Scheduled System Maintenance on May 29th, 2015:
IEEE Xplore will be upgraded between 11:00 AM and 10:00 PM EDT. During this time there may be intermittent impact on performance. We apologize for any inconvenience.
By Topic

IEEE Quick Preview
  • Abstract

SECTION I

INTRODUCTION

THE MOST common approaches adopted today for the recovery of locomotion skills after a stroke take into account different physical treatments combining strength training [1], task oriented exercises [2], electrical stimulation [3], treadmill [4], and robotic aided therapy [5]. According to the literature, the combination of physiotherapy and robotic-based training is likely to improve the walking abilities of post-stroke patients, because it enhances independent walking, increases speed and endurance, without worsening gait quality [4], [5], [6]. Conversely, other authors have not observed significant locomotion-related modifications (e.g., cadence, stride length, symmetry, endurance) when robotic-based therapy is compared to traditional approaches [7], [8], [9]. Alternatively, it is also reported that although robotic-based therapy produces encouraging effects on post-stroke patients during acute and subacute phases, it does not significantly influence patients' performance after three months from the trauma [5].

These discrepancies can be partially ascribed to the many methodological differences in the studies such as duration and intensity of treatments, contribution of physiotherapy, and severity of the pathology. Moreover, literature agrees on the need to better understand the influence of the intervention time on the effectiveness of intensive practice in stroke patients. Many recent studies have especially highlighted that early and intensive rehabilitative treatments can significantly favor the functional recovery of post-stroke patients, such that some authors strongly recommend that neurorehabilitative therapies be provided as early as the first days after the trauma, after adjusting for functional status and severity of stroke [10], [11]. In this respect, a number of research groups have observed that early mobilization seems to be able to enhance future gait independence [12], may reduce depressive symptoms in stroke patients [13], and is an important predictor of stroke functional outcomes [14].

The present paper introduces a new robotic platform, named NEUROBike, which has been developed to provide neurorehabilitative treatments to bedridden patients who have experienced a stroke. It has been designed with a twofold goal: on the one hand, since patients often spend more than half day in bed during the acute phase [15], NEUROBike can allow them to start practicing several exercises at an early stage according to the severity of the pathology, and the intensity required by the programmed therapy. On the other hand, NEUROBike provides intense treatments which can elicit functional motor schemes, comparable to those involved during daily activities (e.g., walking, sitting), in order to promote neural plasticity by a suitable sensory feedback. Passive and active exercises can be selected with NEUROBike according to the patient's clinical situation in order to maximize clinical outcome.

SECTION II

DESIGN CRITERIA

The main specification underlying the design of NEUROBike was that the platform could actively control angular excursions at the hip, knee, and ankle joints in the sagittal plane, in order to elicit flex-extensor motor schemes comparable to those adopted during natural motor tasks.

Another required feature consisted in interacting with patients only through footplates, in order to reproduce pseudo-natural sensory feedback while carrying out the exercises. In particular, robotic platforms currently adopted in clinical fields, are provided with cuffs wrapping limb segments, which represent the main mechanical interface between patients and robot, and allow a direct control of limb joint trajectories [16], [17]. Nevertheless, due to pathology-related abnormal muscle co-contraction, this solution does not suitably distribute mechanical stress at the limb–robot. This may lead to skin irritation [18] and an unnatural sensory feedback which may negatively interfere with the neurorehabilitative treatment.

A. Mechanical Structure

From the mechanical viewpoint, the first prototype of NEUROBike has already been introduced in a previous paper [19]. Briefly, it consists chiefly of two symmetrical cardinal manipulators provided with motor driven footplates representing the robot–patient interface (Fig. 1). Each manipulator is characterized by 3 degrees-of-freedom (DoF) allowing the foot to be translated in the sagittal plane and rotated with respect to the medial-lateral axis.

