Tactile Sensing in Intelligent Robotic ManipulationA Review
Johan Tegin, Jan Wikander Department of Machine Design, Royal Institute of Technology Stockholm, Sweden johant@md.kth.se, jan@md.kth.se

Abstract

A tactile sense is key to advanced robotic grasping and manipulation. By touching an object it is possible to measure contact properties such as contact forces, torques, and contact position. From these, we can estimate object properties such as geometry, stiffness, and surface condition. This information can then be used to control grasping or manipulation, to detect slip, and also to create or improve object models. This paper presents an overview of tactile sensing in intelligent robotic manipulation. The history, the common issues, and applications are reviewed. Sensor performance is briey discussed and compared to the human tactile sense. Advantages and disadvantages of the most common sensor approaches are discussed. Some examples are given of sensors widely available today. Eventually the state of the art in applying tactile sensing experimentally is presented.

1 Introduction

To pour and bring a glass full of milk is an almost trivial task for a human adult. For a robot on the other hand, it is a challenging task. Nonetheless, for robots to be useful as service agents at home, in hard to reach places, hazardous areas, outer space, and elsewhere, they need to be mobile and able to grasp, move, and manipulate objects in their surroundings. Usually, we dene two different kinds of grasps; power grasps and precision grasps. Power grasps are typically used for larger objects and in tasks that do not require more than simple manipulation of the object. Grasping a chair to lift it would be an example. Signicant for a power grasp is that the object is xtured by the palm and as much nger area as possible. More delicate objects are typically held in a precision grasp. When lifting a glass and in other precision tasks, primarily the ngertips are used for contact. The precision grasp has advantages such as enabling better control of contact forces, but it is also typically less stable than the power grasp. To perform more complex tasks such as rotating the object, as in actually pouring something in the glass, we need
to be able to control the motion of the object. Using a power grasp, this can be done by moving the arm and wrist. In a precision grasp, this can be done by moving only the ngers. The objective in using tactile sensing in robotic grasping and manipulation is to extend the possibilities beyond those resulting from using other sensor modalities only. Usually the limiting factor is lack of information (and knowledge of how to use it). Only by touch can we collect information from the very point of contact. Lee and Nicholls [1] dene a tactile sensor as a device that can measure a given property of an object or contact event through physical contact between sensor and object. In robotic manipulation we are primarily interested in the mechanical properties of the contact. Other properties, such as temperature and moistness, will not be covered in this paper. First, we need to know whether we are in contact with the object or not. Seemingly simple, but we can already here see why visual information with a comparatively low resolution, and which also can be occluded needs to be complemented with more detailed contact information. Second, tactile information can inform on the contact conguration. We can sense forces and torques and from that nd out what type of contact it is. A surface, an edge, or a point contact will all give different pressure patterns in a sensor matrix. The vibrations that slip in a contact will cause, can be detected. Then there is haptic exploration; by moving the nger, we can extract data from several points and use that data to create or improve an object model. From that contact information it is possible to derive information regarding the objects inherent properties: Is the object stiff or is it compliant? Is there any texture? What is the friction coefcient? All this information can be used to improve the object model and to improve manipulation quality. Finally, contact data can be used as feedback for control. For complicated tasks in uncertain and dynamic environments, open loop control will not sufce. Knowing the contact properties and whether contact has occurred is essential in work with contact transition issues. The feedback

can be used to control force and torque at a specic contact location, something necessary for manipulating objects and to control slippage. As can be seen, we need a tactile sense for intelligent robotic manipulation. This paper will give an overview over the most common issues and topics in grasping and manipulation. To put things in perspective, a historical overview is presented followed by a section on likely applications. After that, a review of sensing issues starting with a reection on the human tactile sense and how it relates to robotics. Partly because we can use it as benchmark, partly because the human sensory system and manipulation capacity is both a grand challenge to mimic and also a great source of inspiration. A review of some sensor hardware follows and eventually some examples of experimental results.
researchers in visual recognition or tracking can buy out of the box cameras [1], there is to our knowledge no widely available tactile sensor suite for robotic manipulation. There are a few robotic hands out there, some of which are commercial products. But they are then rather costly. Today, there is clearly a need for qualied manipulation hardware together with more advanced methods for actually using the tactile information.

3 Applications

Today, much attention is given to tactile sensing in minimally invasive surgery, keyhole surgery. The case is that much of the tactile information available in open surgery, will now be lost. Articial tactile sensing can restore some of this loss of tactile information. Eltaib and Hewit [9] give an overview of tactile sensing in minimally invasive surgery MIS. But MIS still involves humans in the feedback loop and hence does not cover all needs for performing intelligent robotic manipulation. Outside the laboratory, manipulation is still primarily performed without tactile sensing. In an industrial setting, most variables can be controlled. Even though force/torque sensors are used for grinding operations and for peg-in-hole tasks, the really large benets with a rened tactile sense can be reaped outside such well controlled environments. For a versatile robot in an uncertain environment, tactile sensing will open up new possibilities. Recently we have seen some impressive humanoids, if they cannot be made to interact with their surrounding, they will be of little use. Combining these humanoids with advanced grasping and manipulation capabilities, has for long been a dream that opens up possibilities limited only by imagination. Such robots could be used pretty much anyplace in which it can be cumbersome or dangerous to use humans; as 24-hour household help, for re-ghting, in deep space missions or for ABC warfare clean-up. But even without advanced humanoids, there are certain applications for which tactile sensing can be of great use and that also stand out as the more likely to be the rst to apply it. Perhaps the most common justication for developing robots for autonomous grasping and manipulation is the demographic situation. Within a few decades, many industrialized countries will face a signicant increase of elderly people. A robot could for example be used to perform simple fetch and manipulation tasks. Picking up something that has been dropped, open the door, and work as an interface to other machines around the home are some examples of what such a robot could do. The challenge is to make these robots safe, useful, versatile, user friendly, reliable, and while perhaps not cheap, at least give a good value for the money. In a dynamic and uncertain environment, such as in a home not particularly modied for robot access, this becomes a truly challenging task.

2 Historical overview

In the 1970s, tactile sensing for robotic applications emerged as its own eld of research. Research kept increasing during the next decade and led to one of the rst overview papers on tactile sensing by Harmon in 1984 [2]. The 1980s was also time for the rst advanced robotic grippers such as the DLR ROTEX gripper [3]. It is equipped with laser range nders, tactile arrays, force/torque sensors, integrated actuator, and also analog and digital electronics for communication over a serial bus, all neatly packaged in one single device. In some areas, there has been great progress since Harmons article from 1984. But to a large extent, the problems today remain similar to what they were back then. He foresaw a rapid expansion within automation. This however, has not been the case. Lee addressed this very fact in 2000 [4] when he pointed out that in structured environments, we are able to come very far without tactile information. Hence, he foresees that tactile sensing will be most useful in unstructured environments where object properties and/or the environment is not fully known. The review paper of Lee and Nicholls [1] gives an overview of tactile sensing in mechatronics up until 1998. The yearly number of published articles had steadily increased, an increase that has continued since then and which is exemplied by the development of novel tactile arrays [5] manufactured in a silicon processes, that control is evolving [6], and that system integration is taken to a new level [7]. The focus has somewhat shifted over the years; from the development of new tactile sensing technologies towards data processing. At the same time, strong theoretical models of contact congurations and grasp dynamics have been developed [8]. However, when it comes to the application of such models in association with tactile sensors, only little work has been done. Tactile sensing for robotics is still in its infancy. While
Toys have previously been an underestimated market and may very well turn out to be the rst robotic manipulators with a tactile sense to hit the market. Further, Lee [4] considers natural product processing, such as agriculture and food processing, as yet another probable application area. He also gives a nice overview of the features, needs, issues, and challenges in the applications above.

