examples are Going further in this development direction there will be a lot of
more efficient and bigger compact gear boxes, more cost other important areas for the future use of industrial robots.
efficient motors and drive systems, cheaper carbon composite Examples are flexible automation for disassembly (Steele,
material and more advanced tools for mechanical design. 2004) and sorting during scrap handling, for cutting up of meat,
In the long-term perspective, there will for sure be a for other types of food processing and handling (Fig. 6;
continued robot development for the automotive industry and Hamazawa, 1999), for consumer goods handling, for the
its sub suppliers. What could happen here is that increasing fuel assembly and processing of big components for airplanes
prices, pollution and other environmental problems will make it (KUKA Robotics, 2005), bridges, buildings, ships, trains,
necessary to build smaller and more efficient lightweight cars trains, power stations, windmills, etc., and for a wide spectrum
using, for example, electric drives in combination with fuel of craftsmen tasks. If economically feasible solutions for
cells. Then the question is how this will change the flexible robot automation can be found for these types of
requirements on the manufacturing systems including the applications a huge new market for industrial robots will drive
industrial robots. One possibility is that spot welding of the car the robot technology development in a new direction.
body will be replaced by, for example, laser welding and Looking at driving forces from future new application
soldering (Kobe, 2001) for metal parts and riveting, gluing and processes for robotics, enhancement of the robot performance
finishing of composite parts (Brosius, 2006). This could mean a will be very important. It is, for example, well known that arm-
bigger need of smaller and lighter robots with higher precision type robots are cheaper to manufacture and install than
and rigidity. If electric drives and fuel cells will be commonly Cartesian manipulators. However, the bandwidth, the stiffness
used, then new applications will arise, probably with higher and the accuracy of a Cartesian manipulator can be made much
requirements on robot accuracy and short cycle times. higher than for an articulated robot, but if the performance of
However, since it is still manufacturing of products in large articulated industrial robots is significantly increased these
volumes, probably made by the same industrial organisations robots could take over large market shares from the more
that make cars and car components today, the infra structure of expensive Cartesian manipulators. Examples of applications
the manufacturing system will probably be the same. where arm-type industrial robots could then be used are high
As pointed out in Section 1, one part of the car performance laser cutting, plasma cutting assembly and
manufacturing that is still labour intensive is the final assembly. machining. One example where both improved robot perfor-
Using the robot technology of today for this application, mance and a new flexible automation concept is needed is
problems arise with the geometric complexity and variability of fettling of iron castings (Lauwers, Wallis, Haigh, & Sohald,
the work objects, with the space needed for safety fences, with 2004). In this case, a third aspect is also important and that is the
difficult robot programming and with time consuming handling very unhealthy foundry environment. Having the health
of assembly failures. Only few robot installations can be problems of the workers in mind, several other future driving
economically motivated, as, for example, for the gluing and applications for industrial robots can be found. Examples of this
mounting of car windows. In this special case, the health risk is can be found in slaughterhouses, cold stores, glazier work-
one reason for the introduction of robots. The working shops, fisheries and garbage handling plants.
conditions are also the incitement for the introduction of
equipment to assist at heavy lifts. Here, a development has been 4. Scenarios about future industrial robot control
running for some years to develop lifting tools from being development
passive mechanical systems to be powered and servo
controlled. These intelligent assistant devices (IAD) are Even if it is very difficult to predict the long-term
interesting with respect to their safety requirements and the development directions, some scenarios will be outlined to
 T. Brogårdh / Annual Reviews in Control 31 (2007) 69–79 75

Fig. 6. The food industry will be one of the drivers for the future robot development. Here, the parallel kinematic robot FlexPicker, based on the Delta concept, works
in a bakery.

give an idea about how to make an extrapolation based on the 2001) to measure arm vibrations. These distributed sensors
driving forces that can be observed for the use of robotics. Since could then be used by the servo in a sensor fusion manner based
the automotive industry is the major force driving the robotic on the real time models running in the controller. Important in
development today it could be relevant to start looking at the this scenario will of course be to have accurate dynamic
future of car manufacturing with a scenario based on the earlier models. An industrial robot is a strongly coupled multivariable
proposed need of more accurate, rigid, fast and slender robots system with up to 50 mass-spring elements to be modelled
for such processes as laser welding, soldering, assembly, together with the non-linearities originating from friction and
riveting and gluing to replace the large number of big heavy lost motion in gears and other transmission components.