Figure 1
Fig. 1. CAD representation of NEUROBike. In order to make reading easier, only the manipulator attached to the left limb is reported. The figure also shows the reference frame adopted for analyzing leg joint kinematics during the experimental sessions. In particular, the Formula${\rm x}$ axis is horizontally oriented toward the head, the Formula${\rm y}$ axis is vertical and the Formula${\rm z}$ axis is orientated according to the right hand role. The center of the reference frame is located in the middlepoint between right and left ASISs. Labels in the figure refer to: 1. seat; 2. frame for pulleys, 3 and 4. linear guides leading respectively vertical and horizontal movements of the foot, provided with their related brushless motors (5 and 6); 7. end effector provided with the motor leading the rotation of the foot (8).

The maximum load considered for designing the platform consisted of a person with mass of 130 kg and 1900 mm tall, walking at 1.3 m/s. After estimating the geometrical and inertial features of leg segments by using tables reported in literature [20], kinematics and forces at the end-effector were computed. The results are shown in Fig. 2.

Figure 2
Fig. 2. The figure shows: on the left, both components of linear velocity and angular velocity of the end-effector estimated by the inverse kinematics approach; on the right estimated forces and torque required to move the limb of a person with mass of 130 kg and 1900 mm tall, lying horizontally, and moving his/her lower limbs as during walking at 1.3 m/s. Leg joints were modeled as frictionless hinges.

According to simulations (Fig. 2), for each manipulator, two identical sinusoidal brushless servomotors (Axor Industries SSAX 100; nominal speed: 3000 rpm; power: 2400 W), were chosen for translation of the end-effector in the sagittal plane. They are provided with planetary gearheads (Axor Industries REX 105; ratio 4/5), and motor drivers (Axor Micro B NET). Each motor is connected to a linear guide (Movopart MR-CB200), allowing maximum excursion of 1000 mm along the Formula${\rm X}$ axis, and 1200 mm along the Formula${\rm Y}$ axis. A dc motor (Maxor Motor RE-40; nominal speed: 7000 rpm; power 150 W), provided with planetary gearhead (Maxor Motor GP-42C; ratio 4/1), and motor driver (Maxor Motor ADS-50) rotated the end-effector about the medial-lateral axis. The structural frame is mainly constituted of aluminum extrusions.

NEUROBike is also provided with knee straps (Fig. 1), which passively connect both knees by using a cable that runs through a couple of pulleys positioned above the platform. Their purpose is to avoid movements of the legs in the frontal plane and to prevent knee hyperextension. Specifically, during walking-like cyclic therapy, the knee is supposed to be almost fully extended during the middle stance, entailing a kinematic singularity which may favor hyperextension of the knee in patients with a weak control of knee flexor muscles. To prevent this, a cable of suitable length was connected to both knees, representing a geometrical constraint. The cable helped the leg to be lifted during the middle of the stance thanks to the pulling action of the contralateral leg.

B. Control Architecture

The control architecture of NEUROBike is completely based on open-source software platforms. Specifically, programs developed using SCILAB/SCICOS (http://www.scilab.org; http://www.scicos.org) platforms, running on the operative system Ubuntu (8.04) with patched RTAI kernel (http://www.rtai.org). CoMeDi drivers (http://www.comedi.org) allow interfacing of control algorithms and servomotors through two S626 Sensoray I/O boards. Encoders and current sensors embedded in the servomotors close the feedback control loop. Fig. 3 briefly shows the control architecture designed for NEUROBike.

Figure 3
Fig. 3. Control architecture. Labels SW and HW refer respectively to software and hardware.
SECTION III

KINEMATIC MODELS UNDERLYING LEG JOINT TRAJECTORIES

As already stated, NEUROBike was designed to enable movement of the legs featuring joint angular excursions comparable to those of daily activities such as walking or sit-to-stand. Although these motor tasks have been widely described by many authors [20], [21], to the best of our knowledge, kinematic models have never been reported in literature, limiting the exploitation of the results. The aim of this subsection is to describe the kinematic models implemented in NEUROBike control algorithms.