Compared to a robot, humans also respond to sensory information with a large latency. For the fastest reexes we see latencies of 20-30 ms and much longer times for voluntary responses [10].
4.2 Passive and Active Sensing
Sensing can be divided in many ways. But one of the more important dividing lines is that between passive and active sensing. Passive sensing concerns the analysis of static tactile data, whereas active sensing is when motion is actively used to extract more information. Okamura and Cutkosky [6] write Some types of features, particularly small ones, cannot be sensed accurately through static touch; motion is required. An example is detecting edge sharpness. Placing a nger on the edge will give only little information regarding its sharpness compared to sliding it across the edge. A dextrous hand can also actively manipulate its environment to retrieve information on properties impossible to estimate in other ways. For example, by tilting an object weight or center of gravity can be estimated. By dragging a nger along a surface, friction and texture can be approximated, and so on.

4 Sensing

4.1 Human Sensing
The information quality needed to perform certain robotic manipulation and grasping tasks still remains unknown [1]. Neither is it known exactly how humans manipulate objects. Even if we did know, it is not certain that an anthropomorphic approach would be the best. But there is still much to be learned from what is known about the human sensory system. A condensed overview of human sensing is presented by Howe [10]. He puts tactile sensing in perspective from a human mechanoreceptor viewpoint. To fulll all robotic tactile sensory needs with a single type of sensor is difcult, if not impossible. This is a problem that we humans also have. To overcome it we are equipped with different types of receptors. The fast adapting (FA) mechanoreceptors can sense vibrations but not static stimulation, whereas the slowly adapting (SA) mechanoreceptors respond to static stimuli. The different properties of human cutaneous mechanoreceptors are summarized in table 1. Table 1: Different human mechanoreceptors. From [10]. Receptor Type FA I SA I FA II SA II Field Diameter 34 mm 34 mm >20 mm >10 mm Frequency Range 1060 Hz DC30 Hz 501000 Hz DC15 Hz Postulated Sensed Parameter Skin stretch Compressive stress (curvature) Vibration Directional skin stretch

5 Sensors

5.1 Sensor Specication
There is of course no single sensor that excels with respect to all design criteria. The sensor specication will have to depend upon the task at hand. For this reason, a very large amount of tactile sensing technologies have been developed. But some criteria that always must be considered are [1]: Variables and measurable range pressure, shear forces, torques, slip, etc Resolution in space Response prole accuracy, bandwidth, hysteresis, creep, aging, etc In addition to these, we should add some desirable properties such as simple mechanical integration, low power consumption, and low cost. Humans have a resulution in the ngertips of about 1 mm and are able to sense frequencies close to 1 kHz [10]. A similar specication is often proposed for distributed tactile sensors, for example a spatial resolution of 1-2 mm and a frequency span up to at least 100 Hz [11].

Without visual feedback, humans have a rather dim perception of the position of their limbs. This is because the human proprioception1 is poor, especially compared to what can be achieved in a robot. Even without visual feedback, a robot can thanks to high resolution encoders identify its position and orientation in space much more accurately than humans. This is an advantage that can be exploited, particularly in certain haptic exploration tasks [6].
1 Proprioception The ability to sense the position, location, orientation, and movement of the body and its parts.

5.2 Sensor types

Over the years, many tactile sensors have been proposed. Tactile information from a power grasp can be be collected using distributed sensors covering the phalanges. A ngertip is typically more roomy and allows for a more space
consuming force/torque sensor that can supply detailed information in the case of a precision grasp. The basic principles for a few different ngertip sensors can be seen in gure 1. An excellent overview of tactile sensing devices for manipulation was written by Howe [10]. Saad et al. also present an overview of current sensor technologies [12] although omitting image recognition sensors such as those developed by Ferrier, Hristu, and Brockett2 [25, 26]. Some argue in favor of a compliant ngertip. An issue for a stiff ngertip is that it is more prone to contact transition problems and it also offers poor grip. This can to a certain extent be overcome by covering the sensor with a soft material. But more important, a compliant ngertip not unlike the human is also advantageous from a pressure distribution and stability point of view [13, 14].
gauges is typically more accurate and is often designed to measure all six degrees of freedom. But there is also a larger mass, often the ngertip itself, between the object and the intrinsic sensor. This can be an a disadvantage when measuring small forces and it also makes the sensor sensitive to high accelerations. An additional downside is that when using a force/torque sensor inside the ngertip, we cannot tell the difference between multi-point and single-point contacts [17]. Nonetheless, intrinsic sensing is often used in manipulation tasks using a precision grasp. Using an intrinsic sensor it is possible to determine the contact location without measuring it explicitly. Data from a six DOF force/torque sensor inside the ngertip can be used to compute the point of contact [18]. Bicchi [19] mathematically discusses the issues in computing contact position based on intrinsic sensor data.

Figure 2: The ATI Nano 17 six DOF force/torque sensor. Courtesy of ATI Industrial Automation.

5.3 Extrinsic Sensors

The typical extrinsic sensor is a tactile array, not unlike a laptop touch-pad, or a single-point pressure sensor. It typically senses normal forces and contact positions. The sensors often display a measurable resistance change as a result from compression of a semi-conductive polymer.
Figure 1: Different ngertip sensors; a) Distributed extrinsic, b) Force/torque intrinsic, c) Fluid lled If we consider the very compliant ngertips (see section 5.4) as one sensor type, the remaining sensors can be divided into extrinsic sensors (see section 5.3) and intrinsic sensors. The intrinsic sensors measure forces within the grasping mechanism whereas the extrinsic sensor measures forces that act upon the mechanism. The predominant intrinsic tactile sensor is a small force/torque sensor mounted inside the ngertip [7, 15]. Extrinsic sensors are signicantly more diverse, covering different kinds of single point and distributed sensors. Examples include force sensitive arrays to be mounted on the ngertip and those where the sensor module itself constitutes the ngertip. The choice of whether to use extrinsic or intrinsic sensing depends upon the task at hand. The advantages of an extrinsic sensor include that they can be made to cover large areas as when using a power grasp and that the point of contact is explicitly measured. Most extrinsic sensors only measure pressure. But there are a few advanced extrinsic sensors capable of measuring shear forces on the tactile element level [5, 16]. An intrinsic sensor using strain
Figure 3: Force sensing resistors, one mounted on the ngertip and the other on the proximal phalanx on the nger of a Barrett hand. As an example, the Gifu hand [20, 21] is equipped with tactile sensors covering the phalanges of all ngers (35) and the palm with a grand total of 624 measurement points. The ngertips are not covered by the distributed sensor and instead feature six DOF force/torque sensors.

also section 5.4

An optical waveguide will show nearly no loss of light if properly designed. But when its boundary is affected by touch, frustrated total internal reection will occur. The intensity changes can be measured either by a position sensitive detector or through positioned optical bers and separate detectors [22, 23]. Other measurable properties include birefringency effects resulting from internal stress (photoelasticity) [12] and capacitivity changes resulting from compressing the insulator of a capacitor.