robots used for spot welding. Assuming the use of fibre optics A lightweight slender robot concept could also be useful for
and improved Yttrium Aluminium Garnet (YAG) laser tasks in the final assembly lines of automobile plants. The
technology (Rooks, 2000) and lightweight soldering, riveting scenario in this case must, however, also include the possibility
and gluing equipment, the weight of the load carried around by to make intuitive programming and efficient failure handling in
the robot will be radically reduced from that of a spot welding direct contact between human and robot, otherwise the
gun. This means that it is not far away to look at an
implementation scenario for the development of lightweight
slender robots with wrist- and upper arm concepts having much
lower mass using integrated actuator solutions, fibre compo-
sites and other lightweight materials. Even if a lightweight high
speed motor together with a lightweight high ratio speed
reducer will probably be more expensive than the heavier wrist
drive systems used today, the robot installations will probably
be cheaper since the main axes can be equipped with actuators
having lower power and since less massive frames can be used
for the mounting of the robots. In order to make a slender
lightweight robot stiff with respect to tool forces, the robot
control may need to provide a virtual stiffness, for which
sensors are needed in the arm structure of the robot. Examples
of sensors are capacitive encoders (Fig. 7) with optional
vibration measurement electrodes to measure the angle and the
vibrations of the output shafts of speed reducers, joint torque
sensors (Pfeffer, Khatib, & Hake, 1989), accelerometers (de Fig. 7. Capacitive encoder rings (Hexagon Metrology) integrated into an ABB
Jager, 1994) and strain gauges (Feliu, Garcı́a, & Somolinos, robot joint for accurate measurements of robot arm movements.
76 T. Brogårdh / Annual Reviews in Control 31 (2007) 69–79

automation system will be too complex. Therefore, full safety is load weights and for different performance requirements. The
needed and a low moving mass and compliant control of the implication of this is that also the model-based robot control
robot is important in this respect (DLR, 2006; Hirzinger, Albu- must be modular, including modular kinematics and dynamic
Schäffer, Hähnle, Schaefer, & Sporer, 2001; Ogawa, Haniya, models. To obtain optimal control for all possible module
Okahisa, & Ichibangase, 2005; Zinn, Khatib, Roth, & configurations, automatic servo- and model tuning is needed for
Salisbury, 2004). In order to obtain a high safety level, it is each configuration of the robot modules. Since such a highly
not far away to make use of the sensors needed for virtual modular robot concept will increase the difficulties at
stiffness control also for safety purpose. This could be done by configuration, re-configuration and maintenance, it will be
using the redundant measurement signals to generate residuals important to hide the complexity for the user. This means that
and to make redundant supervision of the robot control. If also a functionalities must be developed for Plug and Play and
six DOF force/torque sensor or joint torque sensors are used for intuitive human–robot interaction throughout the lifecycle of
human–robot interaction, further redundant supervision can be the robots. What this could imply is that robot CAD systems
introduced. need to support robot cell commissioning and design using
A robot concept for flexible automation of the final assembly virtual environment with configuration tools handling Plug and
of cars could also be used for the assembly of other complex Play components for robot- and cell modules (Gauss, Dai,
high volume products, to be found, for example, in the home Zimmermann, Som, & Wörn, 2006). When the selected
appliance manufacturing industries. It could also be used for the modules are delivered to the SME site the robot controller
automation of assembly of products with smaller lot sizes and should perform automatic cell configuration supported by Plug
then the concept has some resemblance with the robot assistant and Play to make the installation of mechanical, electrical and
concepts found in academic research today (Hägele, Schaaf, & software modules very simple. At a re-installation of the robot
Helms, 2002). However, in order to be able to obtain an in another application the same configuration tools can be used
economically feasible solution the robots in the mentioned again, making it easy for an SME to adapt the robot-based
assembly applications cannot work as an assistant to the production to new products and new customer requirements. In
operator but instead the operator should assist the robots to this scenario, new concepts are also needed to be able to support
solve unpredictable problems. The robot installations should easy to use intuitive robot- and cell calibration and robot
then be so robust that one operator could serve several robots. In programming. One component in such a solution could be the
this respect, it is very important to have a good balance between use of multi modal robot–human interaction including lead
the autonomy of the robot control and the tasks for the operator. through (Fig. 8) for the positioning of the robot.