C. Kinematic Model During Walking

Concerning locomotion, we developed a model providing estimated leg joint angular excursions related to a selected walking speed, based on the Fourier sum decomposition of data recorded while healthy subjects underwent gait analysis over a wide range of speeds [22]. This approach was partially introduced in a previous work dealing with the tuning of NEUROBike's PID controllers [23]. Briefly, the trajectory of 20 markers suitably placed (Davis protocol) on the body of nine (five female and four male) young healthy adults while walking on a treadmill, was captured by using a five camera Formula${\rm ELITE}_{\rm PLUS}$ System (BTS, Milan, Italy) with a sample frequency of 100 Hz. Subjects walked at five different speeds calculated in accordance with the Froude number (Formula${\rm Fr}=0.050$; 0.075; 0.100; 0.150; 0.175). Raw data were low pass filtered with a zero lag Butterworth filter with cut off at 10 Hz, and expressed in the reference frame related to the pelvis (Fig. 1). After taking into account data related to 10 strides for each subject and each speed, the respective hip, knee, and ankle angular excursions in the sagittal plane were calculated by using a traditional approach [20]. The data were then post-processed as follows.

For each stride, leg joint trajectories were decomposed into ten harmonics by using the Fourier series of periodic functions in which the period of the fundamental harmonic coincided with the duration of the gait cycle. To this regard, an idealized version of the kinematic trajectories for the Formula$i$th stride and the Formula$j$th leg joint can be described by Formula TeX Source $${f_{ij}}\left(t \right) = {\mu _{ij}} + \sum\limits_{k = 1}^{10} {\left[ {{\alpha _{ij,k}}\cos \left({{{2\pi k} \over {{T_{i}}}}t} \right) + {\beta _{ij,k}}\sin \left({{{2\pi k} \over {{T_{i}}}}t} \right)} \right]} \eqno{\hbox{(1)}}$$ where

  • Formula${f_{ij}}\left(t \right)$ is the time (Formula$t$) based function describing the trajectory of the Formula$j$th leg joint (Formula$j=1$,2,3 refers, respectively, to hip, knee, or ankle) related to the Formula$i$th gait cycle (Formula$i=1,2,\ldots ,20$, refers, respectively, to all bilateral strides);

  • Formula${\mu _{ij}}$ is the overall mean of Formula${f_{ij}}\left(t \right)$;

  • Formula${\alpha _{ij,k}} = ({2 / {{T_{i}}}})\int _{0}^{{T_{i}}} {{f_{ij}}\left(t \right)\cos \left({({{2\pi k} / {{T_{i}}}})t} \right)dt}$ and Formula${\beta _{ij,k}} = ({2 / {{T_{i}}}})\int _{0}^{{T_{i}}} {{f_{ij}}\left(t \right)\sin \left({({{2\pi k} / {{T_{i}}}})t} \right)dt}$ are the Formula$k$th Fourier coefficients;

  • Formula${T_{i}}$ is the duration of the gait cycle.

For all harmonics, the averaged values of Formula${\mu _{ij}}$,Formula${\alpha _{ij,k}}$, Formula${\beta _{ij,k}}$, and Formula${T_{i}}$, across all strides, were computed and considered representative of each subject walking at a specific speed. Then, the linear regression model fitting these parameters with respect to the speed was estimated and considered as representative of the relationship between them and the walking speed. According to this approach, once the walking speed is defined (i.e., independent variable Formula$v$), each parameter (i.e., dependent variables Formula$T$, Formula${\alpha _{j,k}}$, Formula${\beta _{j,k}}$ with Formula$j=1$,2,3, Formula$k=1,2,\ldots, 10$) can be estimated by means of a linear equation where slope and intercept are those reported in Tables IIV.