However, the other way around is more difcult and especially so in the presence of noise. Lee and Nicholls [1] gives an overview and Nowlin [28] discusses ill-posedness and ill-conditioning when trying to compute surface deformation from sensor data.
5.7 Widely Available Sensors
Today, there are many suppliers of sensors that can be used in tactile sensing for robotic manipulation. A few of them and some of their products are presented below as examples of what performance-wise well known hardware is widely available. Hopefully the number will steadily increase by the addition of complete integrated tactile systems, additional robotic hands for dextrous manipulation, drivers, and hard- and software needed to implement tactile sensing in real life applications. Interlink Electronics, Inc. supply force sensing resistors (FSR), both as single elements and arrays like the ones in gure 2. A matrix based sensing system is available from Tekscan, Inc. Pressure Prole Systems, Inc. market a tactile system featuring a capacitive array. (For physical reasons, such a capacity based system has a limited spatial resolution. The minimum tactile element spacing is 2 mm.) The companies mentioned above also supply other kinds of sensors, from single point sensors to conformable arrays. Capacitive and resistive sensors often feature creep, hysteresis, and/or poor long time stability. But they are typically easy to implement and sometimes offer an accuracy high enough to make them very useful. Nonetheless, strain gauge based sensors, such as force/torque sensors, typically perform better with respect to these issues. A miniature six DOF force/torque sensor for use in a ngertip (see gure3) is offered by ATI Industrial Automation, Inc. BL Autotec market a similar sensor. While not commercially available, the DLR sensor is anyway included in the summary in table 2. The force/torque sensors above are all equipped with overload protection. See table 2 for example specications. Table 2: Small Force and/or Torque Sensors Sensor BL Nano 5/4 ATI Nano 17 Bokam Aura/Supra FSR DLR F/T digital Max. Force [N] (Fz =70) 42 (Fz =153) varies 30 Max. Torque [Nm] 0.4 0.5 N/A N/A 0.15 (Mz =0.05) H [mm] [mm] >32.8 14.23 0.2 1.25 16
5.4 Highly Compliant Sensors
A ngertip exploiting the compliant and optical properties of closed cell polyurethane foam was recently developed by Hellard and Russell [24]. Also recently, the deformation of a ngertip membrane lled with a transparent liquid was measured using a camera [25, 26]. The picture data was then used to compute the displacement of the membrane and from that the object shape information was derived.

5.5 Data Processing

The amount of tactile data from many and large tactile arrays can be overwhelming, and even if it can be managed, just getting it to a processor may be difcult. Consider a 1616 tactile array; a minimum of 32 electrical wires are needed. Add to this that data from 256 sensors that need to be processed. As always, one has to prioritize, or try to come up with innovative designs. DLR deals with this by distributing the A/D-conversion and signal processing. An example of this is their six DOF ngertip sensor3 that features integrated electronics for signal processing. The sensor has a purely digital interface [7]. Other force/torque sensors typically deliver analog signals that require an external signal processing unit. Its also possible to perform the signal processing in the analog domain or in the sensor itself. Or simply be satised with not retrieving all information possible from a contact position. One example is the XYZ-pad [11] that only needs four electrical wires and still supply normal force and XYposition. Tactile patterns will be impossible to recognize, and the sensor will also only handle one contact point. But for certain applications, that may sufce. Yet another way to reduce the wiring is to use wireless communication. Optical communication from sensors embedded in a soft ngertip is proposed by Yamada et. al. [27].
The tactile inversion problem
When using a compliant sensor or ngertip, there is usually some elastic material or liquid between the deformed surface and the sensor array. Given the surface deformation, forward computation will give the sensor stresses.

also table 2.

6 Tactile Sensing Implemented
Even though there is a gap when it comes to actually manipulating objects using tactile information, as mentioned in section 2, more and more work is being done. Bicchi [29] presents an overview of grasping where he mentions that one of the most needed advances in robotic grasping is to estimate object compliance. Coelho et. al. [30] have developed models for grasp policies and grasp control and also veried them in simulation. Below are a few examples of real life experiments that until recently have been presented. By combining tactile sensing with vision, Hosada et. al. present a system that learns to detect slip from tactile sensor information [31]. Using information from an intrinsic sensor, Bicchi et. al. present a nice method to reduce the risk of slippage by controlling the normal force [32]. Laschi et. al present an anthropomorphic robotic grasping platform developed for evaluation of neurophysiological and other physiologically inspired theories such as biologically-inspired grasping coordination [33]. A neural approach to software development will be used. The DLR hand is one of the most rened robotic hands of today. Both with respect to mechanics and control. It has demonstrated the catching of a ball, playing the piano, and more [34]. They have also implemented impedance control essential to more autonomous tasks.

References

[1] M. H. Lee and H. R. Nicholls. Tactile sensing for mechatronics - a state of the art survey. Mechatronics, 9(1), October 1999. [2] L. D. Harmon. Automated touch sensing: A brief perspective and several new approaches. In IEEE Int. Conf. on Robotics and Automation, March 1984. [3] J. Dietrich, G. Hirzinger, J. Heindl, and J. Scott. Traditional and Non-Traditional Robotic Sensors, chapter Multisensory Telerobotic Techniques. SpringerVerlag, 1990. [4] M. H. Lee. Tactile sensing: New directions, new challenges. Int. J. of Robotics Research, 19(7), Jul 2000. [5] B. J. Kane, M. R. Cutkosky, and G. T. A. Kovacs. A traction stress sensor array for use in high-resolution robotic tactile imaging. IEEE J. of Microelectromechanical Systems, 9(4), Dec 2000. [6] A. M. Okamura and M. R. Cutkosky. Feature detection for haptic exploration with robotic ngers. The Int. J. of Robotics Research, 20(12):925938, 2001. [7] J. Butterfa, M. Grebenstein, H. Liu, and G. Hirzinger. DLR-Hand II: Next generation of a dextrous robot hand. In IEEE Int. Conf. on Robotics and Automation, Seoul, Korea, May 2001. [8] A. Bicchi and V. Kumar. Robotic grasping and contact: A review. In IEEE Int. Conf. on Robotics and Automation, April 2000. [9] M. E. H. Eltaib and J. R. Hewit. Tactile sensing technology for minimal access surgery - a review. Mechatronics, 13(10):11631177, Dec 2003. [10] R. D. Howe. Tactile sensing and control of robotic manipulation. Advanced Robotics, 8(3), 1994. [11] H. Liu, P. Meusel, and G. Hirzinger. A tactile sensing system for the dlr three-nger robot hand. In Int. Symp. on Measurement and Control in Robotics, May 1995. [12] R.E. Saad, A. Bonen, K.C. Smith, and R. Benhabib. Tactile Sensing, volume The Measurement, Instrumentation and Sensors Handbook, chapter 25. CRC Press LLC, 1999. [13] R. A. Russell and S. Parkinson. Sensing surface shape by touch. In IEEE Int. Conf. on Robotics and Automation, May 1993. [14] R. W. Brockett. Robotic hands with rheological surfaces. In IEEE Int. Conf. on Robotics and Automation, March 1985.