For example, the robot could have some basic failure recovery One interesting possibility when building a modular robot
schemes but infrequent difficult to handle failure situations family is to use parallel kinematics. Actually, this could be
should be handled by the operator by direct interaction with the realistic using recently reported parallel kinematic structures
robot. An interesting aspect of developing robot control for with a small footprint in relation to the work space as for serial
such human robot collaboration is that the software architecture robots (Fig. 9; Cui, Zhu, Gan, & Brogårdh, 2005; Williams,
in industrial robot controllers of today will not be optimal and Hovland, & Brogårdh, 2006). Since the six parallel links in
neither the software architectures for autonomous systems. these new structures only need to transmit axial forces and no
Thus, new efficient control architectures are needed and the use bending or twisting torques, it is easy to build lightweight
of a combination of industrial and academic experiences is robots with very high accuracy and stiffness. These properties
probably necessary to develop high performance scalable are important in, for example, the fettling applications
software architectures for these new robot concepts. mentioned earlier and since high performance combined with
A robot that is able to work automatically according to a low inertia can easily be obtained also for a large work space,
program generated, supervised, edited and recovered by direct the new structures will be useful, for example, for the
physical interaction between human and robot will of course processing and assembly of aerospace components and for the
also solve many of the problems encountered when introducing processing of other large construction parts as for wind mills,
robots in small and medium size enterprises (SMErobot, 2005). bridges, buildings, railroads, etc., as mentioned before.
However, these companies usually have much smaller lot sizes Moreover, because the mechanical robot structure is not
than the automotive industry and they represent a much larger redundant, it is very easy to assemble this type of robots without
spectrum of applications and automation environments than any requirements on mounting accuracy. This makes these
found in the automotive final assembly lines. Moreover, their structures ideal for Plug and Play and it could even be realistic
economy is much more sensitive to big investments and they do to assemble and re-assemble the robots at the customer site
not have the access to the automation and robot expertise found without any advanced mounting equipment.
in the organisation of an automotive manufacturer. The In future robot applications with very complex work objects
question then is if it is possible to develop a robot family at as in disassembly, sorting, cleaning, cutting up etc. the robots
the necessary price level for all the small and medium sized must be supported by advanced vision systems. Probably, 3D
enterprises (SME) applications and requirements. One pre combined with colour and texture recordings (Suppa &
requisite for success in this case could be to find a modular Hirzinger, 2005) is needed for robust recognition- and
robot program, which means that industrial robots will be measurements on the object structures that will be processed
possible to assemble for different work space sizes, for different and the vision systems could need assistance from other sensor
 T. Brogårdh / Annual Reviews in Control 31 (2007) 69–79 77

Fig. 8. Lead through programming is expected to be an important technique in order to make robot programming easier. Here lead through programming is performed
using a measurement arm connected to the ABB TeachSaver robot programming software.

types to measure not visible characteristics of the work objects.

Because of the huge variability in object features and
processing rules efficient data bases must be built to model
the relations between object characteristics and robot tasks.
These process models should be possible to build using the
craftsmen’s experience. One scenario here is that the teaching is
made by the craftsmen themselves by demonstrations on real
objects using lead through of the robot while vision systems and
other sensors collect object features and the robot controller
updates data bases or other repositories with the rules that will
be used for the robot task execution. In order to make use of
rules and data from other similar installations the robot
controllers could be embedded in networks to make use of
experience and process models generated on other manufactur-
ing sites. Such system support together with the direct
interaction between craftsman and robot could give the
possibilities to make future robotics as flexible and easy to
use as necessary for the introduction of flexible automation in
applications, where manual work force is the only economic-
ally realistic solution today.
Beside the more exotic scenarios described above there will
of course also be an ongoing development of the robot
technology of today. This means further optimisation of the
performance/price ratio both with respect to the manufacturing
cost of the robots and the life time costs of the robot
installations. As discussed earlier robot control is very
important in order to get as much performance as possible
Fig. 9. A new parallel kinematic robot, developed by ABB, with a work space out of a robot structure. Model-based control using accurate
as a Scara robot but with performance characteristics as a Delta robot. models from system identification of each individual robot will
78 T. Brogårdh / Annual Reviews in Control 31 (2007) 69–79

give the future possibilities to push the performance/price ratio References

further, especially if sensors are introduced in the arm structure
of the robot. However, even with the most accurate control the ABB-2. (2001). IRB7600 power robot. http://www.rimrockcorp.com/products/
robots/7600/brochures/default.htm.
robot performance cannot surpass the fatigue limits of the robot ABB-1. (2003). ABB’s industrial solutions for the automotive industry. http://
structure and its mechanical components and the torque-, www.abb.com/global/seitp/seitp161.nsf/99ad595c32e0c2d9c12566e1000a
current- and temperature limits of the drive system. Thus, to go 4540/27b64a329396d49bc1256bf8003588c5/$FILE/Automotive%20
further in the optimal use of industrial robots these limits must Industrial%20IT%20brochure.pdf.
