Secondary Logo

Journal Logo


The ToMPAW Modular Prosthesis: A Platform for Research in Upper-Limb Prosthetics

Kyberd, Peter J. PhD; Poulton, Adrian S. PhD; Sandsjö, Leif PhD; Jönsson, Stewe CPO; Jones, Ben MEng; Gow, David BSc

Author Information
JPO Journal of Prosthetics and Orthotics: January 2007 - Volume 19 - Issue 1 - p 15-21
doi: 10.1097/JPO.0b013e31802d46f8
  • Free

The design of an upper limb prosthesis presents mechanical, electrical, and cosmetic challenges. The major concern in the provision of each limb prosthesis is to create a solution that is most appropriate to the user. The level of loss, the strength and mobility of the residuum, and the needs and abilities of the user must be taken into account. Because every user is different, a high degree of customization is needed. Variations include degree of loss, side, size, gender, and preferred method of control. This makes every prosthesis as individual as its wearer, which complicates prosthetic assembly and fitting.

The solution described here—to adopt a modular approach—has not been feasible because commercial suppliers have not supported such thinking. Conventional prosthetic limbs provide different mechanical substitutes for each joint. Although there is some interoperability among different joints, each joint has different requirements, so linking the control systems becomes complex. In addition, although the joints are separate units, a single controller operates the entire prosthesis; thus, a failure of the controller means a failure of the prosthesis.

Often it is difficult to select a suitable control strategy because there are several degrees of freedom to be controlled with only a small number of separable control signals available. The conventional response is to make the commands sequential, and the order of commands and the input signal range are best tested first. Thus, it is important that different strategies can be easily tried. Microprocessor-based controllers have begun to fulfill this objective1,2; however, there has been little standardization in this area. An international “Totally Modular Prosthetic Arm with high Workability” (ToMPAW) consortium addressed these problems. Funded by the European Union to stimulate research in prosthetics technology, the consortium built on earlier experiences of the participants, including limb fitting, orthopedic, technological, and manufacturing approaches.


Principle among the problems associated with creating a practical prosthetic hand is making it easy to operate while improving its functional range. The natural hand is controlled by the motor and sensory cortexes of the brain. Thus, when a person wants to pick up an object, for example, the central nervous system (CNS) selects the most appropriate grip and then adapts the hand's shape and grip force to maximize the contact area and minimize the contact force. The person is not generally aware of or even thinking about the act. By comparison, a person operating a conventional prosthesis has minimal information fed to the CNS. Operation of the multiple degrees of freedom required for very functional devices is difficult, often impossible, to achieve, so motions and control are limited and controlled sequentially in an open loop. To add functionality based on current control principles to increase the hand's functional range would make the hand burdensome, perhaps negatively affecting basic control of the device.

An anthropomorphic hand is generally set in a tips (or precision) grip because 60% of all tasks use this grip. Consequently, this hand has the incorrect grip for nearly half of all normal activities. Although great facility can be achieved via specialist terminal devices, it does pose limitations in the devices range. The most likely solution would then have a mechanism with a number of independent motions. This would be at the cost of either slow serial control of each motion or some more complex multifunctional control. To be practical, any control paradigm chosen must reduce the intellectual effort required to manage the hand. Thus, there is a need for less mentally demanding control principles addressing the control issues in multifunctional prostheses.


The following research programs form the background knowledge of the members of the ToMPAW consortium.


Developed in Sweden during the 1970s, the SVEN hand embodied a number of important concepts. The key design aspects of the hand were wrist rotation of the glove to be referred up the arm to the elbow (allowing for less strain and unattractive wrinkles of the glove), powered wrist flexion (something few other prostheses have ever achieved), and an adaptive grip accomplished by a single wire driving all four fingers and passive wrist deviation. However, the most significant contribution was the innovative ideas on how to control the hand, making use of the “phantom limb” experience.

To use these signals, the control of the individual prosthesis was derived from EMG recordings from six electrodes placed on the residual limb while the subject performed the basic hand movements: finger flexion, finger extension, pronation, supination, wrist flexion, and wrist extension. By means of a pattern-recognition technique implemented in the prostheses, the EMG signals from the residual limb could then be used to control the hand prosthesis. This way the prosthesis was adapted to the myoelectric patterns produced by the user, rather than the opposite, thereby reducing the time needed for the user to learn to control the multifunctional device.3,4

The original EMG processor was an analog system. More recently a digitally based experimental system was realized, which demonstrated the method to be applicable using modern low-cost microprocessor technology.5,6