7 Conclusions

Although many sensor technologies and strong theoretical models have been developed, there is still much left to be done in intelligent grasping and manipulation. In particular, there is a gap in applying the theoretical models in association with tactile sensing. In the past, progress has been slow and it will likely stay that way for a while. But the applications, in humanoids and other applications, are closer than ever. The humanoids of today can walk and dance, but they can only perform very simple manipulation tasks. On the road towards more advanced manipulation lies many interesting challenges. From a technology standpoint, with more and more research being published, better hardware, more powerful computing, the current development in MEMS and wireless solutions, and an increased interest from commercial players and academia, the conditions for growth and advance are better than ever.

Acknowledgment

This work was supported by the Swedish Foundation for Strategic Research through the Centre for Autonomous Systems at the Royal Institute of Technology, Stockholm, Sweden.
[15] X. H. Gao et al. The HIT/DLR dextrous hand: Work in progress. In IEEE Int. Conf. on Robotics and Automation, September 2003. [16] Y. Yamada and M. R. Cutkosky. Tactile sensor with 3-axis force and vibration sensing functions and its application to detect rotational slip. In IEEE Int. Conf. on Robotics and Automation, May 1994. [17] A. M. Okamura, N. Smaby, and M. R. Cutkosky. An overview of dextrous manipulation. In IEEE Int. Conf. on Robotics and Automation, April 2000. [18] J. K. Salisbury. Interpretation of contact geometries from force measurements. In IEEE Int. Conf. on Robotics and Automation, March 1984. [19] A. Bicchi. Intrinsic contact sensing for soft ngers. In IEEE Int. Conf. on Robotics and Automation, May 1990. [20] H. Kawasaki, T. Komatsu, K. Uchiyama, and T. Kurimoto. Dexterous anthropomorphic robot hand with distributed tactile sensor: Gifu hand II. In IEEE Int. Conf. on Systems, Man and Cybernetics, 1999. [21] H. Kawasaki, T. Komatsu, and K. Uchiyama. Dexterous anthropomorphic robot hand with distributed tactile sensor: Gifu hand II. IEEE Trans. on Mechatronics, September 2002. [22] S. Begej. Planar and nger-shaped optical tactile sensors for robotic applications. IEEE J. of Robotics and Automation, 4(5), October 1988. [23] H. Maekawa, K. Tanie, and K. Komiriya. Tactile feedback for multingered dynamic grasping. IEEE Control Systems Magazine, February 1997. [24] G. Hellard and R. A. Russell. A robust, sensitive and economical tactile sensor for a robotic manipulator. In Australian Conf. on Robotics and Automation, November 2002. [25] D. Hristu, N. Ferrier, and R. W. Brockett. The performance of a deformable-membrane tactile sensor: basic results on geometrically-dened tasks. In IEEE Int. Conf. on Robotics and Automation, April 2000. [26] N. J. Ferrier and R. W. Brockett. Reconstructing the shape of a deformable membrane using image data. The Int. J. of Robotics Research, 19(9):795816, Sep 2000. [27] K. Yamada, K. Goto, Y. Nakajima, N. Koshida, and H. Shinoda. A sensor skin using wire-free tactile sensing elements based on optical connection. In SICE, August 2002.

[28] W. C. Nowlin. Experimental results on bayesian algorithms for inpterpreting compliant tactile sensing data. In IEEE Int. Conf. on Robotics and Automation, Apr 1991. [29] A. Bicchi. Hands for dexterous manipulation and robust grasping: A difcult road towards simplicity. IEEE Trans. on Robotics and Automation, 16(6), Dec 2000. [30] J. Coelho, J. Piater, and R. Grupen. Developing haptic and visual perceptual categories for reaching and grasping with a humanoid robot. Robotics and Autonomous Systems, 2001(37):195218, 2001. [31] K. Hosada, Y. Tada, and M. Asada. Internal representation of slip for a soft nger with vision and tactile sensors. In IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, October 2002. [32] A. Bicchi, J. K. Salisbury, and P. Dario. Augmentation of grasp robustness using intrinsic tactile sensing. In IEEE Int. Conf. on Robotics and Automation, May 1989. [33] C. Laschi, P. Gorce, J. Coronado, F. Leoni, G. Teti, N. Rezzoug, A. Guerrero-Gonzlez, J. L. Pedreo Molina, L. Zollo, E. Guglielmelli, P. Dario, and Y. Burnod. An antropomorphic robotic platform for experimental validation of biologicallyinspired sensory-motor co-ordination in grasping. In IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, 2002. [34] C. Borst, M. Fischer, S. Haidacher, H. Liu, and G Hirzinger. DLR hand II: Experiments and experiences with an antropomorphic hand. In IEEE Int. Conf. on Robotics and Automation, September 2003.

Measure

The bar is made of material with a low thermal expansion coefficient. It is placed in several positions on a rigid fixture 6 throughout the working envelope of the robot and provides a measure of the repeatability of the robots self-adjustment. The test is based on the norms used for classic CMM, such ISO 10360-2 and the ASME B89.4.1 as well as the robot performance standard ISO 9283.
The hole-bar test provides a measure of accuracy and repeatability of the robots self-adjustments.
Operational flexibility is also enhanced because: Different car models can be produced on a single line and new models easily added to running lines. Instant feedback is provided when tooling adjustments are made (necessary, for example, when a new model is introduced). Calibrated measurement points are easy to add using off-line programming. Standard robots can be used instead of those with a built-in absolute accuracy option. No more time-consuming and cumbersome correlation with CMM or Gold Model.

Conclusion

Implementation history
2001: First performances tests with IRB 4400 and Perceptron vision system. 2002: Validation of absolute accuracy performances with Dynalog. 2003: Industrial validation with Renault. 2004: First four industrial robots stations at Renault Flins plant.
ABBs robot based measurement system (RBMS) makes manufacturing accurate and repeatable workpieces from robots cells simpler and faster. Tolerance to all kinds of changes that have to be reckoned with on a realworld factory floor and which could otherwise degrade product quality is improved. And despite living in a sub-optimal real world, production to within 0.4 mm of the ideal CAD world is possible. ABB is making sure your car doors shut perfectly the first time and every time!

