Characterization and Integration of Muscle Signals for the Control of an Exoskeleton of the Lower Limbs during Locomotor Activities

Abstract

Daily activities are a source of fatigue and stress for people with lower extremity spasticity. The possible aids must be introduced while maintaining priority control by the patient. This work aims to develop such an application in the context of walking on the exoskeleton developed at the Systems Engineering Laboratory of Versailles (LISV). The application results are based on data recorded at the END-ICAP laboratory with gait sensors for healthy subjects, people with CPs, and people who had a stroke. Our contribution is the proposal of a new method of neuromotor control for a rehabilitative exoskeleton. It consists in determining and assisting the motor instructions for the movements of a patient while retaining his expertise; the assistance as needed and the detection of its intention based on a fusion of information. The results show that the proposed index characterizes the relationship of the angle difference with a reference movement for each joint. It dynamically compensates for movements efficiently and safely. This index is applicable for gait pathology studies and robotic gait assistance.

1. Introduction

Our body movements are the result of a complex interaction between the central nervous system (CNS), nerves and muscles. Damage to any of these components can lead to movement disorders [1]. In the world, these lesions represent the leading cause of disability. A first example is that of cerebrovascular accident (stroke) (CVA). There are more than 700,000 new cases of stroke each year - one every four minutes. Stroke is the leading cause of acquired physical disability in adults, in cases where the patient survives. It can occur at any age, 25% of patients are under 65 and 10% under 45. The number of cases in young people has increased significantly. A second example is cerebral palsy CP which affects a newborn every 6 hours or one in 800 children. What was previously called cerebral palsy is the leading cause of motor disability in children. These neurological conditions produce disorders that lead to slowed or absent voluntary movement caused by muscle spasticity. They affect the speed, quality, and ease of day-to-day activities of the human body [2], spasticity being an increase in muscle contractions that causes stiffness or contraction of muscles which impedes movement. It causes difficulty in walking, locomotion, or maintaining normal posture and balance [3].

As part of this work, we were interested in the design of rehabilitation systems for stroke and CP patients. The system considered is that of the rehabilitation exoskeleton, as it exists in clinics. The exoskeleton is a mechatronic device, worn by the patient, designed to increase physical performance, which is adapted to the shape of the human body and the function targeted for rehabilitation. This orthosis is used to provide high-intensity training to human limbs, tailored for each user, to promote recovery from a disease or neurological disorder [4]. The exoskeleton works mechanically in parallel with the human body [5] and can be controlled in a way passive or active. As a general rule, an exoskeleton considered as a whole is a system that has means of perception to acquire physiological signals of interest, of means of calculation to apply a treatment to extract the relevant parameters, and of means of action to give control instructions to mechanical effectors. In the case of walking disorders, the autonomic exoskeletons are frequently used for rehabilitation. This type of exoskeleton represents great challenges in terms of control: each patient has their own motor skills which make generic control of an exoskeleton difficult [6]. It is, therefore, necessary to find solutions that allow better control of the functions of these exoskeletons which place the patient in an expert position in his movement. This approach helps to promote rehabilitation to compensate for difficulties until almost normal movement is achieved. The patient retains the possibility of direct control by an interactive control method, between the exoskeleton and the patient, which takes into account the components of normal walking to establish a diagnosis and perform the analyzes necessary to understand walking pathological. The use of biomechanical parameters makes it possible to meet the medical needs concerning the gait of the patients to be treated [7]. First, the temporal-space parameters characterize walking overall, by integrating notions such as cadence, speed, or the number of steps [8]. Second, the kinematic parameters of gait are obtained by analyzing joint angles during a cycle by direct visual observation or by 3D video analysis for greater precision [9]. Finally, electromyographic (EMG) signals vary with time and can be characterized by their amplitude, frequency and phase [10].