Based on a program of investigation dating to the 1960s, the most recent version of the Edinburgh arm is composed of compact joints made from worm and wheel combinations linked by carbon-fiber tube sections.7 The motors are mounted within tubes that make up the links between the joints. They are mounted entirely on one side of the joint. The worm gears take up very little of the arm on the other side of the joint, allowing for greater design flexibility. If the user has a long distal segment, the motor lies below the joint. If the final segment is sufficiently short and there is space for the motor to be positioned above the joint, the weight is more efficiently distributed above the joint. The motor need not raise itself as well as the arms load, allowing the design of a modular mechanism and a powered shoulder system to be applied clinically. Before ToMPAW, these systems were controlled with conventional analog electronic controllers. The addition of the ToMPAW technology allowed for the modularity to be extended to the controller level.


The philosophy adopted at Southampton University's Control Engineering Group by Professor Jim Nightingale and team was that feeding information back to the wearer was difficult or impossible because it would tend to cause information overload for the operator. Instead, sensory information was sent to a computer-based controller that coordinated the joints in a multijointed arm prothesis, and the user had to give only simple instructions to reach, grasp, or release an object.8

A difference was observed between control of the hand and that of the arm. The hand is required to achieve the correct grasp only once the arm has placed it in the correct position and orientation. Thus, the hand controller focused on selection of different grasp forms. Initial studies were conducted in the laboratory, but recent work at the Oxford Orthopaedic Engineering Centre has conducted clinically based studies.9–11 A Southampton Hand can have any number of independent motions (as many as six have been realized so far12); the use of sensors and a controller means that only one general “grasp” instruction is required. This frees the operator to make only strategic decisions about the object and the environment and leaves the “reflex” control to the hand.

The concept of hierarchical control can be used in a complete arm. The Southampton group demonstrated that when the arm was driven by three inputs that control the position of the wrist in space (X, Y, Z), an arm with more than three degrees of freedom can be easily operated.13,14 Individual joint positions were calculated and maintained autonomously by the controller. Additional refinements to the system included the use of the distribution of forces across the surface of the hand to allow the arm to balance the forces and maintain a more stable grip. One significant conclusion was that the need for more localized numerical processing called for microprocessors at each joint, a task unfeasible at the time, but one of the targets of the ToMPAW system.



The initial stage of the project was to survey the various groups with interest and knowledge of the field, namely users, prosthetists, and prosthesis manufacturers. From this a series of qualities was drawn up, (C. Wartenberg et al., unpublished observations) followed by the creation of a set of requirements, which were used in the design of a modular prosthetic system.15

Along with the understandable desires for lighter and more anthropomorphic arms, the qualities identified from the survey that particularly informed the design choices of the ToMPAW system were:

  • Quiet operation of joints without the sounds of brakes being applied or removed
  • A wide range of grip shapes and good control of the grip force (for example, not too great a grip on initial contact)
  • Reliable operation, free from breakdowns and interference from electrical noise.

This latter requirement meant that a bus-based modular system was an inevitable choice. Its use in other industries has shown that the separation of functions allows for easy application. Upgrades and modifications are simple. The separation of the controller from a single point to a distributed system ensures that a single failure does not cause the total failure of the device. In addition, the systems can be made from a smaller set of standardized components.

The ToMPAW system aimed to provide a modularized set of components from which prostheses to fit a wide range of users can be assembled. The system is modular not only mechanically but also in terms of electronics and software. This is facilitated by the use of a network of microprocessors, each controlling one or two joints. Each processor is close to the sensors and actuators that interface with it. This has reliability advantages as well as the low stock-keeping costs and ease of maintenance associated with modular parts. In addition, the bus architecture can be easily interfaced to the other networks so a node that provides a TCP/IP interface will allow remote diagnostics, testing, and adjustment for the arm system across the Internet.


The design of the controller took into account the requirements of modularity and reliability. This meant that there was a need to keep interconnections to a minimum (because these tend to be one of the most unreliable features of an electronic system). The preferred approach was to use a microprocessor network of the Fieldbus type having built-in network communication support and protection mechanisms against data corruption. In addition, a network solution allows for graceful degradation where a fault in one node (or processor) of the network need not affect the others. The network technology selected for this project was Echelon's LonWorks (Palo Alto, CA).

A “master” (the most proximal node making up the interface to the user) node accepts as many as to eight analog inputs and can be used for overall control of a system (Figure 1). However, the hand controller can also accept two EMG inputs, so a hand prosthesis does not require a master proximal board when the hand is stand-alone. This allocation of a separate device to each joint means that the same device handles the motor and sensors associated with each joint. This minimizes the network traffic.