Volumetric accuracy test

To additionally observe accuracy and repeatability of this equipment, a volumetric accuracy test is used. This hole bar test uses a bar with ten equidistant holes and an ABB IRB 4400 robot carrying a Perceptron FlexiCam sensor.
Fabrice Legeleux ABB MC Beauchamp, France fabrice.legeleux@fr.abb.com

Learning skills

Robotics technology in automotive powertrain assembly
Hui Zhang, Zhongxue Gan, Torgny Brogard, Jianjun Wang, Mats Isaksson The past 40 years have seen industrial robots establish their superiority over humans in most areas of manufacturing requiring endurance or repeatability. One important application domain, however, has so far lagged behind the industrys expectations: mechanical assembly. As fast, precise and dependable as they are, traditional industrial robots just dont seem able to perform certain assembly operations as well as a skilled human worker.

The ABB solution

The technology was initially tested on IRB 4400 and IRB 6400 robots, with payloads ranging from 30 to 200 kg, used to assemble different automotive powertrain parts. These tasks, which can be considered complex, included inserting forward clutch hubs and the assembly of F/N torque converters 1. The tests demonstrated consistently superior performance in terms of cycle time, acceptable insertion force, reliability and ease of programming.
motion should be changed in a controlled manner so as not to further increase the contact force. Using the concept of maximum achievable passive admittance as a foundation, ABBs engineers constructed intelligent control methods that integrate seamlessly with existing advanced position control methods and guarantee stable, gentle contact in most common production environments. The design also ensures a smooth transition between the force control and position control modes.

Easy to program

To emulate human behavior, robots need to exhibit force-sensing capability and compliance, be able to direct forces in a controlled way and react to contact information.
While the literature on robot force control is formidable, the results have generally been less than impressive. Achieving acceptably fast robot movement while assuring contact stability has persisted as a challenge. Many promising intelligent-control methods have been investigated, but superimposing slow force control on position control has typically resulted in poor performance. The concept of impedance and admittance1) is also helpful in understanding force control in a robot. Along each degree of freedom the instantaneous power flow between two or more physical systems can be defined as the product of two conjugate variables, an effort (force) and a flow (velocity). An obvious but important physical constraint is that no one system may determine both variables. Along any degree of freedom a manipulator may exert a force on its environment or impose a displacement or velocity on it, but it may not do both. Thus, an assembly robot should have the property of admittance, accepting force (input) and yielding motion (output). The understanding is that once contact force is sensed during assembly, the robots

Footnote

Harmonious position and force control
Although a robot with active force control has the advantage of being versatile and programmable for different applications, it requires a more advanced control system and adapted programming to specify how the robot should interact with external constraints. Research so far has focused on the control strategy and its capacity for enabling the robot to establish stable, gentle contact while interacting with the environment. At present, there exists neither a high-level programming language nor a suitable programming concept with which to exploit the force control capability. Introducing force feedback only enables an industrial robot to respond to an environmental force. In no way does it tell the robot how to move when mating parts. A force control enabled compliant robot can therefore only try to avoid high contact force; it lacks a mechanism for mating parts, such as gears, according to their geometrical contours. Jamming of the assembly pieces is prevented, but no help is provided with aligning them. The presumption that a robots position can be modified via the interaction forces is difficult, if not impossible to put into practice when mating uncertainty is high and there are so many possible contact scenarios that they are mathematically impossible to handle. The ABB solution relieves users of the burden of complex programming by introducing the concept of attraction force. Coupled with admittance-based fast force control, this ensures not only that contact is made gently, but also that the part being mated is positioned for accurate alignment. As soon as all the alignment requirements are satisfied, the robot can begin its partSpecial Report Robotics

Bearing Installation Bearing liner Insertion Gear pump assembly Piston assembly Spark plug assembly Valve Insertion Reverse servo installation

Trans /gearbox

Bearing Installation Bearing liner Insertion Clutch plate assembly Forward clutch hub assembly Splined shaft insertion Sun gear assembly Torque converter assembly Transmission accumulator assembly Transmission differential assembly Triple clutch assembly
Tests carried out with advanced force control in industry have demonstrated its ability to improve cycle time and agility in different assembly applications.
ABBs robotic solution for this task started with the choice of robot, an IRB 6400. This was selected on the basis of its payload capacity of 150 kg the weight of the parts it can support, without help, and without contributing to any undesirable contact forces. Tests showed that the IRB 6400 is able to handle a total weight of 75 kg (the combined weights of the part, gripper, force sensor, etc) and still limit the contact force to less than 200 N. Advanced force controlled ABB robots, as the tests demonstrated, are capable of extremely delicate assembly operations, even when heavy parts are involved. Arm movement is feather-light but still powerful, a combination that suits assembly applications in a wide range of industries.
Superior performance, reliable operation
Developed primarily for assembly applications in the automotive industry 4 , advanced force control has potential benefits for many other areas of industry. Quick market acceptance is expected, especially where absence of the human element in mechanical assembly the ability to get it just right, first time has been a problem in the past. In bringing this innovation to market, ABB is once again underscoring its position as the industry leader in robotics technology.
The article was first published in ABB Review 1/2004 pp 1316. For more information see www.abb.com/robotics

Cylinder head

Bearing Installation Bearing liner Insertion Cylinder head assembly
Bearing Installation Bearing liner Insertion Planetary carrier assembly Planetary pinion insertion Splined shaft insertion Sun gear assembly
Dr. Hui Zhang Dr. Zhongxue Gan Dr. Jianjun Wang ABB Corporate Research Windsor, CT, USA hui.zhang@us.abb.com Dr. Torgny Brogardh Mats Isaksson ABB Automation Technologies AB Vsteras, Sweden torgny.brogardh@se.abb.com

Bearing Installation Bearing liner Insertion Piston assembly
The tests carried out with advanced force control in the automotive industry have convincingly demonstrated its ability to improve cycle time and agility in different assembly applications.