The main challenge facing interactive exoskeletons is the direct relationship between biomechanical signals and the desired behavior of the exoskeleton. One of the most difficult cases addressed by robotics research is walking rehabilitation with an autonomous exoskeleton. The best-known autonomous systems are HAL [11], Wandercraft [12] or Ekso GT [13]. There are two known control strategies for these devices: (i) impedance or admittance control, generally predetermined and not taking into account the user’s physical condition [14, 15], (ii) a control using electromyography by the detection of muscle activation, but this type of control is not satisfactory because the EMG signals recorded on patients with muscle disorders lead to incorrect operation [14, 15, 16]. Thus, these two strategies can only be used with a predefined behavior that the patient should follow. They are not suitable for patients with CP or after a stroke. Thus, these two target populations of our study require a new approach, based on continuous measurements of biomechanical signals.

The signals acquired must be analyzed, interpreted, and used to drive the trajectory of an online exoskeleton. Off-line approaches limit the patient’s movement and do not take into account the variability of gait [9]. These cause fatigue and pain in the patient during a rehabilitation session with the exoskeleton [17]. Moreover, it does not allow to distinguish the deviations from the compensations due to the influence of the walking speed [9].

The approach we had followed is based on recording EMG signals from muscles involved in the knee and hip movements [18].

Like o novelty from the other research, we introduced a new muscle co-contraction index for a control strategy capable of assessing a compensatory joint angle suitable for CP or stroke patients. EMG signals contain information about the patient’s intention [19] and can be used to assess muscle co-activation around the joint [20]. It has been suggested that muscle co-activation indicates the achievement of motor skills [21, 22] and it is also linked to joint stability [22]. Co-activation is considered to be an important factor contributing to the ineffectiveness of pathological movements [23]. Most researchers and clinicians rely on EMG measurements to express it as a comparison between the EMG measured for the muscles involved and the reference values. These values are also often examined using the co-contraction index [24] for a given joint. Although, the assessment of co-activation is suitable for off-line analysis and diagnosis, it does not meet the constraints of a control strategy because it is calculated at each phase of the walk cycle and in a specific way.

2. Theoretical framework

Locomotion is the ability to move. More specifically, this term refers to the way we travel from one place to another. Running, swimming, jumping, or flying are examples of types of locomotion. Human locomotion begins with signals from the central nervous system (CNS), which are transmitted to muscles to move joints. The CNS is made up of many nerve cells that are interconnected. Voluntary movement results from sending a nerve impulse from the brain that travels through the spinal cord and then the motor nerves. The brain plays a major role in the balance and coordination that are essential for our locomotion. The limbs are the main elements that drive movement in humans. The structures used during locomotion include the bones that support the body and help maintain the shape of the body and provide a surface for muscle attachment. Then there are the joints which are the points of contact between the bones and which allow the bones to move against each other without friction. Tendons connect muscle to bone and transfer the force generated by muscle contraction into the movement of the skeleton. Finally, the muscles that work in agonist/antagonist pairs move the bones of a joint by varying the angles of the joint. Locomotion is a behavior capable of providing information on motor control strategies [25, 26, 27]. Locomotor activity is described using specific anatomical terms that are used to determine the processes of movement. The terminology used describes this movement as a function of its direction with respect to the anatomical position of the joints as shown in Figure 1a. In general, the movement is named according to the anatomical plane in which it occurs (Figure 1b), where two axes of a joint are near or far from each other (Figure 1c). Anatomists use a unified set of terms to describe most movements, although other more specialized terms are needed to describe the uniqueness of movements such as those of the hands or feet [28].

Figure 1.

Anatomical characterization of a movement: a) anatomical position; b) anatomical axes; c) anatomical planes.

A subject is considered to be in the anatomical position, shown in Figure 1a. When standing in an upright posture, face straight in front of him, feet close together and parallel, and palms of hands facing [29].

This is a standard position for which planes and axes are defined. To perform a practical analysis of human movement, anatomical movements can be defined as the act of moving body structures or changing the position of one or more joints of the body. Joint actions are described to the anatomical position which is the universal starting position for describing movement [30].