Figure 1.:
Schematic of the ToMPAW system, showing the different controller nodes at each of the joints. User input is at the proximal node and can take a variety of different inputs. Each joint has a distal node that performs the closed loop control of the joint. The hand node requires more input and output lines, especially when the hand is in stand-alone mode.

The electronic subsystems of the arm were tested using a hybrid system consisting of the existing mechanical assemblies of the Oxford Intelligent Hand11 and the Edinburgh Arm.7 Two such arms were constructed.


One advantage of the ToMPAW system is that additional joints and functional units can be added simply. One such application was a laboratory-based demonstration of coordinated joint control.

A physical user interface for the control system used a joystick as the primary input device. It was modified to provide force feedback by the addition of two motors, each with a beam connected to the joystick lever (Figure 2). Each beam had a strain gauge attached to enable the force applied to be measured. This value was then input to an extra ToMPAW node used exclusively for the input/output to the joystick. The arm controller was extended by the simple addition of the extra node and changes to the input software. The advantage of the modular system was shown because no other changes to the system were necessary.

Figure 2.:
Instrumented joystick used to investigate coordinated joint control of the ToMPAW arm. The joystick has strain gauges on two beams that are driven by two motors so strain against the beam is measured and the joystick can be driven at the same time as the joints of the arm.

For laboratory experimentation the joystick was operated by subjects seated next to a support made for the joystick that allowed the motion of the acromion to drive the joystick. Two control formats were tested.

Direct control, in which each of the two inputs at the joystick was mapped to the control of individual joints, so the prosthetic shoulder was controlled with shoulder elevation and elbow flexion by shoulder proretraction.16

Coordinated control, in which the position of the wrist was mapped to a position in the plane of the arm motion. For any given position of wrist within a reachable workspace there was a unique pair of angles for the shoulder and elbow. The centered joystick was the rest position for the arm (fully extended position hanging down). The values could be accessed from a look-up table in the microprocessor's memory and then sent on the bus to the controllers of the individual joints, which controlled the detailed motions. The envelope within which the arm moved was positioned relative to the center of rotation of the shoulder. For the experiments this center was placed 15 cm laterally to the center of the subject's own shoulder. Three targets were placed within the envelope close to the edge of the area.

Raising the shoulder caused the wrist to be raised; forward motion of the shoulder propelled the wrist forward (Figure 3). The detailed control of the two joints (shoulder and elbow) was then maintained by the additional LonWorks node. Five able-bodied users were tested to gauge the control and mental load required to perform a targeting task.16

Figure 3.:
Demonstration of coordinated joint control of the ToMPAW arm. The joystick is propelled by the acromion. The arm motion responds by mapping the XY position of the joystick to the position of the wrist in a plane. The detailed control of the different joints is performed by the ToMPAW controllers.


The purpose of this generation of the ToMPAW system was to demonstrate the adaptability of the approach. It was then applied in four transradial prostheses and two transhumeral prostheses (Table 1). More advanced controllers were built with little or no changes to the basic form of the arms.

Table 1:
Characteristics of the users of the intelligent arm systems


The hands were based on the Southampton Hand with hierarchical control and two degrees of freedom.11,17 They were originally fitted with bulkier electronics, which were changed for the smaller bus-based controller with the same functionality. One example was used for more than 4 months before the hand was refitted with the new electronics. This user had not employed electromyography to control his hand previously (usual prosthesis was a Steplon hand and an olecranon suspension socket). It was found that once the electrode placement was refined, the control of the hand was easy to achieve.

Laboratory experience with the hand with a female user showed that it was easy to learn a control format that differed from conventional methods. The hand is proportional voluntary opening/involuntary closing (VOIC) with the degree of opening made proportional to one muscle, so as the muscle relaxes so does the prosthesis. This is in contrast to the subject's usual device, which is thresholded voluntary opening/voluntary closing (VOVC). She had no particular problems in relearning the control or switching between the two formats. This was not surprising because the former method is analogous to the way body-powered devices are controlled and users are expected to be able to switch routinely between body-powered and electric hands.

A laboratory-based user of an earlier hand system18 (controlled in the same manner but with different electronics and mechanism) after a break of 5 years, found it easy to use the newer hand.

A second user of the conventional electric hand used the hand at home for a number of months, finding no particular problems with the control format, switching between the hands without great mental effort.


Two hybrid arms were constructed, linking hand, wrist, and elbow. Both formats of elbow (i.e., the motor below and above the joint) were used in this pair. The advantage of the microprocessor-based system is that the control can be altered to suit the user. The hands and arms were fitted to users who attend the Limb Fitting Clinic at the Nuffield Orthopaedic Centre in Oxford, United Kingdom, or the Sahlgrenska University Hospital, Göteborg, Sweden.