Welding sees the light

Precision welding with lasers and robots
Fabrice Legeleux The automated production of complex metal parts would hardly be possible without robots. Robots repeat the same movement sequences endlessly and tirelessly, delivering high quality workmanship at low cycle times. However, with robot technology alone, this would hardly be possible. The tools used by the robots are an equally important part of the process. They must fulfill the same high demands on accuracy, repeatability and reliability as the robots themselves, and they must interface optimally with the robots.
Welding requires a heat application of around 103 W/cm2. If too much energy is applied, the metal evaporates. If not enough is applied, the heat is conducted away too quickly. In laser welding, a powerful laser delivers this energy. Laser welding technology must additionally consider such aspects as focusing and controlling the position of the welding spot. Different laser welding technologies exist. One common approach is the CO2 laser. This technology, despite what its name may suggest, typically uses a gas mixture with 70 % He, 15 % CO2 and 15 % N2. A drawback of CO2 lasers is that the light cannot be transmitted along fiber optic cables. Instead, a system of mirrors is used. Very precise coordination is necessary to direct the light to the correct location. The Nd:YAG laser (neodymium yttrium aluminum garnet) solves this problem. It is a solid state pumped laser whose light can be transmitted along flexible fiber optical cables. Because of this, Nd:YAG lasers are finding increasing use in body in white applications. The technology does have its drawbacks: The efficiency of an Nd:YAG laser source is limited to around 23 %, and the light quality is inferior to that of a CO2 laser. The flexibility offered by being able to use fibre optics make many customers prefer this technology however. To offset the low laser efficiency, the technology must strive for the highest possible productive laser uptime. This is achieved using a mirror that switches the ray between cells: ideally, one cell starts welding the moment the other cell stops and the light is never wasted! The new IRC5 industrial robot controller provides unparalleled capabilities with its MultiMove functionality1) precisely synchronizing multiple robots with extreme accuracy. It also controls the mirror that switches the ray, ensuring that this perfectly matches the movements of the robot. The use of a fiber optic connection has another important advantage: The welding source is not rigidly attached to the robot 1. This saves on timeconsuming alignment and adjustment

Complex process

The MIG/MAG process is very complex. To fully understand the theory behind it requires in-depth knowledge of arc physics, fluid dynamics, material science, arc-electrode interaction, amongst others. Very few people in the world have all this knowledge, and of those that do, it is unlikely that any have any practical hands-on welding experience.
Dick Skarin, Brith Claesson, Gran Bergling
Transforms welding from black art to practical science

VirtualArc

On the other hand, many very skilled welders in the industry perform perfect welds without any special knowledge of arc physics and the science behind it. Rather, their choices are guided by so-called feeling for the process itself. Such knowledge can therefore be found only in the head of each individual welder. Unfortunately, the number of such skilled welders will most certainly decrease in the future, making it increasingly difficult for the welding industry to be successful.

The challenge

The concept of the traditional welder at work in his smithy is being replaced by a more high-tech view of the craft. Who isnt familiar with pictures of robot welders at work on production lines? Designers and manufacturers take precise and clean mass-produced welds for granted. But teaching a robot to weld isnt as easy as it may seem. Whereas a human welder draws on experience and intuition in choosing the correct welding technique and settings, a robot welder must be instructed in every detail of the procedure. Despite the availability of considerable theoretical knowledge, a series of test runs are often required to determine the correct settings. Such runs waste test materials and block robots that could otherwise be in revenue generating use. The costs and time involved mean that the number of test scenarios must be limited and the optimum can easily be missed.
Welding equipment suppliers must combine the science behind the MIG/MAG process with practical expertise: Welding must evolve from todays black art to a modern fully controllable manufacturing process. The welding industry itself is interested in raising cost efficiency by using simulation tools as much as possible

Virtual Arc

for production optimization and planning. ABB have taken up the challenge of serving robot arc welding customers by fulfilling these criteria. A unique simulation tool called VirtualArc has been developed to meet the arc welding requirements of customers. VirtualArc offers robot programmers, operators and welding engineers an expert system providing in-depth analysis of the arc welding process. Its use leads to improvements in process control, final welding quality and productivity. The simulation tool incorporates stateof-the-art technology, facilitating prediction of the dominant phenomena in the weld. Implementation time and

Those manufacturers, who have been quick to adopt and install new packaging systems, are stealing an edge over their rivals.
Until now, wrapping a new product has been a real challenge for the packaging industry. Whereas humans are able to quickly adapt to pack different products, building robots with the same flexibility is another matter. Until recently, modifying machinery to package different products was too great a challenge. Now, those manufacturers, who have been quick to adopt and install new packaging systems, are stealing an edge over their rivals. This is brought about by specially designed machinery and software developments. Good packaging companies also provide training to make adoption of new technology straightforward and trouble free. Changeover is especially quick and convenient when the entire packaging process comprises one integrated system.
The deployment of robots in packaging lines is an area where technology is quickening the pace and making human hands redundant. Also, unlike humans, robots do not suffer from repetitive strain, fatigue, boredom or any of the related illnesses or conditions these provoke. Establishing the need for robots in packaging lines, however, requires consideration. First, the investment needs to be cost justified, and then the system must be modelled to make the most of the capabilities of robots. Indeed, robots often replace human workers, which is why their acceptance is often a subject of negotiation. However, robots do present a strong case of their own and this is why we are seeing more and more getting to work in food industry. Incidences of strain industries and lost time in the baking industry are a motivator for many companies to seriously consider the role of the robot. As the USA, UK, and elsewhere in Europe become more litigious, the costs against claims against businesses from workers suffering workplace related injuries will rise.
Henrik J. Andersson ABB Automation Technologies Vsters, Sweden henrik.j.andersson@se.abb.com
The development of the IRB 6650S shelf robot a successful cooperation between supplier and customer
Ola Svanstrm When a customer needs a powerful and agile robot in a very short time and no existing product covers the niche, the chances of success may seem slim. But if the supplier is well-prepared and has a strong modular platform to develop from, then the situation is totally different. In close cooperation with Gestamp Automocion, ABB was able to develop a new shelf robot in under seven months. The IRB 6650S is handling the customers tough requirements with ease these translate into an enlarged working range, an increased handling capacity and a high moment of inertia all factors which increase the productivity of the customers lines.

Shelf life

Gestamp Automocion is a specialist in automation of press body shops for automotive industries and a longstanding customer of ABB robots. In early 2003, Gestamp contacted ABB explaining that they needed improved shelf mounted robots with a better payload and working range. The time frame for development was very tight but made possible by the recently introduced new large robot platform. This had been launched the previous year and provided a solid basis for further development based on a modular and well thought out concept.

