The Dawn of Powered Lower-Limb Prostheses: Part 3

Content provided by The O&P EDGE
Current Issue - Free Subscription - Free eNewsletter - Advertise

The Dawn of Powered Lower-Limb Prostheses series of articles describes the progress of prosthetic technology toward powered lower-limb prostheses. In the first article, the two approaches most prevalent in the academic literature, the BiOM foot and the Vanderbilt knee, were described (The O&P EDGE, March 2013). The second article described some of the alternative approaches to external power, such as pleated pneumatic actuators, SpringActive’s robotic tendon designs, Össur’s POWER KNEE™, and the antagonistic pairing of series elastic actuators in a more energetically conservative approach to an externally powered knee joint (The O&P EDGE, September 2013). This installment describes the progress that is being made as researchers work toward control of powered prostheses through the user’s myoelectric signals.

The Vanderbilt Approach

Within the modern era of such efforts, the first study to describe volitional control of a powered prosthesis was published by the same team that developed the Vanderbilt knee.1 Because the Vanderbilt knee already uses various inputs to control its activity during weight bearing activities such as standing and walking, the aim of these initial efforts was to evaluate the potential control of the powered knee during non-weight bearing activities.

The authors described their method of placing electrodes over the affected quadriceps and hamstrings of three seated subjects with transfemoral amputations who then used their electromyographic (EMG) signals to regulate the movement of a powered knee unit mounted to an adjacent bench in a manner consistent with the subjects’ seated positions. During the experiments, subjects were shown a computer monitor that depicted a real-time desired knee angle trajectory along with the immediate knee angle of the prosthesis as measured by its joint angle sensor. Subjects were asked to try and track various types of knee joint movements using this feedback.

During each of three training sessions, the subjects performed this action across four different patterns or trajectories. In trajectory A, the subjects were asked to quickly change their knee joint angles, in increments ranging from 8 degrees to 45 degrees, and hold these positions for five to ten seconds. Trajectory B was made up of various sloped trajectories, which measured the subjects’ abilities to move the prosthesis at various constant velocities. Trajectories C and D were sinusoidal waves representing two different speeds as the subjects attempted to move the leg up and down smoothly. At the conclusion of the final training session, each subject completed each of the four trajectories and tracking errors were derived. While there were subtle differences between subjects and trajectories, the collective mean error rate was only 6.2 degrees.1

To place this value in context, the subjects repeated the exercise using their sound-side knees with an instrumented knee brace, trying to mimic the desired knee motions through the same four trajectories. The error rate for this exercise was 5.2 degrees, suggesting that the EMG tracking of the powered knee was, on average, within a single degree of that obtained with the sound knee.

Virtual Control at RIC

This study was quickly followed by a research letter published in The Journal of the American Medical Association detailing similar efforts taking place at the Rehabilitation Institute of Chicago (RIC), Illinois.2 In these trials, four subjects with transfemoral amputations were seated with electrodes placed over nine muscle groups on their residual limbs: the semitendinosus, sartorius, tensor fasciae latae, adductor magnus, gracilis, vastus medialis, rectus femoris, vastus lateralis, and long head of the biceps femoris. As with the previous study, directions were provided through computer software. In this case, subjects were instructed to complete the following motions: flexion and extension of the knee, plantarflexion and dorsiflexion of the ankle, internal rotation of the tibia and femur, and relaxation. The computer used in the trial learned to recognize these patterns though pattern recognition algorithms.

With pattern recognition established, subjects were asked to complete several iterations of two different virtual tasks. The first was a virtual prosthesis with 2 degrees of freedom, which required the subjects to flex and extend a virtual knee and plantarflex and dorsiflex a virtual ankle. The second was a virtual prosthesis with 4 degrees of freedom in which the subjects were asked to perform the previous two tasks along with femoral and tibial rotation.2 Performance was characterized according to the accuracy of the EMG classifications, the times required to complete the motions, and the motion completion percentages. Accuracy rates were high, averaging 91 percent with the 2 degrees of freedom task and 87 percent with the more complex 4 degrees of freedom task. Completion percentages were also high at 97 percent and 85 percent respectively. It is significant that although none of the subjects had undergone RIC’s targeted muscle reinnervation (TMR) procedure, they were still able to generate EMG signals discrete enough to control both knee and ankle movements.2