The movement of the limbs is depicted with the aid of reference planes, which can be seen in Figure 1c. From the reference anatomical position, three imaginary section planes are defined, crossing a subject’s center of gravity and perpendicular to each other. These three planes are the sagittal plane, the frontal (or coronal) plane, and the transverse plane. The sagittal plane is a vertical plane perpendicular to the ground, which passes through the middle of the body and divides it into two symmetrical parts, lateral and medial or right and left. The frontal (or coronal) plane is a vertical plane perpendicular to the ground and to the sagittal plane, which separates the two anterior (front or ventral) and posterior (rear or dorsal) parts. The transverse plane, also called horizontal, axial, or median, is a horizontal plane parallel to the ground, separating the upper and lower parts of the body [31].

At the intersection of two planes constituting an axis, three anatomical axes are defined: the sagittal axis (anteroposterior), the transverse axis, and the vertical axis (longitudinal). The vertical axis is a longitudinal axis perpendicular to the ground when the subject is standing, the axis transverse is perpendicular to the vertical axis and the sagittal axis is an axis crossing the body from front to back and perpendicular to the previous two [31].

This work is particularly interested in pathological gait, and for this, it is necessary to understand normal gait and the terminology used to describe it. It is the benchmark against which a patient’s gait can be compared. Even if the patient’s gait differs from normal in one way or another, this does not necessarily imply a clinical or social need to transform it into a “normal” gait. Many abnormalities or walking disorders are compensation for certain problems encountered by the patient. Therefore, although abnormal they are useful.

A definition of walking is proposed by Smidt [32] as “The way of moving the body from one place to another by alternately and repetitively changing the position of the feet”. In more detail, it is a translational progression of the body produced by coordinated rotational movements of the body segments. This essential activity, cyclical in nature, results from a series of rhythmic movements. It is characterized by alternating propulsive and restraining movements of the lower limbs which cause the center of gravity and the body to move forward by putting one foot in front of the other in a repetitive manner [33, 34]. Walking allows you to move around autonomously and independently. However, it is a complex activity, learned after a long process of trial and error before it matures. Once acquired, walking becomes an “automatic” activity that no longer requires special attention and is broken down into a series of movements that are repeated according to a specific cycle. The principles of gait are described in the literature of different disciplines: for example, from a medical point of view where the study of gait is carried out for surgery and prostheses [35] or else from the point of view of robotics, which aims at the design of artificial systems that move while walking [36].

Walking is accomplished using a complex, coordinated set of nerve signals, sent to muscles, to move joints, limbs, and the rest of the body. The central nervous system that produces these nerve impulse patterns is made up of neurons located in various parts of the brain and spinal cord.

Human locomotion relies on a rhythm-generating system located in the spinal cord and controlled by cortical neurons. This system reacts according to the information transmitted by the sensors of the muscles, joints, and skin of the legs [37]. At the cortical level, walking involves many structures such as the visual, cerebellar, vestibular, and proprioceptive cortices. In addition, the locomotor areas, trunk, and gray nuclei play a major role in controlling the activity of the gait spinal generator, which in turn controls muscle activity. The basal ganglia participate in the initiation of walking and its proper progress. The cortex plans the action and the choice of a motor program when the cerebellum is dedicated to controlling gait and balance, as shown in Figure 1 [38].

The arthrology of walking is linked to the lower limbs, it is dependent on three main joints: hip, knee, and ankle. These joints are mainly stressed in one or two of the three planes defined in Figure 1. The directions of these movements for the hip, knee, and ankle are shown in Figure 2.

Figure 2.

Joints and main movements of the lower limbs.

The possible movements of these joints, which take place in the sagittal plane, are flexion and extension, these same movements are also called dorsiflexion and plantarflexion for the ankle. Abduction and adduction movements take place in the frontal plane and are visible in Figure 2. Internal and external rotations (medial and lateral) take place in the transverse plane [31]. The muscles ensure the movement of the limbs by limiting the articular angles by their co-contraction. From a biomechanical point of view, muscles are distinguished by their role in mono or biarticulate movement, that is, by dividing the movements into anatomical units according to each joint. This distribution takes into account, due to the complexity of the musculoskeletal system, the fact that muscular action is not limited to a single degree of freedom mobilized voluntarily. According to Rasch et Burke [39] we distinguish: agonist muscles which are muscles whose contraction tends to cause the desired movement and antagonist muscles which are muscles whose contraction serves to produce an exactly opposite joint action. To fully understand normal walking, it is necessary to know which muscles are active as agonists and antagonists during different parts of the gait cycle to achieve the desired joint angles [13].