Table 1
TABLE I SLOPE AND INTERCEPT FOR Formula$T$, Formula$\mu_{A}$, Formula$\mu_{K}$, AND Formula$\mu_{H}$
Table 2
TABLE II SLOPE AND INTERCEPT FOR FOURIER COEFFICIENTS RELATED TO ANKLE PLANTAR-DORSIFLEXION
Table 3
TABLE III SLOPE AND INTERCEPT FOR FOURIER COEFFICIENTS RELATED TO KNEE FLEX-EXTENSION
Table 4
TABLE IV SLOPE AND INTERCEPT FOR FOURIER COEFFICIENTS RELATED TO HIP FLEX-EXTENSION

The main advantage of the Fourier-based kinematic model is that leg joint trajectories referring to two consecutive steps are not characterized by discontinuity because they are estimated as the sum of sinusoidal functions whose periods are integer fractions of stride period (i.e., Formula$T$). Fig. 4 shows a representative set of leg-joint angular excursions using the reported model, for five values of walking speed.

Figure 4
Fig. 4. Leg joint angular excursions estimated by the proposed model at different speeds.

Noticeably, regarding cycle duration, NEUROBike also allows the user to select a custom value of Formula$T$ which does not depend on the walking speed. This allows training to be characterized by wide angular excursions (comparable to excursions at fast walking speed), and provided slowly enough to avoid eliciting clonic movements.

Figure 5
Fig. 5. On the left, angular excursions at hip, knee, and ankle joints of subjects for the sit-to-stand motion are represented in gray. The averaged curves are reported in black. On the right, angular excursions at knee and ankle as a function of the hip joint excursion are represented in gray. Averaged curves and polynomial fitting models are respectively represented in black and red. Fitting models are cubic polynomials expressed by the following equations:–Formula${\hbox {knee}} = 0.00012 \times {\hbox {hip}}^{3} + 0.0096 \times {\hbox {hip}}^{2} + 1.1 \times {\hbox {hip}} + 5.8$;–Formula${\hbox {anke}} = -0.000091 \!\times\! {\hbox {hip}}^{3} \!+\! 0.00067 \!\times\! {\hbox {hip}}^{2} \!+\! 0.18 \!\times\! {\hbox {hip}} - 0.6$.

D. Kinematic Model of Leg Joint Angular Excursions During Sit-to-Stand

According to literature, leg joint angular excursions during sit-to-stand are characterized by a certain degree of correlation [21] such that it is possible to express two of them as depending on the third one, thus reducing the computational load of implemented algorithms. Therefore, a group of seven healthy subjects (five males and two females; age Formula$26\pm 1.0 {\hbox {years}}$) was asked to carry out this motor task while the kinematics of the subset of markers described in the previous section, related to legs and pelvis, was captured (six-camera based Vicon 512 Motion Analysis System, sample rate of 100 Hz). Recording during both sitting and standing phases started with an acoustic signal and stopped 0.3 s after onset. Since sitting and standing are characterized by basically mirrored leg joint kinematics [21], only data referring to sitting were considered for further processing consisting of: 1) low-pass filtering (zero lag Butterworth filter with cutoff at 10 Hz); 2) reporting angular excursions at the knee and ankle as a function of the hip excursions. Finally, cubic polynomial fitting models were computed and adopted as base functions for the control algorithm. Fig. 5 shows both the experimental data and polynomial models.

E. Ellipse-Based Trajectories

A third paradigm of exercises developed for NEUROBike was based on ellipsoidal trajectories which are implemented both in passive and active mode (Fig. 6). The length of major and minor axis, the orientation, and the duration of a cycle can be modified by the user. This exercise was developed to allow patients affected by hypertone in specific muscle groups to carry out the treatment. In particular, due to the features of the ellipse, NEUROBike was expected to differently emphasize the range-of-motion (ROM) at the leg joints.