University of Michigan: Weight Bearing In-Socket EMGs

With two centers having examined EMG signals in non-weight bearing activities without a socket interface, the next step was to determine if EMG signals could be recorded within a weight bearing socket environment. Researchers from the University of Michigan (U-M), Ann Arbor, were the first to report on this progress.3

Their investigation was carried out with 12 subjects with transtibial amputations and 12 able-bodied subjects who acted as a control group. Surface electrodes were applied to the tibialis anterior and the medial and lateral heads of the gastrocnemius on the residual limbs of the individuals with amputations and to the right legs of the control subjects. Once electrode locations on the residual limbs were verified via EMG signals, the amputee subjects donned their liners and sockets. In addition to the lower-leg EMG recordings, surface electrodes were applied to targeted muscles of the upper leg including the vastus lateralis, rectus femoris, biceps femoris, and gluteus medius.3

The first stage of the trial sought to determine if the subjects with amputations could isolate contractions of the tibialis anterior and gastrocnemii. While most subjects were able to do so adequately, consistent with collective observations of myoelectric signals in upper-limb prosthetic applications, some subjects had difficulty isolating the individual muscle groups and performed co-contractions instead. Others were unable to maintain adequate muscle activation levels.3

In the second phase of the trial, muscle activity was recorded during treadmill walking at progressively faster speeds. The authors reported that the activation patterns of the lower-leg muscles in the amputee cohort were characterized by high intersubject variability and differed substantially from those observed in the control subjects. However, the interstride variability of these activation patterns among those with amputations was similar to the consistency observed within the control subjects. Restated, while the muscle activation patterns were quite unique from amputee to amputee, they were fairly consistent from stride to stride for each individual subject.3 By contrast, there was considerably less variability between the muscle activation patterns recorded on the upper leg for the amputee cohort compared to the controls.3

The implications of this pilot-level research are that, while EMG signals can be recorded within a transtibial prosthetic socket during ambulation, there appears to be considerable variability among the individual activation patterns. Further, the ability of individuals with transtibial amputations to activate a powered prosthesis volitionally through isolated EMG may vary from individual to individual. What is not known is whether the consistencies of both phenomena might improve with iterative training and biofeedback while subjects watch their powered prosthesis move.3

Clarkson University and the Transfemoral Myo-Socket

The myosignals recorded in the U-M study were obtained from electrodes that were positioned over the appropriate muscle bellies prior to the donning of the liner and the socket, with trailing electrical wires exiting the proximal aspect of the socket and liner. Daily use of a volitionally controlled, externally powered prosthesis would require the integration of electrodes within the socket wall itself such that donning the prosthesis puts the appropriate muscle bellies in contact with myoelectrodes enclosed within the socket. This was first accomplished in a case study authored by researchers from Clarkson University, Potsdam, New York.4

The study follows the progression of an otherwise healthy subject with a unilateral transfemoral amputation. Using palpation, the authors were able to identify contractile activity in three residual knee extensors (vastus lateralis, rectus femoris, and vastus medialis) and one knee flexor (biceps femoris). With the muscle bellies identified, the subject donned a clear test socket to identify the appropriate positions for the socket-mounted electrodes. This done, commercially available Ottobock electrodes with silicone aprons (MYOBOCK® Suction Socket Electrode 13E202), commonly used in upperlimb applications, were installed within the test socket. Two of the knee extensors (rectus femoris and vastus lateralis) and the knee flexor produced usable EMG signals at this stage.4

Though the individual extensor muscles could not be isolated and co-contraction was common with contraction of the biceps femoris, the patient was able to produce EMG signals on command. At this stage, the diagnostic myo-socket was attached to the subject’s legacy prosthesis, which included an Ottobock C-Leg, to record EMG activities during ambulation. While significant co-contractions persisted with ambulation, there was sufficient muscle activity and separation at this evaluation stage.