For example, the action of the gluteus mediums, illustrated in Figure 3a, depends on a fixed point. If the fixed point is on the pelvis, the gluteus mediums abduct the thigh, with maximum effectiveness when the angle between the pelvis and the thigh is between 30° and 35°. The anterior fibers are brought into play as he rotates the thigh medially on the pelvis while the contractions of the posterior fibers cause the thigh to rotate sideways on the pelvis.

Figure 3.

The muscles: (a) gluteus medius, (b) hamstrings, (c) quadriceps, (d) tibialis anterior and (e) triceps sural.

If the fixed point is on the femur, the gluteus medius muscle acts as a stabilizer of the pelvis. In addition, the gluteus medius performs a homolateral tilt of the pelvis. The hamstrings, visible in Figure 3b, are divided into semi-tendinous, semi-membranous, and biceps femoris. It causes knee flexion, thigh extension, pelvic retroversion. It participates in the stabilization of the knee and the pelvic girdle. In addition, they participate in the adduction of the thigh to the pelvis. During knee flexion, the biceps femoris rotates the knee sideways while the semi-tendon muscle rotates medially.

The quadriceps, visible in Figure 3c, is divided into 4 heads: the rectus femoris, the vastus medial, lateral, intermediate. Stabilization of the knee is affected by the vastus lateralis and medialis. The rectus femoris and vastus intermedius play a dynamic role, the rectus femoris muscle being a biarticular muscle, flexes the thigh on the pelvis. The tibialis anterior muscle is a flexor muscle where it exerts dorsiflexion of the ankle, this muscle is shown in Figure 3d. In addition, it causes an inversion of the ankle. Along with the contraction of other muscles, the tibialis anterior participates in triple flexion.

The sural triceps is divided into gastrocnemius medial, lateral gastrocnemius, and soleus Figure 3e. He exerts plantar flexion of the ankle which allows him to lift the whole weight of the body on the tip of the foot, as can be seen in Figure 2. There is little stress on the gastrocnemius during open chain knee flexion. The triceps sural participates with other muscles in knee flexion, but to a limited extent [Lacôte et al., 2014], its action is essential during the support phase for dorsiflexion control.

The role of muscles has been studied through electromyography (EMG), first by Scherb in 1940 who began his experiments by touching the muscles when walking a subject on a treadmill and then using the EMG. In 1981, the reference base for the study of muscle activity was made available by the group of “Verne Inman” at the University of California at San Francisco and at Berkeley. This group has contributed to significant advances in the understanding of muscle activity and many aspects of normal walking [40]. The use of surface EMG in gait analysis has since received much attention. We will detail two muscle groups that are particularly important for this work because they involve in the movements of the hip and the knee, which are the joints studied in this work. These are the quadriceps and the hamstrings. These two groups of muscles are biarticular and act on both the knee and the hip. At the front of the thigh, the quadriceps serves to straighten the leg and flex the thigh over the pelvis. At the rear, the hamstrings provide flexion and rotation of the knee as well as the extension of the thigh over the pelvis. The work is mainly carried out in the sagittal plane.

To take on the role of these muscle groups, it is important to study their action after neuromuscular activation. The kinematics is characterized in this plane by flexion of the hip and the knee. It is associated with median and posterior neuromuscular activations of the hamstring during the initial contact phase of the leg with the ground. The kinematics, in this plane, are also associated with an increased lateral neuromuscular activation of the quadriceps, during the maximum load phase of the movement [41]. We can thus see that:

• the rectus femoris (rectus femorus), which is part of the quadriceps, flexes the hip and allows the knee to extend;

• the biceps femoris, which is part of the hamstring, plays an opposite role to that of the rectus femoris and flexes the knee to allow hip extension;

• semimembranous and semitendinous, extend the hip and flex the knee.

The roles of these muscle groups are detailed in Table 1 for each type of movement and each joint.

	Movement	Agonist	Antagonist
Knee	Flexion	Quadriceps	Hamstring
	Extension	Hamstring	Quadriceps
Hip	Flexion	Hamstring	Quadriceps
	Extension	Quadriceps	Hamstring

Table 1.

