Realizing the Promise of Robotic Leg Prostheses

See allHide authors and affiliations

Science Translational Medicine  06 Nov 2013:
Vol. 5, Issue 210, pp. 210ps15
DOI: 10.1126/scitranslmed.3007312


Recent advances in robotics technology are enabling the emergence of robotic leg prostheses that can emulate the full biomechanical functionality of the healthy limb. The behavior of such prostheses is software-controllable, in an analogous manner to the way in which the central nervous system controls the human musculoskeletal system. Although these prostheses have the capability of reproducing the biomechanical behavior of the healthy limb, their ability to do so is a function of how well the prosthesis control system coordinates the movement of the leg with the movement of the user.


Major lower-limb amputations are most commonly of the transtibial (below-knee) or transfemoral (above-knee) type. Consequently, the two most common lower-limb prosthetic components are prosthetic knee joints and prosthetic feet (which are intended to replace the biomechanical function of both the foot and ankle joint). Prosthetic feet have, for the past few decades, consisted primarily of carbon-fiber leaf springs, whereas prosthetic knees have consisted primarily of rotational dampers that provide resistance to movement between the thigh and shank segments. Such knee damping provides a reasonable emulation of healthy knee biomechanics during the swing phase of walking. Some knee units have additionally incorporated an electronic control system in order to modulate the degree of damping at the knee. The essential physical behaviors of conventional knee and ankle joints are determined by their physical design, even when endowed with some degree of electronic modulation. The ankle is a spring, whereas the knee is a damper. In contrast, the musculoskeletal system is capable of a wide variety of physical behaviors and, at any given moment, behaves in a manner governed by the nervous system, depending on the activity in which the user is engaged.

Several recent technological advances have made robotic prostheses viable, including advances in lithium-ion and lithium-polymer batteries; rare-Earth magnet brushless motors; micro-electromechanical systems (MEMS)–based sensors, particularly inertial measurement units (IMUs); low-power complementary metal oxide semiconductor (CMOS) design of microcontroller architectures; and high-power metal oxide semiconductor field-effect transistors (MOSFETs). Rather than provide a predetermined physical behavior at each joint, these robotic limbs are developed with a structure quite similar to that of the human neuromuscular system. Effectively, the motors and structure of the prosthesis provide the equivalent of the musculoskeletal system; the sensors and microcontroller are the equivalent of the neural control system (peripheral and central); and the battery is the equivalent of the metabolic power supply (digestive and circulatory systems). In contrast with conventional prosthetic legs, which are capable of reproducing a relatively small subset of the behaviors of the healthy limb, a robotic prosthesis can reproduce any motor behavior exhibited by the healthy limb. The ability to do so, however, is highly dependent on the ability of its control system to coordinate movement with the user’s central nervous system (CNS).

Although individual embodiments vary, robotic leg prostheses contain mechatronic components equivalent to the biological components of the healthy limb, as well as the equivalent of biological components that lie outside the limb. The essential components of a robotic leg prosthesis are illustrated in Fig. 1. An electric motor is incorporated for each generalized agonist/antagonist muscle pair being replaced by the prosthesis. A typical set of proprioceptive sensors includes sensing of the joint angles and angular velocity, which might be combined with kinesthetic sensors, such as those for measuring forces or torques within the prosthesis. A typical robotic prosthesis additionally includes an IMU, consisting of a three-axis accelerometer, combined with a three-axis gyroscope—the combination of which provides information regarding the movement of the prosthesis through space. As such, this sensor provides information that, in the biological system, would be derived from the proprioceptive sensors in the limb, combined with inertial information from the vestibular system. The combined set of information from the proprioceptive, vestibular, and kinesthetic sensors is fed into the microcontroller, which provides the equivalent function of the CNS. The microcontroller contains a coordination controller, which determines the immediate physical behavior of the joint based on the set of sensory information it receives. Upon determining the appropriate behavior, the microcontroller uses the equivalent of efferent nervous channels to affect the appropriate behaviors from the joint motors. Last, because the robotic limb is isolated from the metabolic power supply (the user’s circulatory system), the prosthesis must carry its own power supply, which is most often in the form of an electric battery (Fig. 1). Some robotic leg prostheses that exemplify this construction have been described in the recent literature (1–5).