Figure 6
Fig. 6. The figure shows the representative set of ellipses, in terms of axis dimension and orientation, used during performance evaluation. Values may be customized according to therapy's needs.

To the best of our knowledge, this paradigm has never been implemented in other robotic-based platforms aimed at providing neurorehabilitation of walking. Although the leg joint kinematics directly achieved is not comparable to that obtained with daily motor tasks, these rehabilitative training exercises are expected to accommodate the abnormal muscle hypertone or co-contraction following a neuropathy, which often occurs in specific muscle groups.

SECTION IV

PERFORMANCE EVALUATION

This section describes a set of preliminary experimental tests carried out to evaluate the influence of NEUROBike-mediated robotic treatments, on the kinematics and muscle activation of a group of healthy subjects.

A. Subjects and Procedures

Synchronized electromyographic (EMG) and kinematic data were recorded while seven healthy adults (five males and two females, age Formula$26\pm 1.0 {\hbox {years}}$, body weight Formula$61.2\pm 11.7 {\rm kg}$, height Formula$1.70\pm 0.08 {\rm m}$) underwent exercises provided by NEUROBike, after giving their informed consent.

The 3-D trajectory of a set of 11 spherical markers placed on body landmarks of the pelvis and left lower limb (left and right anterior superior iliac spines, prominence of the greater trochanter external surface, lateral epicondyle of the femur, head of fibula, lateral malleolus, first and fifth metatarsal head, tip of the toe and additional markers rigidly attached to wands over the mid-femur and mid-shaft of the tibia), and three markers located on the left NEUROBike footplate, was recorded by using a six-camera based Vicon 512 Motion Analysis System (Oxford, U.K.) with a sample rate of 100 Hz.

EMG signals of seven muscles belonging to the right leg were recorded (NORAXON, Telemyo 2400T, V2) by using bipolar Ag-AgCl surface EMG electrodes placed on previously shaved and cleaned skin. Recorded muscles were: Gastrocnemius Lateralis (GL), Soleus (SOL), Tibialis Anterior (TA), Rectus Femoris (RF), Vastus Medialis (VM), Biceps Femoris (BF), and Tensor Fascia Latae (TFL). Sampling rate was 1500 Hz and the gain of the amplifiers was 1000. Electrode placement was tested through suitable movements [24] in order to verify the absence of cross-talk among muscle signals. The experimental set-up is reported in Fig. 7.

Figure 7
Fig. 7. Experimental setup adopted while carrying out performance evaluation.

B. Motor Tasks

The kinematics of lower limb and EMG activity was recorded while subjects carried out both passive and active exercises mediated by NEUROBike. The connection between feet and footplates was fastened by straps. During the passive motor tasks, the subjects were fully assisted while NEUROBike either imposed angular excursions at the leg joint, comparable to walking excursions (according to the model described in Section III-A), or led the end-effector along ellipsoidal trajectories. According to the configuration of the platform (see Section III-A), angular excursions during passive walking were related to three different speeds (i.e., slow, comfortable, and fast respectively corresponding to 0.1, 1.1, and 1.6 m/s) and the duration of the cycle was always fixed at 5.5 s. Concerning the ellipsoidal trajectories, the end effectors tracked the three different ellipses reported in Fig. 6 in clockwise and counterclockwise directions, with a period of 4 s.

Both sit-to-stand (see model described in Section III-B) and ellipse-based exercises were conducted in active mode, that is, the isokinetic movement of the end-effector through the extension phase of its trajectory was triggered by a force of the foot on the foot plate was greater than a certain threshold (for performance evaluation 8 kgf was set). During both ellipse and sit-to-stand exercises the movement of the legs was alternated (i.e., a leg was actively extended while the contralateral one was passively flexed).

Records related to passive exercises lasted about 60 s while those related to the active exercises ended after 10 cycles.