Agonist and antagonist bi-articular muscles, during hip and knee flexion and extension movements.

3. Biomechanics of walking

Biomechanics is a discipline that proposes to consider biological systems as objects of applied mechanics. His research methods are a combination of mechanics, anatomy, and physiology. It is a science of movement through the study and reproduction of the mechanisms which result in a definite movement of the body. Joint mechanics focus in particular on the joints, which provide both movements of a limb segment and its stability.

From the point of view of biomechanics, walking speed is decisive in evaluating the contribution of each segment of the body [42]. Walking speed primarily affects the stability and balance of the lower limbs. Joints can produce greater ranges of motion through greater muscle responses. In the bipedal system, the three main joints in the lower body and pelvis work in sync as muscles and impulses move the body forward. The degree to which the body’s center of gravity shifts during translation defines the efficiency of the movement. The body’s center of mass shifts sideways and up and down while walking. Walking is a repeating pattern comprising steps, with a stride denoting a complete cycle of walking. The characteristic frequency is defined by the split time, which measures the time elapsed between the heel touch of one leg and the heel touchdown of the contralateral leg. The width of a step can be described as the mediolateral space between the two feet [43].

Analysis of the gait cycle is important in examining biomechanical mobility to gain insight into lower limb dysfunction during dynamic movements [44]. By gait cycle analysis, it is best to examine each joint separately [42]. Both objective and subjective methods can be used [45, 46]. The cyclical nature of walking gives it a certain uniformity, but sometimes acyclic elements result from accidental causes such as tripping and other more natural causes such as spinning are severe. Since walking allows movement from one point to another but requires maintaining balance, dynamics play an important role Figure 4.

Figure 4.

The gait cycle and its phases [47].

The automation of this activity facilitated its study, mainly by allowing biomechanical standards to be measured and established. The establishment of standards is essential for the study of pathological movement because they allow understanding by comparison. Walking is described in the state of the art as a cyclic function. A cycle is determined by all the events occurring between two successive identical events. The gait cycle is the time interval between two successive occurrences of one of the repetitive locomotion events [48]. The onset of the gait cycle is most often represented by the initial contact of a foot with the gait surface [13, 32, 47, 49, 50]. This cycle includes the support and oscillation phases. The oscillation phase refers to the moment when the foot is in the air for the progression of the limb. The support phase is the period when the foot is in contact with the ground as shown in Figure 4. Walking can also be defined as including initial double support, single-limb support, double end support, and a swing. The support phase represents 60% of the walk cycle, where each interval of double support is 10% and single limb support is 40% [50]. The oscillation phase designates the remaining 40%. Support from one limb is equivalent to tilting the other because they occur at the same time [47].

According to Perry and Davids [47], the walking cycle can be divided into eight phases, which correspond to three basic tasks as shown in Figure 5. (i) the first phase, from 0 to 2% of the cycle, begins with the initial contact or with the contact of the contralateral foot with the ground; (ii) the second phase, from 2 to 10% of the cycle, is the initial double push phase; (iii) the third phase, from 10 to 30% of the cycle, consists for the first half of the phase of simple support going from the detachment of the contralateral foot until the passing of the center of mass of the vertical; (iv) the fourth phase, from 30 to 50%, is the second half of the single support phase ending when the contralateral foot touches the ground; (v) the fifth phase, 50–60%, is the final double-support phase in which the weight is transferred from one limb to the other. The support phase ends with the toes peeling off; (vi) the sixth phase, from 60 to 73%, is the start of the oscillating phase and begins when the foot leaves the ground and ends when the foot is aligned with the contralateral foot; (vii) the seventh phase, 73–87%, is the mid oscillating phase, this phase ends when the oscillating member is forward; (viii) the eighth phase extends from 87 to 100% of the walk cycle, this is the end of the oscillating phase. This phase ends when the foot comes into contact with the ground. Advancement of the limb is complete when the leg segment is located in front of the thigh segment, all the steps are illustrated in Figure 5.

Figure 5.

Decomposition of the walking cycle.