be tuned to the tasks of the individual robots. One scenario for ABB-3. (2004). Product specification pickmaster. http://www.abb.nl/global/
nlabb/nlabb034.nsf/99ad595c32e0c2d9c12566e1000a4540/bbce2cfac
this is to introduce adaptive robot performance. This means that
f64e51ec1256f5b004835fd/$FILE/ProductSpecificationPickMaster2_
the controller automatically tunes the drive system parameters 30.pdf.
to optimise the robot performance for the robot programs that ABB-4. (2006). Robot application builder. http://www.abb.com/global/abbzh/
are running. For this thermal and mechanical fatigue models abbzh251.nsf!OpenDatabase&db=/global/seitp/seitp327.nsf&v=9AA
must be executed together with the dynamic robot models in C910011&e=us&m=9F2&c=4A08ADF3B8BD7976C12570FC00348B3F.
ATI, Jr3, AMTI, & Fanuc (2006). Links to robot force sensor manufacturers.
real time to estimate temperature and mechanical stress in
http://www.ati-ia.com/, http://www.jr3.com/, http://www.amti.biz/, http://
critical components and structures. This adaptivity will result in www.fanucrobotics.com/file_repository/fanucmain/Intelligent%20Robots
better use of the installed robot and moreover it will make the %20with%20Force%20Sensors.pdf.
robot design more efficient since it will not be critically Braintech-ref. (2006). Vision guided robotic software & solutions. http://
dependent on worst case movements. www.braintech.com/index.html.
Bredin, C. (2005). Team-mates, ABB MultiMove functionality heralds a new
era in robot applications. ABB Review 1 http://www.abb.co.uk/global/
5. Conclusions abbzh/abbzh251.nsf!OpenDatabase&db=/global/gad/gad02077.nsf&v=96
DE&e=us&c=A86EAFECBB07B12AC1256FBE0055E7CE.
Industrial robot development has for sure not reached its Brosius, D. (February 2006). SMC makes a big comeback in production models,
limits and there is still a lot of work to be done to bridge the gap while carbon fibre sustains interest in high-end and aftermarket applica-
tions. Detroit auto show a showcase for new composites applications. http://
between academic research and industrial development and to
www.compositesworld.com/ct/issues/2006/February/1179/2.
intensify academic research in directions that target new Colgate, J. E., Peshkin, M., & Klostermeyer, S. H. (2003). Intelligent assist
applications and new flexible automation concepts. When devices in industrial applications: A review. IROS 2003. http://othello.-
developing robot control for real industrial use a lot of mech.northwestern.edu/peshkin/pubs/2003_IROS_IADsReview.pdf.
unforeseen problems arise and many of these problems need Cui, H., Zhu, Z., Gan, Z., & Brogårdh, T. (2005). Kinematic analysis and error
modeling of TAU parallel robot. Robotics and Computer-Integrated Man-
applied research to be solved. It is then very important with a
ufacturing, 21(6), 497–505.
close collaboration between researchers, industrial robot d’Angelo, P., & Corke, P. (2001). A robot interface using WAP phone. In
developers, automation system builders and robot users. Some Proceedings of Australian conference on robotics and automationhttp://
research- and development tasks in the robot control area for www.araa.asn.au/acra/acra2001/Papers/dAngelo.pdf.
such collaboration have been outlined in this presentation and de Jager, A. G. (1994). Acceleration assisted tracking control. IEEE Control
Systems Magazine, 14(5), 20–27.
the following is a short summary of the R&D directions that
DLR. (2006). Light-weight robots. http://www.dlr.de/rm/en/desktopdefault.
have been discussed: aspx/tabid-417/.
ESALAN Systems. (2006). Programmable electronic systems with safety
- Sensor-based robot control for safe interaction between function. http://www.elan.de/tweb/gb/2sicherheitsgerichtet/downloads/
human and robot and simultaneously for high performance ESALAN_systems.pdf.