Figure 8
Fig. 8. Leg-joint angular excursions measured during passive walking-like exercises. Each subplot shows the averaged curve (black line), one standard deviation wide error band, Formula${\bf rms}$, and mean Formula${\mmb \rho}$.
SECTION V

RESULTS

A. Kinematics During Passive Exercises

During passive walking, leg joint angular excursions appeared comparable to those estimated by the model above all at proximal joints (see Figs. 4 and 8). More in detail.

  • – Hip angular excursions were similar to those expected (mean Pearson correlation coefficient, Formula${\mmb \rho} > {0.80}$) with low inter-subjects variability at all speeds (root mean squared error, Formula${\bf rms}< {5.27}^{\circ}$); as the speed increased, the ROM at the hip increased from 25° to 45°, the inter-subject variability (Formula${\bf rms}$) decreased, and the similarity between measured and expected angular excursions (Formula${\mmb \rho}$) increased.

  • – Measured knee angular excursions were similar to those expected (Formula${\mmb \rho} > {0.95}$) and the inter-subjects variability (Formula${\bf rms} < {9.79}^{\circ}$) was higher than that at the hip joint; as the speed increased, maximum flexion increased from 60° to 70°, the Formula${\bf rms}$ decreased, and the similarity between expected and estimated angular excursions (Formula${\mmb \rho}$) remained almost constant; noticeably, the knee appeared flexed by an offset of about 10°–20° that did not allow flex-extension as during the loading response (0%–10% of the cycle).

  • – Ankle angular excursions were dissimilar to those expected (Formula$0.35< {\mmb \rho} < {0.62}$) and showed medium inter-subjects variability at all speeds (Formula${\bf rms}< {7.43}{^\circ}$); during 0%–40% of the cycle, the ankle was more plantarflexed than during natural walking and dorsiflexion occurred during 40%–60% of the cycle; at the end of the cycle (80%–90%), the ankle remained plantarflexed and reached the starting value late.

During ellipse-based passive exercises, leg joint trajectories at hip, knee, and ankle were consistent across cycles and were characterized by an Formula${\bf rms}$ between about 5° and 15° (Fig. 9). The knee joint generally showed greater Formula${\bf rms}$ than the others, which increased when the ellipse was vertically orientated. Noticeably, leg-joint angular excursions during counterclockwise rotations were more symmetric with respect to the cycle than those obtained performing clockwise ellipsoids.

Figure 9
Fig. 9. Leg-joint angular excursions measured during passive ellipse-based exercises. Each subplot shows the averaged curve (black line), 1 standard deviation wide error band, Formula${\bf rms}$, and mean Formula${\mmb \rho}$. Columns labeled: A and B refer to the ellipse oriented at Formula$- \pi /6 {\rm rad}$, respectively, clockwise and counterclockwise directions; C and D refer to the vertically oriented ellipse, respectively clockwise and counterclockwise directions; E and F refer to the ellipse oriented at Formula$\pi /6 {\rm rad}$, respectively, clockwise and counterclockwise directions.

B. Kinematics and EMG Activity During Active Exercises

Leg extension during sit-to-stand (Fig. 10) was characterized by angular excursions at the hip, from 60°–5°, at the knee, from 75° to 15°, and at the ankle, from Formula$-15$° to Formula$-5 ^{\circ}$, comparable to those expected (Fig. 5). Kinematics during this exercise was characterized by an Formula${\bf rms}$ slightly greater than that related to the passive exercises.

Figure 10
Fig. 10. The figure shows leg-joint angular excursions during both flex and extension cycles related to the sit-to-stand based exercise.

Regarding muscle activation, it was possible to distinguish between the different contribution of recruited muscles during flexion and extension cycles. In particular, the extension scheme was mainly carried out by the simultaneous activation of RF, VM, and BF, respectively, leading the extension of knee and hip joints. During the flexion cycle, the TFL and a lower RF contribution helped the flexion of the hip while the TA controlled the dorsiflexion of the ankle. Calf muscles (SOL and GL) were slightly activated only during the flexion cycle (Fig. 11).