During walking, significant movements occur in the three sagittal, frontal and transverse planes. However, the study of gait movement for exoskeletal rehabilitation focuses on the sagittal plane. The kinematics of gait shows the quantification of joint angles in flexion and extension, Figure 6. The hip flexes and stretches once during the cycle. The flexion limit is reached around the middle of the first phase, then the hip is kept flexed until the first contact. Maximum extension is reached before the end of the support phase, and then the hip begins to flex again. The knee has two flexion peaks and two extensions during each walk cycle. It is nearly extended before initial contact, followed by flexion, narrowing towards extension during the downforce phase, before resuming flexing to peak in the swing phase. Finally, it stretches out again to prepare for the initial contact of the next step. The ankle is usually a few degrees from the neutral position on initial contact. After the initial contact, the plantarflexion control of the ankle brings the forefoot back to the ground. Halfway through, resumes dorsiflexion. Before the initial opposite contact, the angle of the ankle changes again, major plantarflexion occurs just after the toes have lifted off. During the swing phase, the ankle returns to dorsiflexion until the forefoot has cleared the ground, then the ankle returns to a neutral position which is maintained until initial contact, of the next step.

Figure 6.

Illustrations in the sagittal plane of the articular angles of the hip, knee and ankle, during a gait cycle performed by healthy adults at the spontaneous speed [adapted from (Lacôte et al., 2014)].

Muscle co-contraction (CCM) has a functional role in ensuring the stiffening of the joint to limit imbalance during limb advancement [51]. This role is particularly visible in children, in whom the establishment of this mechanism during the first steps has been well described [52, 53]. Gait patterns in people with a neurologic impairment are characterized by abnormal total voluntary contraction (TVC). This is reflected in particular by problems in postural stability [54], since it is important for providing joint stability [54, 55, 56, 57, 58], adequate movement precision and energy efficiency [59], as well as to adapt to environmental requirements [60]. From a neurological point of view, CCM is particularly important for rehabilitation, in robotics approaches to rehabilitation, it is even considered as a distinct element of motor control in many theoretical models of motor control [61, 62, 63, 64, 65].

TVC is a contraction of the antagonist’s muscle, triggered by the command on the agonist, which creates an opposing or even reversing torque of the desired movement [66]. It characterizes the simultaneous activation of agonist and antagonist muscles within the same joint, and which act on the same plane [67].

TVC can be characterized physiologically and biomechanically. In this work, we will focus more particularly on bio-mechanical aspects. In the case of normal movement, the presence and degree of muscular co-contraction (CCM) are still subject to debate [68, 69]. The level of CCM depends on several parameters: it is proportional to the speed [70, 71], it is variable for the degree of inertia [72], to the muscle group under consideration [73] and maybe partially dependent on sex and age [74, 75, 76]. Other factors can also be taken into account to explain these variations [25, 53, 77, 78, 79] in the literature, but it is always calculated as a ratio between agonist and antagonist muscles and as a function of the walking sequence (double support, unipodal phase, etc.). From a scientific point of view, TVC has been defined in different ways to facilitate its interpretation. For walking, several components are thus put forward, such as magnitude, time or the temporal evolution of the amplitude. From these definitions, different formulas or computational approaches have been used to quantify the TVC [80]. All of these methods limit the comparison of data between studies and the understanding of TVC mechanisms.

The CCM assessment uses indices that have been created to contribute to a better understanding of the underlying mechanisms. These clues shed light on the role of TVC in walking in people with stroke or CP. During walking, TVC is presented as the time and magnitude of simultaneous contraction between opposing muscles [80]. The studies on the evaluation of TVC are purely descriptive, they do not involve any intervention or program of work with the patients. Only two experimental studies used control groups and assessed walking before and after an intervention [81, 82, 83].