Fig. 1 The anatomy of a robotic leg.

The major defining components of a robotic leg prosthesis and their biological equivalents are shown.



Although robotic prostheses can potentially restore healthy biomechanical function to a much greater extent than conventional prostheses can, their ability to do so depends heavily on the extent to which they operate in concert with the user’s intact neuromuscular system. The need for a system and structure that integrate the movement of the prosthesis with the movement of the user is an issue specific to robotic limbs. The only motive power source for a conventional, energetically passive prosthesis is the user. As such, a passive prosthesis is inanimate; it cannot move without the user swinging his or her residual limb to propel the prosthetic leg forward. Because the user must “sling” the leg around, such prostheses fundamentally move in concert with the user—albeit in a biomechanically deficient manner.

Unlike an energetically passive prosthesis, a robotic prosthesis contains its own motive power source and can both act and react. The nature of the control interface between the prosthesis and user therefore becomes substantially more important with a robotic leg. Such legs have the capability to emulate healthy biomechanical function, but this capability also enables them to move independently of the user, potentially in discord with the user’s movement or intent. This is again analogous to the function of the biological system, in which the ability of the musculoskeletal system to provide biomechanical functionality is only as good as the coordination and control provided by the neuromuscular system (primarily the CNS). A robotic leg interface and control system must provide appropriate coordination within an activity, such as walking, and must additionally recognize a user’s intent to perform a given activity or change from one activity to another—for example, from walking to stair ascent. Intent represents neural control function primarily performed by the spinal cord, brain stem, and potentially cerebellum, whereas change represents more cerebellar and cortical functions.

Because the prosthesis control system must act in unison with the CNS, it requires some ability to measure information from it (it should be neurally integrated in some form). Several approaches exist to measure information from the CNS, which can be characterized by widely varying degrees of invasiveness: in order of increasing invasiveness, (i) a physical sensor interface, (ii) a surface or implantable electromyography (EMG) interface, (iii) an implantable peripheral nerve interface, and (iv) an implantable cortical interface. A physical sensor interface uses only physical (as opposed to physiological) sensors on the prosthesis, which can measure movement associated with the prosthesis, in addition to forces and torques within it. Although these sensors do not provide a direct measure of neural information, the movement and loads associated with the prosthesis are strong functions of the user’s movement, which is a direct result of CNS activity. The control interface can also, or alternatively, measure electrical activity from muscles in the residual limb (or elsewhere) via electromyography (either from the skin surface or via implanted electrodes). Such muscular output is clearly linked directly to the CNS. Alternatively, electrodes can be implanted in the peripheral nerves, which have the potential advantage (relative to electromyogram measurement) of higher specificity in the measurement of motor commands and additionally provide the possibility of relaying afferent information to the CNS. Last, electrodes can be implanted in the motor cortex of the brain, which arguably provide less specificity (relative to peripheral measurements) but perhaps have the potential to provide a more integrated measurement of movement intent. Like peripheral implants, (sensory) cortical implants additionally provide the potential to relay sensory information to the CNS. Approaches that entail a greater degree of invasiveness must obviously justify the invasiveness with substantive functional advantage, relative to less invasive methods.

Whichever measurement method (or methods) is used to acquire information regarding the user’s movement and movement intent, this information must be used within a cooperative control framework. Some cooperative control frameworks have been described that use physical sensor interfaces (1, 48) or the combination of physical and EMG sensing within the control structure (911). Although these frameworks have demonstrated effective control of robotic prostheses, the development and validation of such control frameworks is a nascent field of research. As these control frameworks become increasingly able to coordinate the action of the robotic prosthesis with the movement intent of the CNS, we expect that the full promise of robotic prostheses will increasingly be realized and translated to improving patient mobility and quality of life.