Figure 11
Fig. 11. The figure shows the superimposition of EMG signals recorded during both the sit-to-stand (green) and the ellipse (gray) based exercises. In this last regard, on the subplot A) we reported EMG signals related to the vertically-oriented ellipse, on the subplot B) we reported EMG signals related to the ellipse inclined by Formula$\pi /6$. Noticeably, EMG signals are normalized with respect to their maximum values and time-interpolated over 1000 points to make flexion and extension phases comparable across exercises. Vertical lines represent the onset of the extension and flexion phases (see labels E and F on the horizontal axis).

Regarding active ellipse-based exercises, we observed that kinematics and inter-subjects variability (e.g., Formula${\bf rms}$) were comparable to the respective passive exercises. Conversely, muscle activation was characterized by a different modulation than that observed during the sit-to-stand exercise. In particular, due to the orientation of the ellipse and the direction of the movement (i.e., clockwise or counterclockwise), both the timing and the amplitude of the coupled activity of RF and VM, and/or BF and TFL, could be modulated. Moreover, a significant contribution of the calf muscles was observed during the flexion cycle. A representative example is reported in Fig. 11.

SECTION VI

DISCUSSION

NEUROBike is a robotic platform designed and developed to provide neurorehabilitation for the recovery of walking abilities in bedridden post-stroke patients. This approach is based on the hypothesis that early and task-oriented treatment can enhance the outcome of therapy and favor cardiorespiratory training. By separately controlling each DoF, several exercises can be planned that involve different leg-joint angular excursions into the sagittal plane.

At the moment NEUROBike is able to provide exercises in both passive and active modes. Specifically, it leads the patients' lower limbs and induces leg-joint angular excursions similar to natural walking (Fig. 8) or sit-to-stand (Figs. 5 and 10), according to subjects' anthropometrical features. Moreover, it can drive the end-effector while tracking differently oriented ellipses into the sagittal plane, in order to modulate the ROM at the leg joints. Furthermore, sit-to-stand and ellipse-based exercises are implemented actively, that is, subjects are not assisted by the robot but they have to start, perform, and complete the task by themselves, while delivering a specified force.

Comparison of leg-joint angular excursions obtained during the first preliminary tests with those reported using exoskeleton-based robotic platforms [25], shows that the proposed approach involves an unexpected flexion offset at the knee (about 10°–20°, Fig. 8) combined with a slight plantarflexion (about 10°–15°), which was also reflected in a slight flexion of the hip joint. We specifically observed that subjects asked to practice passive walking used a further DoF at the metatarsophalangeal joint which redistributed lower limb movement among four DoFs, and, consequently, affected kinematics at the leg joints. Although this could be potentially due to the stretch reflex of the calf muscles, this hypothesis was rejected due to the fact that the EMG activity of SOL and GL was negligible during the passive exercises and, consequently, uncorrelated with the movement. Therefore, it was possible to conclude that due to both ergonomic factors and the absence of a rigid link between feet and footplates, the healthy subjects adopted postural adaptations of the leg that presumably helped them to manage the exercise in a more versatile manner. Further development of the platform will consider solutions for reducing the movement of the foot with respect to the footplate. A suitable seat will also be designed to provide more comfortable excursion of the leg joint while subjects undergo treatments.

To facilitate the training of patients affected by hypertone, or prone to generating clonic movements, ellipse-based exercises appeared as useful alternative treatments. In particular, by modifying the orientation of the ellipses, different combinations of angular excursions may be obtained at each joint while the legs continue to have cyclic and alternate movements as when walking. This possibility allows the therapist to select the best axis inclination for each patient, by either emphasizing or attenuating the ROM at each articular joint according to the patient's clinical picture. Further possibilities, such as modifying the axis length or the center of the ellipses (here not investigated), can favor the development of customized exercises strongly focused on the patient's aptitude to carry out the rehabilitative treatment.