The first user could not produce sufficiently reliable muscle signals (EMGs) to make proportional control successful, so that the arm was driven VOVC; thus, extension caused the elbow to extend, and the flexion value set the arm to flex. Each joint was similarly driven. The different joints switched between each other in a cycle by a bump switch on the inside of the arm. The sensors in the hand prevented the hand from crushing the object as it contacted the hand.

The second user possessed a short humerus following a traumatic amputation 20 years previously. His muscle bulk was small on that arm, and he tended to use a conventional arm with a fixed shoulder and cosmetic hand with a cable-operated elbow. Despite that he had not used the arm for years, he was able to learn to operate the arm with two EMG inputs. Training started with using a visual analog type display of LEDs. The user then progressed to controlling the hand when it was positioned on a base on the table in front of him so that he could build up his muscles and techniques, before progressing to wearing the arm for short periods.

The control of the two muscle channels was not fully independent. A “winner takes all” algorithm was implemented, in which the larger signal is taken as the exclusive input, rather than the difference of the two channels.

The different joints were controlled sequentially, and controlled alternation used a switch built into the harness so scapular abduction would shift control to the next joint. If no input was received by the controller for more than 10 seconds, the control reverted to the hand. As the user's strength and skill improved, his confidence did also. He was able to distinguish between the voluntary bidirectional driving of the wrist and elbow and the VOVC of the hand.


The able-bodied users of the two control formats found the coordinated control format was easier to guide the arm and required less concentration than did the direct control of the joints.


The success of the project is demonstrated by the application of the technology to existing mechanisms, allowing the mechanisms to be extended beyond their original designs (changing control formats, adding different additional degrees of freedom to the system and computer-based controllers) with little alteration. This allowed more advanced controllers to be realized simply, such as the addition of an extra node to provide feedback control of the joystick. It is the capability of the system to be extended that remains its biggest strength. Other groups have also investigated the possibility of remote teleoperation and diagnostics.19 The ToMPAW system uses a computer networking system that was designed to be used in a range of circumstances from local to remote. The nodes can communicate with each other along their proprietary network or through gateways to other protocols, including TCP-IP. Parameters can be set or examined using Web-based tools allowing remote diagnostic, training, or evaluation, without the need for specialized communications.

With increasing numbers of laboratories and programs investigating the possibilities of using pattern-recognition techniques on EMG and other biological signals,20–24 there is a need for a computer-based controller and a multi-axis arm that can use these instructions. The network solution means that if the controller requires additional specialized processor hardware,24 it can easily be added to the network.

During the trials, limitations of the processing power of the local nodes were observed. The local memory and processor speed restrictions can be resolved by the adoption of recent, more powerful variants of the devices. The modular aspect of the electronics means that any advanced signal processing required could be performed using additional microprocessors that would then pass the results on to the other ToMPAW nodes to implement the arm control.

The aim of the project was to demonstrate the use of a systematic approach to the design of a prosthetic arm. This suggested a modular solution to all parts of the system, which in turn allows the application of the different joints of the arm according to the requirement of the particular user. Although this is the dominant approach in the information technology, it has not extended as far within the field of prosthetics. It is clear that it should do so. Modularity has the potential to cut limb-fitting costs by reducing time and effort expended throughout the supply and fitting process. Fewer parts need to be stocked at the local center (or by the manufacturer) because the smaller number would cover the whole range of users. However, it would still allow better access to parts for fitting and maintenance. The reduced stockholding also limits the investment required to maintain an active limb service or busy supplier. In addition to the economic gains, users would benefit from the change because they would be more likely to become readily equipped with the most appropriate devices for their needs more rapidly. If the advantages enjoyed by adopting this approach in other industries were extended to prosthetics, it could result in more durable and dependable devices that are standard across the entire industry.25


This work was supported under the European Commission's Telematics Applications Programme (Project No. DE4210) and by the New Brunswick Innovation Foundation.