Among the most common activities in human locomotion is level walking. Because a robotic prosthesis can emulate all aspects of muscular function, the prosthesis is able to reproduce a number of biomechanical features during level walking that are not reproduced by conventional prostheses, including a biomechanically appropriate loading response at heel strike and a biomechanically appropriate push-off (at the ankle) in late stance. The loading response consists of the combination of knee flexion and ankle plantarflexion at heel strike (both characterized by energy absorption), followed by knee extension (characterized by energy return), which provides shock absorption that enhances intact joint health and establishes a foot flat condition, which decreases the likelihood of slip. Ankle push-off in the late stance phase (characterized by power generation) serves biomechanically to initially propel the center of mass of the body forward and subsequently impart momentum to the swing leg to establish the initial conditions for a ballistic swing phase. In the case of a passive prosthesis, the amputee compensates for the loss of push-off by exerting three to four times the hip torque and power during late stance (12) and sometimes vaulting on the sound leg to compensate for the underpowered swing phase (sound-side plantarflexion during mid-stance, which increases the prosthetic foot clearance in swing). As described in (13), the compensatory power provided by the hip cannot create the equivalent forward impulse of powered push-off; therefore, the efficiency of walking is decreased, and (assuming a unilateral amputation) the user must provide a greater push-off with sound side to maintain a given gait speed (to overcome the energy lost at heel strike in both limbs).

Restoring powered push-off at the ankle in late stance would presumably offer a number of biomechanical benefits, including decreased ipsilateral hip effort on the prosthetic side, which would presumably mitigate stress on the hip joint and help prevent long-term musculoskeletal degeneration of the joint. Additionally, one would expect biomechanically appropriate push-off to lower the metabolic cost of transport, for two reasons. The most obvious is that the powered leg has an external power source, which is contributing to the energy required for locomotion but not using the metabolic energy source. Second, as asserted above, imparting the power for push-off at the ankle enhances the biomechanical efficiency of walking. Some preliminary investigations of the energy cost of walking with powered push-off with transtibial subjects have supported these assertions (4). Among the health and functional benefits of providing these biomechanical features to the user are decreased hip effort, decreased metabolic cost of transport, increased step symmetry, and increased self-selected walking speed.

One of the most substantial benefits of a robotic prosthesis is the ability to provide appropriate biomechanics across a variety of activities and terrain types. Recall that the behavior of passive prostheses is fundamentally restricted (to store or dissipate energy) by the nature of their hardware. Since passive prostheses have relatively little functional flexibility, they are generally optimized for level walking. As a result, the biomechanical function provided by passive prostheses is substantially compromised during other activities and on other terrain. For example, biomechanically healthy stair and steep-slope ascent requires considerable net power generation at the knee joint, which is absent in the case of energetically passive knee joints. By restoring biomechanically appropriate power generation at the knee and ankle joints during such activities, one would expect a significant decrease in metabolic cost of transport relative to energetically passive prostheses. Further, the step-over biomechanics enabled by a powered prosthesis (for example, during stair ascent) obviously promotes an enhanced symmetry of gait, which is an asset to many amputees. The ability of a robotic prosthesis to restore appropriate biomechanical behavior has been shown in (6, 7).

In the case of slope and stair descent, the behavior of both the knee and ankle joints is principally dissipative. The typical passive prosthetic ankle/foot complex, however, is designed to store rather than dissipate energy and, as a result, is generally unable to provide appropriate biomechanical behavior during slope and stair descent. The inadequacy of such ankle behavior induces associated compensatory actions, such as the strategy of placing the prosthetic foot across the edge of a step in stair descent (as a substitute for ankle dorsiflexion) and relying on the sound leg to perform the majority of energy dissipation required to descend a stair or slope in a controlled manner. Thus, when walking down slopes and stairs, one would expect enhanced stability and decreased stress on intact joints.