Concerning active exercises, the preliminary experimental tests highlighted that both sit-to-stand and ellipse-based exercises involve the recruitment of suitably coupled muscle groups in healthy individuals. The main difference between the two sets of exercises is that ellipse-based exercises allow a different modulation of the EMG activity, mainly in terms of timing. This result appeared relevant for the performance of neuro-rehabilitative treatments in post-stroke patients. It is well known that muscle activation in post-stroke patients appears to be characterized by more rigid coordination compared to healthy subjects, such that their neural control signals reflect a reduced flexibility of control schemes [26]. For this reason, previous authors have adopted strategies (e.g., split-belt treadmills [27]) aimed at readapting inter-limb and intra-limb coordination after traumas. We believe that a customized treatment mediated by NEUROBike may help patients to gain flexibility in lower limb control leading to significant improvements in motor performance during walking. Finally, the low inter-subjects variability in the joint trajectories observed in all exercises provided by NEUROBike, highlights the system's ability to provide highly repeatable exercises, a feature that makes it a suitable tool for clinical and research applications.

From a clinical viewpoint, the platform will be involved in a set of pilot studies aimed at verifying its compliance with the needs of patients affected by different pathologies with increasing degree of severity. Then, it will be adopted as an additional tool, in the training of post-stroke patients.

SECTION VII

CONCLUSION

The paper introduces a new robotic platform aimed at providing neuro-rehabilitative treatments to bedridden post-stroke patients. In particular, mechanical structure, control architecture, kinematic models implemented in the control algorithm (i.e., walking, sit-to-stand, and ellipse-based), and passive and active control strategies leading the robot have been described. A set of pilot tests was carried out to verify whether the platform could achieve expected performance. Results showed that NEUROBike is a suitable tool for training of post-stroke patients provided early after the trauma.

ACKNOWLEDGMENT

The authors would like to thank Dr. L. Bassi Luciani for his valuable support. This work was supported by the Fondazione CaRiPisa within the framework of the project “NEUROBike: progettazione e realizzazione di un sistema biomeccatronico per la riabilitazione dell'arto inferiore in soggetti emiparetici in fase acuta.”

Footnotes

This work was supported by the Fondazione CaRiPisa within the framework of the project “NEUROBike: progettazione e realizzazione di un sistema biomeccatronico per la riabilitazione dell'arto inferiore in soggetti emiparetici in fase acuta.”.

V. Monaco, M. Coscia, and D. Martelli are with The BioRobotic Institute, Scuola Superiore Sant'Anna, 56127 Pisa, Italy (e-mail: m.coscia@sssup.it; v.monaco@sssup.it; d.martelli@sssup.it).

G. Galardi is with the Fondazione Istituto San Raffaele, 90015 Cefalu', Italy (e-mail: giuseppe.galardi@hsrgiglio.it).

S. Micera is with The BioRobotic Institute, Scuola Superiore Sant'Anna, 56127 Pisa, Italy, and also with Translational Neural Engineering Lab, Center for Neuroprosthetics, Swiss Federal Institute of Technology Lausanne (EPFL), CH-1015 Lausanne, Switzerland (e-mail: s.micera@sssup.it).

Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org.

References

No Data Available

Authors

No Photo Available

Vito Monaco

No Bio Available
No Photo Available

Giuseppe Galardi

No Bio Available
No Photo Available

Martina Coscia

No Bio Available
No Photo Available

Dario Martelli

No Bio Available
No Photo Available

Silvestro Micera

No Bio Available

Cited By

No Data Available

Keywords

Corrections

None

Multimedia

No Data Available
This paper appears in:
No Data Available
Issue Date:
No Data Available
On page(s):
No Data Available
ISSN:
None
INSPEC Accession Number:
None
Digital Object Identifier:
None
Date of Current Version:
No Data Available
Date of Original Publication:
No Data Available

Text Size