For more information on investment casting automation, visit www.abb.com/robotics
skilled workers to achieve the high casting quality that customers demand. However, there is a definite automation trend. Shell making, where the wax trees are dipped in an alcohol or water based slurry and where the ceramic shells are continuously built using special sand is often robotised 1. Robots are also used for post processing applications, such as grinding and polishing of the cast part. Progressive companies are now looking at automating the wax tree mounting area of the shop floor. Global competition drives us to keep looking for automation possibilities to increase our productivity, says Bertil Bredin, CEO of TPC Components in Hallstahammar, Sweden 2. Quality control using vision systems is just one example of this trend. TPC Components produces some 1000 different articles every year some in very small, prototype series and oth38 IRB 7600 helps increase output
ers at volumes of up to 250,000 parts per year. The ceramic shell making process is practically completely automated. An old hydraulic Unimate robot previously produced the large shells. Because of its outstanding reach and large handling capacity, it was retained in use for many years. Only recently was it replaced it with ABBs new IRB 7600 long-arm version, which features a reach of 3.5 m and a handling capacity of 150 kg. Thanks to these characteristics, it was possible to introduce the new robot into the existing production line without making any changes to the machinery. Only minor changes of the control system were necessary. The new robot reduces cycle time, allowing the robot cell to handle more parts per unit of time 3. The number of manual operations are reduced and final output has so far increased from 1400 trees per day to 1800. When the production cycle is optimised, we plan to produce approximately 2300 trees per day, says
Claudia Berg ABB Automation Technologies AB Vsteras, Sweden claudia.berg@se.abb.com
David Marshall, Nigel Richardson, Chris Miles
Zinc die caster opts for robot-based automation

The die is cast

The South Yorkshire based PMS Diecasting is a small but highly dynamic company. It has positioned itself at the forefront of its field by adopting the latest techniques.
High volume zinc diecasting manufacturer PMS Diecasting, based in Rotherham, South Yorkshire, has automated part of its plant by installing an ABB IRB 140 FoundryPlus six-axis robot, to tend a Frech DAW 20 RC real time control, zinc diecasting machine. The installation, carried out by robot integration experts Geku Industrial Automation Systems, is unusual in that the ABB robot is tending a relatively small (20 tonne) machine, while the Frech, using real time control (RC), is one of only four such machines operating in the UK.

the Geku-designed cell has had a most significant effect. It is very compact and highly efficient, scrap is reprocessed immediately rather than being allowed to build up in bulky skips and an arduous, boring and labour intensive task has been fully automated, allowing staff to use their skills on much more challenging and fulfilling operations. He adds: The introduction of full lights-out operation is the next milestone and a significant development for the future success of our business.
David Marshall ABB Automation Technologies Division Milton Keynes, UK david.marshall@gb.abb.com Nigel Richardson Geku Industrial Robot Systems Chatham, Kent, UK geku@compuserve.com www.geku.co.uk Chris Miles FMPR Keighley, West Yorkshire, UK chris@fmpr.co.uk

TeachSaver a time saver

An innovative programming tool opens great new applications for robotics
Per strm Even so, some 80% of foundry cleaning work is still performed manually. The principle barrier to robotization is not the complexity of the task itself, but lies in the programming of the robots. Such programming is time-consuming. With many production batches being small, this leads to excessive down time of the costly equipment. All this is set to change with ABBs new TeachSaver. This product, an add-in to RobotStudio, brings with it an array of welcome benefits to foundries across the world. The use of TeachSaver slashes programming time by up to 90 %. In addition, it offers better quality products thanks to more accurate calibration and processing methods. Repeatability is a further bonus as well as dramatic timesavings throughout the life cycle of the work cell. And, as programming is carried out off-line, there is never any need to disturb production. For decades, the foundry industry has been searching for effective, profitable ways to automate its cleaning

A Time Saver

The foundry-casting process is not always as perfect as many people imagine. Casting leaves excess material known as burrs and flashes and this must be cleaned away before the part can be used as intended. The deburring and de-flashing process is dirty, noisy and hazardous. Additionally, scrap rates are high and quality inconsistent. It might seem that the stage is set for the entry of robots.
operations. The programming time and the fact that the production has to be stopped during the programming is one of the biggest obstacles to a more expansive use of robots and has also restricted automation to only the largest batches. With TeachSaver, robots can be programmed off-line and time cut dramatically from weeks to hours. This opens up new possibilities for the automated cleaning of smaller batches and complex parts with a high cleaning content.
The use of TeachSaver Step-by-Step Process path programming
TeachSaver provides high-accuracy methods for calibration of the tools and of the relation between the robot and the workpiece. This is automatically handled in TeachSaver and the results can be repeated time after time.

Working closely with a major Japanese automaker, ABB began to pursue the idea applying it not only to bumpers. As carmakers struggle to position themselves in a highly competitive global market, offering consumers a greater choice in color and features is one way to set themselves apart. At the same time, the cartridge concept promised to reduce VOC (volatile organic compounds) emissions and paint waste, an important considera-
ABBs cartridge bell system (CBS). The proximity of the paint to the bell cup minimizes paint and solvent loss when the color is changed. atomizer paint cartridge
A paint robot changing its cartridge in the manipulator. The empty cartridge is automatically refilled.

bell cup

solvent

solvent delivery line

piston

cylinder

server motor
tually attracted to it electrically. The major issue raised by this method is sparking. A spark can easily ignite the flammable mixture of vaporized paint and solvent and air. The developers had to find the ideal pressure to hold the cartridge in the robot arm. Too much of a vacuum would have led to sparking. As these hurdles were cleared, a working concept began to develop. A conventional robot painting systems involves large tanks for the main color, smaller tanks for special colors, a tank for solvent, paint supply lines leading to a Color Change Valve unit (CCV), and a Flushable Gear Pump unit (FGP) mounted on the robot to regulate paint flow. They all have to be cleaned for paint changes. In the new concept 1 the robot was to be separated from the paint delivery system. There would still be paint and solvent tanks, albeit much smaller and for special colors. These would feed
to a cartridge handling unit, where the paint would be filled into the cartridges. Each cartridge would contain a piston driven by solvent to force out the paint as painting progressed. A full cartridge would contain only paint below the piston, and no solvent; as solvent was pumped into the cartridge, the paint would be driven out, until the piston was fully extended with only solvent above it. The robot required only a solvent-feeling line; all other lines would be singleusage and lead only to the cartridge station. A surprisingly simple and straightforward idea! At least, as an initial concept. The cartridge change time requirement made the supply and the removal of solvent in the cartridge an important issue to overcome. What was desired was an almost instantaneous and constant delivery of solvent to the cartridge, resulting in a similarly constant and smooth delivery of

with the first robot. A single instruction in the work handling robots program is sufficient to move it from the start to the finish of its trajectory along the arc. However, because the welding robot applies two spots in different spatial positions and, therefore, requires two instructions in its program, the handling robot must also have two instructions. Therefore, the arc movement must be accomplished using two move instructions, one to a midpoint and another to the end of the arc, which are executed synchronously with the two move instructions in the welding robot program.
that their relative position may be adjusted and then switched back again for the coordinated jogging to continue. This is a powerful tool in fine-tuning MultiMove programs and is only offered by ABB. Recovery from a production stop due to equipment or process failure is a potential problem due to the complexity of the choreography in MultiMove operations. Not only has the robot at fault to avoid work and tooling, but it must also coordinate with its partners during its retraction to a safe position as well as on returning to its last position. The problem is eased by the IRC5 controller because of its path recording functionality. This is activated for every robot in a MultiMove operation. Knowing the path leading to the error point enables the faulty robot to retract in synchrony with the coordinated robots to a safe point that is identified in its RAPID error recovery routine. The same path data will similarly be used after recovery to return all the coordinated robots to the program positions at which the error occurred. Some errors necessitate the re-execution of a command (otherwise known as a retry) rather than returning the robot to its last known position. An example of this is an arc failure during arc welding. In this case an arc
Shorter lead times, increased productivity and improved quality are some of the generalised potential benefits of multiple robot operation with the new IRC5 controller.
Another feature of MultiMove is the ability to jog multiple robots using the joystick on the FlexPendant 2. During coordinated jogging, the relative positions of all the devices remain constant and are exactly the same as during the full speed execution. At any point, any of the devices can be switched to an independent jog so
ABBs Flexpendant (left) is part of the IRC5 package. It supports robot programmers (right) through its ergonomic design, customized menus and touch screen.