1.Poulton A, Boman A, Gow D, et al. Use of microprocessor technology in upper limb prostheses. Paper presented at the 2nd International Conference on Advances in Osseointegration, Surrey, United Kingdom, June 17–18, 2000.
2.Lake C, Miguelez JM. Evolution of microprocessor based control systems in upper extremity prosthetics. Technol Disabil 2003; 15(2):63–72.
3.Herberts P, Almström C, Kadefors R, Lawrence P. Hand prosthesis control via myoelectric patterns. Acta Orthop Scand 1973;44:389–409.
4.Almström C. An electronic control system for a prosthetic hand with six degrees of freedom. Technical Report, Research Laboratory of Medical Electronics Chalmers University of Technology 1977;1:77.
5.Berntsson L-O, Wiklund P, Almström C, et al. Control of multi-functional hand prosthesis for below elbow amputees by means of identification of EMG signal patterns (in Swedish). Technical Report, Annual General Meeting of the Swedish Society of Medicine, SLS, Stockholm, Sweden, 1994.
6.Ewaldz M, Almström C, Herberts P, et al. A system for refining control coefficients for myoelectric multi-functional hand prosthesis based on pattern recognition (in Swedish). Technical Report, Annual General Meeting of the Swedish Society of Medicine, SLS, Stockholm, Sweden, 1994.
7.Gow D, Douglas W, Geggie C, et al. The development of the Edinburgh Modular Arm System. Proc Inst Mech Eng Part H 2001;215:291–298.
8.Nightingale JM. Microprocessor control of an artificial arm. J Microcomp Appl 1985;8:167–173.
9.Kyberd PJ, Tregidgo R, Sachetti R, et al. The Marcus intelligent hand prosthesis. In: Ballabio E, ed. Rehabilitation Technology: Strategies for the European Union. Amsterdam: IOS Press;1993:98–102.
10.Kyberd PJ, Chappell PH. The Southampton Hand: an intelligent myoelectric prosthesis. J Rehabil Res Dev 1994;31(1):326–334.
11.Kyberd PJ, Evans M, te Winkel S. An intelligent anthropomorphic hand, with automatic grasp. Robotica 1998;16:531–536.
12.Light CM, Chappell PH, Kyberd PJ. Quantifying impaired hand function in the clinical environment. Proceedings of the MEC '99 Myoelectric Controls Symposium, Fredericton, New Brunswick, Canada, August 25–27, 1999.
13.Storey N. Control of an arm prosthesis [PhD thesis]. Electrical Engineering Department, University of Southampton, 1997.
14.Swain ID. Adaptive control of an arm prosthesis [PhD thesis]. Electrical Engineering Department, University of Southampton, 1982.
15.Poulton A, Kyberd PJ, Gow D. Progress of a modular prosthetic arm. In: Keates S, Langdon P, eds. Universal Access and Assistive Technology. London: Springer; 2002:193–200.
16.Jones B, Kyberd PJ. Improved control for an artificial arm. Proceedings of the MEC '05 Myoelectric Controls Symposium, Fredericton, New Brunswick, Canada, August 17–19, 2005.
17.Kyberd PJ, te Winkel S, Poulton A. A wireless telemetry system for user training of artificial arms. Prosthet Orthot Int 2002;26:78–81.
18.Kyberd PJ, Gow D, Scott H, et al. A comparison of upper limb prostheses users in Europe. Proceedings of the MEC '99 Myoelectric Controls Symposium, Fredericton, New Brunswick, Canada, August 25–27, 1999.
19.Lemaire ED, Fawcett J, Nielen D, Leung AKL. Telehealth strategies for remote prosthetic applications. Technol Disabil 2003;15(2):145–150.
20.Kelly MF, Parker PA, Scott RN. The application of neural networks to myoelectric signal analysis: a preliminary study. IEEE Trans Biomed Eng 1990;37(3):221–230.
21.Gallant PJ. An approach to myoelectric control using a self-organizing neural network for feature extraction [thesis]. Queens University, Kingston, Ontario, 1993.
22.Ajiboye AB. Neuro-fuzzy logic as a control algorithm for an externally powered multifunction hand prosthesis. Proceedings of the MEC' 02 UNB Myoelectric Controls Symposium, Fredericton, New Brunswick, Canada, August 21–23, 2002.
23.Englehart K, Hudgins B, Chan ADC. Continuous multifunction myoelectric control using pattern recognition. Technol Disabil 2003;15(2):95–104.
24.Kajitani I, Iwata M, Harada M, Higuchi T. A myoelectric controlled prosthetic hand with an evolvable hardware controller. Technol Disabil 2003;15(2):129–144.
25.Stavdahl Ø, Mathisen G. A bus protocol for intercomponent communication in advanced upper-limb prostheses. Proceedings of the MEC '05 Myoelectric Controls Symposium, Fredericton, New Brunswick, Canada, August 17–19, 2005.

hierarchical control; microprocessor; modular design; myoelectric control; prosthetic limbs

© 2007 American Academy of Orthotists & Prosthetists