In addition to better restoring the biomechanics of healthy gait across a variety of terrain and activity types, robotic leg prostheses may also reduce the incidence of falls in the lower limb amputee population. Specifically, recent studies indicate that the annual incidence of falls in the lower-limb amputee population exceeds that of the elderly population, the rate of seeking medical attention as a result of such falling is comparable with that of the institution-living elderly, and the incidence of falling (and consequently requiring medical attention) is higher in younger amputees than in older amputees, presumably because younger amputees are less restrained in their choice of activities and terrain (1417). Robotic prostheses provide several behaviors that could reduce the incidence of falls. First, the improved biomechanics provided by robotic prostheses will promote robustness in locomotion. For example, one would expect push-off to enhance the robustness of the swing phase and thus decrease the incidence of scuffing. If scuffing were to occur during swing, a robotic knee has the capability of providing the equivalent of a concentric muscle contraction, to reflexively restore the angular momentum of the swing leg and thus the likelihood of resultant falls. Similar logic applies to a greater degree in upslope walking. In the case of slope and stair descent, the ability of the foot to conform to the ground and of the ankle to provide dissipative behavior promotes stable and controlled descent, which in turn would presumably decrease the incidence of falls. In the case of standing, an enhanced support polygon, enabled by the ability of a robotic leg to conform to and provide body weight support on varying ground topography, would also presumably lead to a decreased incidence of imbalance and thus a decreased incidence of falls. Furthermore, in the event of a perturbation the healthy limb uses recovery reflexes (such as a stumble recovery reflex) to reject or mitigate the perturbation, so that a stumble is less likely to result in a fall. Such reflexes, which are fundamentally active responses, can be programmed into a robotic prosthesis, to provide the amputee with improved tolerance for perturbations.

Last, as previously alluded to, the biomechanical deficiencies of an energetically passive prosthesis generally necessitate a number of compensatory actions from the intact joints, which increases the stress on these joints and presumably exacerbates the propensity for musculoskeletal degeneration. A robotic prosthesis would presumably mitigate such degeneration because the restoration of healthy biomechanics in the prosthetic joints precludes the need for the variety of compensatory actions required by conventional prostheses. Thus, although clinical studies have yet to be conducted in this regard, one would expect the overall level of compensatory effort and loading in the intact joints to decrease when using a robotic prosthesis—and therefore also the incidence and/or severity of intact joint degeneration.


The translation of robotic leg prostheses onto the lower limb prosthetics market has recently begun, and the process of translation is expected to continue over the next few decades. Specifically, robotic leg prostheses with physical sensor interfaces have started to emerge on the commercial market, and studies regarding biomechanical benefits of these prostheses have appeared in the literature (18, 19). As previously discussed, as control frameworks for integrating the prosthesis control system with the CNS become more capable, so also will the functionality of the prostheses. Further, enhanced neural integration may offer enhanced functionality and control, particularly if sensory information, such as proprioceptive and kinesthetic information, can be conveyed in a meaningful way to the user. Such capability will depend on the continued development of emerging neuroprosthetic methods, such as those described in (2022).

In addition to the continued development of control interfaces, robotic prostheses will have to gain regulatory approval. With regard to the U.S. market, single-joint prostheses are usually considered as Class 1 devices by the U.S. Food and Drug Administration (FDA). A conventional transfemoral prosthesis has traditionally been constructed by combining two single-joint prostheses (a separate knee and ankle joint), and thus the composite prosthesis remains classified as a Class 1 device. In the case of a robotic prosthesis, the knee and ankle joints should move together in a coordinated fashion; in order to do so, they must be at least electronically connected. Such a multijoint coordinated device is considered by the FDA to be a Class 2 device. In the absence of a clear predicate device, evidence will need to be established regarding the safety and efficacy of such prostheses in order to gain FDA approval.

Robotic prostheses are inherently complex devices. The clinicians that prescribe prostheses are not trained in the field of robotics. In order to establish robotic prostheses as a viable technology for people with lower-extremity amputation, either these devices must be rendered usable for a clinician without robotics training; the clinician must receive robotics training; or the manufacturer must provide some viable combination of device usability and clinician training. If these translational issues can be successfully addressed, robotic leg prostheses should offer to people with lower-extremity amputation an unprecedented level of functionality.


  1. Competing interests: M.G., B.E.L., and A.H.S. hold patent applications through Vanderbilt University that have been licensed to Freedom Innovations, a United States–based prosthetics manufacturer.
View Abstract

Stay Connected to Science Translational Medicine

Navigate This Article