The brain of a synchronised robot cell is the IRC5 controller.
restrike makes more sense than a retraction. Therefore, a retry will be specified in the RAPID error recovery routine. In MultiMove all devices need to be coordinated during the retry. To make it easier to recover from such errors in arc welding, ABB has developed a new asynchronously raised error function in RAPID. For instance, in the above example it is most likely that the arc failure will occur along the programmed path after the instruction has been executed but before the robot has completed its movement to the end of the path. In this case, it is necessary that the error recovery routine is executed at the point of the arc failure and not at the completion of the instruction. The asynchronously raised error function allows this to occur in MultiMove as well as in single robot routines. Shorter lead times, increased productivity and improved quality are just some of the generalised potential benefits of multiple robot operation with the new IRC5 controller 3. Even in totally independent robot operations, time and costs are reduced due to the efficient internal communications and minimal handshaking of the single controller. When some degree of synchronisation is introduced, waiting times can be minimised leading to further reductions in cycle times.
Better product quality is a high potential benefit of MultiMove. This can be achieved, for example, with two or more robots working together in order to balance the load on the workpiece. An illustration of this is simultaneous arc welding to eliminate the risk of distortion due to uneven shrinkage on cooling. Another is the use of two or more robots to handle delicate or flimsy workpieces that may flex or bend under their own weight. It is also possible to expand the parton concept with MultiMove by coordinating a workpiece-handling robot with one or more process robots, helping to simplify and reduce tooling and fixturing. This can also reduce cycle time as the time to place the workpiece in the fixture has been eliminated and the process robots may be able to start their operations as soon as the part is picked up. Moreover, the 6-axis robot has more dexterity in manipulating the workpiece as compared to a rigid fixture or even a servo-controlled positioner. This could mean, for instance, that the process robots are able to access all areas of the work, allowing the operation to be completed in one handling with no intermediate stops for reorienting the work. This is called onestop or one hit processing. Another advantage of coordinating a work-handling robot with two or
more processing robots is the higher relative speeds attainable, for instance, between the weld torch and the workpiece, leading to possibly better quality welds and/or shorter cycle times. A further benefit is in lifting heavy loads. It may be less costly to employ two smaller robots to lift the load rather than a larger one, or the load may be heavier than the capacity of the largest robot but not of two robots working together. The unique functionality that MultiMove brings to the whole ABB robot range sets new standards in robot technology and opens up a range of applications that were previously impractical or uneconomic. Its development has been backed-up by the knowledge gained from the previous four generations of ABB robot controllers and aided by the expertise generated from over 125,000 ABB robots installed worldwide. MultiMove further strengthens ABBs lead in advanced robot systems.

reducing mechanical stress on the complete system or individual axes. After constraints have been defined, MultiMove PowerPac automatically generates the robot paths.
ABB is the first and only supplier to offer an offline planning, programming and simulation tool for a MultiMove system.
In the following, the example of a two-robot arc-welding cell is considered. The process of automatically generating robot paths defines which tasks each of the two robots should execute, and defines the relationship between the two independently controllable robots. One of the robots can hold an object in a desired orientation while the second robot performs the arc weld along the specified path. When paths have been decided, the robots motion is automatically generated. Their interaction can then be verified in a simulation and subsequently downloaded to the controller of the real robots.
Thanks to MultiMove PowerPac, users can evaluate different cell layouts and production concepts in a very short time. MultiMove PowerPac makes a MultiMove system as easy to program and as flexible to modify as a single robot system. With MultiMove PowerPac, ABB offers the best MultiMove offline-programming tool in the industry. There is, at present, no other working solution on the market.
RobotStudio, VirtualRobot, MultiMove, PowerPac, IRC5, FlexPositioner are trademarks of ABB.
Programming with MultiMove offers great time savings.
Time required for programming a robot compared to MultiMove PowerPac
MultiMove PowerPac FlexPendant Competitor average
Jonas Anselmby ABB Automation Technologies Mlndal, Sweden jonas.anselmby@se.abb.com

Product and tools

Virtual and real
Bertil Thorvaldsson How do you teach a robot? Until recently, programmers had a robot at their disposal and took it through the cycle of instruction, test and correction until everything worked as desired. Under todays drive for higher productivity and competitiveness however, this method has one great drawback: The costly robot and all associated equipment have to be removed from the production process during programming. ABBs RobotStudio offers an alternative. Instead of using a real robot, the programmer uses a virtual one. The robot is replaced by a desktop PC running the RobotStudio software. The software models the robot and all associated tooling and workpieces the programmer can set up and test programs from the comfort of his office. But this is more than a mere simulation tool. The program that runs on the virtual robot can be

The operator specifies target points using a hand-held device directly in relation to the object to be painted. The simulated result is displayed as a composite image on headworn displays 1.
on a head-worn display with an integrated camera 2. In addition to capturing the real world scene, the camera system also comprises a vision-based tracking system that enables correct registration of the simulated process result. The tracking is made possible by attaching automatically configurable visual markers to the object. Programming and editing is done through a wireless pointing device mimicking the real process paint gun. The device is tracked with six degrees of freedom enabling the specifications of path waypoints and process related
information directly in relation to the object. The pointing device has a number of interaction buttons through which the operator can easily specify or change program properties; the operator simply pushes a button to specify or delete a waypoint.
new position. The virtual graphics are continuously updated in such a way that the changes made to the program are visualized instantly. For example by changing the width of a paintstroke, the process quality related to paint coverage can be inspected. After the robot program has been specified, robot instructions are generated automatically and downloaded to the robot controller for execution.

Prototype testing

Programming and editing is done through a wireless pointing device mimicking the real process paint gun.
Editing of the manipulator path is simply done by dragging a waypoint to a
Initial benchmarking tests showed that robot programming using this novel robot programming prototype system achieved the goal of being at least
twice as fast as conventional programming methods. Todays prototype makes up a complete system with full functionality for generating a new robot program including set-up and configuration, creating the robot program, real-time reachability check, editing and generation of robot programming code. The prototype system fulfils all the requirements regarding set-up and configuration (less than 1 day) and training of (new) users in less than 3 days. The prototype system also fulfils the requirement of accuracy (x, y and z deviation below 5 mm). gy will revolutionize other applications - not just within robot programming, but also for other applications such as service and maintenance. AR represents a paradigm shift for Easeof-use of technical systems. Corporate Research Center, Norway, has introduced a totally new but very powerful technology.