Feliu, V., Garcı́a, A., & Somolinos, J. A. (2001). Gauge-based tip position
control of low cost robot mechanics.
control of a new three-degree-of-freedom flexible robot. Robotics Research,
- Sensor-based human–robot interfaces for intuitive robot 20(8), 660–675.
programming and cell calibration. Frey, J. -E., Endresen, J., Kreitz, A., & Scheibe, G. (2005). Unplugged but
- Efficient scalable software architectures for interactive connected. ABB Review 4 http://www.abb.it/global/abbzh/abbzh251.nsf!
robotics. OpenDatabase&db=/global/abbzh/abbzh254.nsf&v=5952&e=us&c=699
BD5D4CB7E7706C1256DAB003085DA.
- Tools to be used in robot installations for automatic Gauss, M., Dai, F., Zimmermann, U. E., Som, F., & Wörn, H. (2006). A standard
identification and tuning of model-based robot control communication interface for industrial robots and processor based periph-
parameters, especially for highly modular robot mechanics. erals—XIRP. In Proceedings of the 37th ISR2006 (International Symposium
- De facto standards and easy to use tools for planning, on Robotics).
optimisation, configuring, calibration, programming and re- Hägele, M., Schaaf, W., & Helms, E. (2002). Robot assistants at manual
workplaces: Effective co-operation and safety aspects. Proceedings of
configuring of robot automation systems. Important concepts
the 33rd ISR2002. http://www.morpha.de/download/publications/IPA_
could be Plug and Play for both virtual and real components Final_Manuscript_ISR_2002.pdf.
and knowledge databases for process- and automation Hamazawa, M. (1999). Robot applications in the Japanese food industry.
deployment. Industrial Robot 26(4).
- Further integration of process control into the robot controller, Hirzinger, G., Albu-Schäffer, A., Hähnle, M., Schaefer, I., & Sporer, N.
(2001). On a new generation of torque controlled lightweight robots.
especially in applications where force/torque sensors and 3D
IEEE International Conference of Robotics and Automation, 3356–
vision will be used. 3363.
- Extending model-based control with features for adaptive Katzel, J. (October 2004). HMIs in robotics advanced systems bring new
robot performance. dimensions in data display, portability, integration, and safety to robots
 T. Brogårdh / Annual Reviews in Control 31 (2007) 69–79 79

in manufacturing. Control Engineering, Cover Story. http://www.manufac- Sciavicco, L., & Siciliano, B. (2000). Modelling and control of robot manip-
turing.net/ctl/article/CA469968.html. ulators. London: Springer. http://www.springer.com/east/home/engineer-
Kobe, G. (August 2001). Manufacturing 2010—Robotics used in auto industry. ing?SGWID=5-175-22-2093898-0.
Automotive Industries. http://www.findarticles.com/p/articles/mi_m3012/ Siciliano, B., & Villani, L. (2000). Robot force control. The international series
is_8_181/ai_79007331. in engineering and computer science, Vol. 540. http://www.springer.com/
KUKA Robotics. (2005). Companies team to create new production methodol- east/home/engineering?SGWID=5-175-22-33356734-0.
ogy for aerospace manufacturing. http://www.roboticsonline.com/public/ Skarin, D., Claesson, B., Bergling, G. (2004). Virtual arc transforms welding
articles/details.cfm?id=2039. from ‘black art’ to practical science. ABB Review 3 http://library.abb.com/
Lauwers, B., Wallis, R., Haigh, P., & Sohald, S. (2004). Development of robot GLOBAL/SCOT/scot271.nsf/VerityDisplay/BB77D94391BF547FC1256F
based fettling cell for castings in low series. News from Castings Technology 0900462245/$File/41-44%203M876.pdf.
International. http://www.castingstechnology.com/secure/securedownloa- SMErobot. (2005). The European robot initiative for strengthening the com-
d.asp?dID=378. petitiveness in SMEs in manufacturing. Integrated Project in EU Sixth
Lippiello, V., Siciliano, B., & Villani, L. (2005). An experimental setup for Framework Programme. http://www.smerobot.org/.
visual servoing applications on an industrial robotic cell. IEEE/ASME Steele, J.W. (2004). Strategies for the disassembly of post-consumer automo-
international conference on advanced intelligent mechatronics. biles. North Carolina Division of Pollution Prevention and Environmental
Mattone, R., & De Luca, A. (2003). Fault detection and isolation in robot Assistance. http://www.p2pays.org/ref/04/03340.pdf.
manipulators. IFATIS report IRAR002R01. Dynamic fault tolerance frame- Suppa, M., & Hirzinger, G. (2005). Ein Hand-Auge-System zur multisensor-
work for remote robots. IEEE Transactions on Robotics and Automation. iellen Rekonstruktion von 3D-Modellen in der Robotik (an eye-in-hand
Ogawa, M., Haniya, K., Okahisa, M., & Ichibangase, A. (2005). New applica- system for multisensor surface reconstruction of 3d-models in robotics).
tion with new robot. Proceedings of the 36th ISR2005. Automatisierungstechnik, 53(7), 322–331.