The myo-socket was then attached to an externally powered active-knee prosthesis prototype. At this stage, the EMG input of the rectus femoris was abandoned for the sake of simplicity, leaving the vastus lateralis and the biceps femoris to regulate the knee joint movement. The subject began his use of the powered prosthesis using non-weight bearing motion-tracking tasks similar to those described at the beginning of this article. This occurred over four three-hour sessions. As the subject became more adept at myoelectric control of the prosthesis, he was moved to weight bearing tasks of quiet standing (i.e. standing still), squatting, and sit-to-stand transfers. When proficiency with these tasks was observed, he was moved to level ground ambulation within the security of a suspension harness. Though significant co-contraction activity persisted, the subject was able to control the prosthesis during walking through myo-EMG signals.4

EMG control of the BiOM

It was near this time that researchers from the Biomechatronics group at the Massachusetts Institute of Technology (MIT) Media Laboratory, Cambridge, reported on their efforts to use EMG as a control mechanism within the BiOM powered ankle-foot system.5 In its commercially available version, the amount of positive ankle work produced during a given step is regulated internally according to ankle joint moments. In this report, the BiOM control mechanism was supplemented with real-time EMG input. A case subject wore a custom liner with conductive fabric electrodes positioned over the plantar flexors of his residual limb and donned the adapted BiOM. He walked at three predetermined speeds in two conditions. In the first, his residual gastrocnemius was idle, such that the regulation of the BiOM was controlled by its standard internal control system. In the second, the subject fired his residual limb muscles during the controlled dorsiflexion of midstance and terminal stance to modulate the responsiveness of the powered foot. Ultimately, the performance of the foot under the two control strategies was similar, suggesting a potential control strategy that would be more amenable to real-time transitions in transtibial walking activities through activation of the residual limb muscles.5

The New England Journal of Medicine Updates the State of the Science

The final case study considered in this article comes from The New England Journal of Medicine. The September 2013 report describes the case of an individual with a unilateral knee disarticulation amputation who underwent TMR of his semitendinosus and biceps femoris at the time of his amputation. It reports upon the results that occurred when this subject walked over challenging terrains using the Vanderbilt prosthesis, coupled with active EMG controls from RIC’s pattern recognition of his natively innervated muscles, and further supplemented by the additional control inputs derived from his reinnervated residual muscle bellies.6 While the 13 mechanical controllers of the Vanderbilt powered prosthesis allowed this subject to reasonably transition between walking tasks on level ground, ascending and descending ramps, and negotiating stairs, a 13 percent movement misclassification error rate by the powered knee was recorded. By augmenting the control inputs of the knee with active EMG input from the natively innervated residual limb musculature, this error rate was reduced to 2.2 percent. When the additional inputs of the reinnervated muscle bellies were included within the control strategy, the error rate was further reduced to 1.8 percent. Critical errors, defined as errors that might cause the patient to fall, were infrequent using the mechanically regulated prosthesis but did occur. With the EMG-enhanced control strategy, no such errors occurred despite the continued transitions between very different walking environments.6


The promise of real-time myoelectric control of externally powered lower-limb prostheses is encouraging due to both the pace of the technology’s advances and the number of institutions that are pursuing this objective. In just a few years, the science has gone from non-weight bearing control of virtual prostheses to real-time control of ambulatory devices across varied walking surfaces. While barriers to the broad application of these expensive solutions exist, these research efforts give some indication of what might be possible in the coming era of externally powered prostheses.

Phil Stevens, MEd, CPO, FAAOP, is in clinical practice with Hanger Clinic, Salt Lake City, Utah. He can be reached at


  1. Ha, K. H., H.A. Varol, and M. Goldfarb. 2011. Volitional control of a prosthetic knee using surface electromyography. IEEE Transactions on Biomedical Engineering 58 (1):144-51.
  2. Hargrove, L. J., A. M. Simon, S. B. Finucane, and T.A. Kuiken. 2011. Real-time myoelectric control of knee and ankle motions for transfemoral amputees. JAMA: The Journal of the American Medical Association 305 (15):1542-4.
  3. Huang, S., and D. P. Ferris. 2012. Muscle activation patterns during walking from transtibial amputees recorded within the residual limb-prosthetic interface. Journal of Neuroengineering and Rehabilitation 9:55.
  4. Hoover, C. D., G. D. Fulk, and K. B. Fite. 2012. The design and initial experimental validation of an active myoelectric transfemoral prosthesis. Journal of Medical Devices 6 (1).
  5. Wang, J., O.A. Kannape, and H.M. Herr. 2013. Proportional EMG control of ankle plantar flexion in a powered transtibial prosthesis. Proceedings of the IEEE International Conference on Rehabilitation Robotics. Jun:1-5.
  6. Hargrove, L. J., A. M. Simon, A. J. Young, R. D. Lipshutz, S. B. Finucane, D. G. Smith, and T. A. Kuiken. 2013. Robotic leg control with EMG decoding in an amputee with nerve transfers. The New England Journal of Medicine 369:1237-42.

Bookmark and Share