Ostring, M. (2002). Identification, diagnosis and control of flexible robot arm. UNECE, The United Nations Economic Commission for Europe. (2004). World
Thesis No. 948. Department of Electrical Engineering, University of Robotics 2004 (and 2006). http://www.unece.org/press/pr2004/04robots_
Linköping. http://www.control.isy.liu.se/research/reports/LicentiateThesis/ index.htm.
Lic948.pdf. Watanabe, A., Kazunori, B., Warashina, F., & Kumiya, H. (2005). Practical bin
Pettersson, M. (2005) On drive train optimization in industrial robot design. picking by the intelligent robot. ISR2005.
Thesis No. 1226. Division of Machine Design, Department of Mechanical Williams, I., Hovland, G., & Brogårdh, T. (May 2006). Kinematic
Engineering, University of Linköping. error calibration of the Gantry-Tau parallel manipulator. Proce-
Pfeffer, L., Khatib, O., & Hake, J. (1989). Joint torque sensory feedback in the edings of the IEEE International Conference on Robotics and Automa-
control of a PUMA manipulator. IEEE Transactions on Robotics and tion.
Automation, 5(4), 418–425. Zhang, H., Zhongxue, G., Brogårdh, T., Wang, J., & Isaksson, M. (2004). Robotics
Pires, J. N., Ramming, J., Rauch, S., & Araújo, R. (2002). Force/torque sensing technology in automotive powertrain assembly. ABB Review 1. http://
applied to industrial robotic deburring. Sensor Review Journal, 22(3), 232– www.abb.com/global/abbzh/abbzh251.nsf!OpenDatabase&db=/global/gad/
241http://robotics.dem.uc.pt/norberto/jr3pci/ft_sensors_III.pdf. gad02077.nsf&v=3B5E&e=us&c=468E728B1CA2767CC1256E
RIA and NIST. (2000). Open architecture control for the robotics industry. 3C00330D31.
Workshop Orlando. Florida, February 9–10. http://www.isd.mel.nist.gov/ Zinn, M., Khatib, O., Roth, B., & Salisbury, J. K. (2004). A new actuation
projects/openarch/Orlando_Workshop/RIA_Workshop.pdf. approach for human-friendly robot design. International Journal of
Robot Manufacturer links (with respect to robot applications). (2006). http:// Robotics Research, 23(4/5), 379–398.
www.motoman.co.uk/applications.htm, http://www.fanucrobotics.com/
index.asp?item=993&site=23, http://www.abb.com/robotics, http:// Torgny Brogårdh is company senior specialist in mechatronics at
www.kuka. com/germany/en/solutions/solutions_search/, http://www. ABB Robotics. He got his PhD 1975 for Prof. Hellmuth Hertz in instru-
reisrobotics.com/us/us/Welding_technique.html, http://www.comau.com/ mentation technology at the University of Lund and became assistant
page-bu-robotics.asp?INSIDEID=1025&INSIDEMODULO=PAGINE&ID professor there in 1976. Between 1982 and 1986, he was associated
MENU=1021&PATHMENU=%7CProducts+%26+Services%7C. professor at the Royal Institute of Technology in Stockholm. From 1976
RRS. (2006). http://www.realistic-robot-simulation.org/. to 1986, he worked at the ASEA Central Research Laboratory, where he
Rooks, B. (2000). Lasers become the acceptable face of precision welding and developed new industrial measurement methods and a new fibre optic
cutting. Industrial Robot, 27(2), 103–108. measurement technology. In 1986, he went to ABB Robotics, where he
Servorobot. (2006). http://www.servorobot.com/page.asp?no_page=62&o= works with research in motion control, mechatronic design, sensors and
5&o2=1&pp=4&pp2=62&lg=1. parallel kinematics. He is a member of the research board CENIIT at the
Schraft, R. D., Meyer, C., Parlitz, C., & Helms, E. (2005). PowerMate—a safe University of Linköping and a member of the Swedish Foundation of
and intuitive robot assistant for handling and assembly tasks. IEEE Robotics Strategic Research. In 1998, he got the Swedish Royal Academy of
and Automation Society: IEEE international conference on robotics and Engineering gold medal for his contributions to the robotics development
automation (ICRA) (pp. 4085–4090). at ABB.