TVC of agonist and antagonist muscles causing flexion/extension movement of a joint can be assessed in several ways using surface electromyography (EMG): (i) the first method is the visual estimate of the EMG amplitude or the percentage of overlap of the considered muscles; (ii) the second method aims to determine the relationship between the activity of the agonist and that of the antagonist; (iii) the third method quantifies the antagonistic moment by a mathematical model; (iv) the last method determines the ratio of the antagonist EMG to the EMG of the same muscle during its maximum agonist contraction. However, external factors can interfere with the measurements. For the first method, a standardization to allow comparison between subjects is not all time possible. For the second method, a decrease in agonist recruitment may introduce a bias to consider high co-contraction indices when they are only high by the decrease in agonist activity. This situation is all the more common as the quantitative values of these two EMGs are recorded under different conditions [66]. The hypothesis at the heart of the third method postulates a linear relationship between the EMG and the moment, the EMG is used as a surrogate for the force produced with all the inaccuracies that this can cause. The latter method has an advantage, which is that the antagonist is also measured in its role as an agonist. However, this method does not take into account the resulting joint moment, although it can be assessed separately.

The EMG thus plays a role in detecting the patient’s intention [84], an intention that is considered to be a necessary parameter for the control mechanism of a rehabilitation exoskeleton. The objective of the work is to find a relationship between muscle co-contraction and joint angle to control a gait exoskeleton. This objective led us to use the last method, this choice is justified by a study of the indices of the proposed CCMs. Among the different methods of calculating the CCM index, three are more robust and more precise. In addition, these methods are evaluated in cases of neurological diseases, such as those studied in this work. These CCM indices are shown below:

(i) the first modified method that was introduced by Unnithan et al. [77] and by Frost et al., [53]:

$${\text{ICCM}}_1\left(t\right)=\frac{\mathit{\text{surface commune}}}{\mathit{\text{surface totale}}}\times 100$$E1

where the common area is the integral of the sum of the antagonistic and agonistic EMG amplitudes and the total area is the total number of data points;

(ii) the second method has been proposed by [81]:

$${\text{ICCM}}_2\left(t\right)=\frac{2\times \mathit{\text{surface commune}}}{\mathit{\text{surface}}{\mathit{EMG}}_{\mathit{\text{antago}}}+\mathit{\text{surface}}{\mathit{EMG}}_{\mathit{ago}}}\times 100$$E2

where the EMG surface is the integral of the normalized muscle EMG values.

(iii) the third method is the one defined by Falconer and Winter [85]:

$${\text{ICCM}}_3\left(t\right)=\frac{2\times ‖\mathit{EMG}{\left(t\right)}_{\mathit{\text{antago}}}‖}{‖\mathit{EMG}{\left(t\right)}_{\mathit{\text{antago}}}‖+‖\mathit{EMG}{\left(t\right)}_{\mathit{ago}}‖}\times 100$$E3

Note that these CCM indices are estimated in the state of the art as values like values discrete for each sub-phase of a walking cycle.

4. Experimental research

The experimental research has the goal to define one new form of the CCM indicators (4, 5, 6) and neuro motor indices (INM). The precise determination of neurological impairment in terms of CCM during gait requires robust measurement techniques that take into account the environmental conditions under which gait is assessed [86]. For example, walking on a floor surface rather than on a treadmill, walking at different speeds and for longer distances or times increase TVC recruitment and variability between subjects [87, 88]. TVC during walking has been studied in patients with stroke [54, 81, 89, 90, 91, 92, 93], and in the case of CP patients [77, 81, 94, 95, 96, 97, 98, 99].

Bio kinematic analysis relies on two-measure detection: (i) first, the angles of the knee and hip in the sagittal plane when flexing and extending these joints when walking. From these angles, a percentage interpolation is estimated over a full gait cycle; (ii) next, EMGs, designed to assess TVC during functional movement. Their study requires an analysis of the relative variations in contraction over time between the agonist and the antagonist [100].

There are standards that have been developed for the different stages of processing of these signals, such as SENIAM for analog and digital acquisition and analysis [101], their implementation however remains variable. Concerning the most appropriate techniques for analyzing EMG signals, differences remain as for the choice of the normalization technique, which leads to important differences between the studies [102].

I have set up the following chain to process the EMG data. EMG measurements are filtered with a 4th order bandpass (10-400 Hz) Butterworth filter, centered to eliminate their average, then rectified by calculating their absolute value, then filtered by a low-pass filter with a frequency f_c between 4 and 6 Hz, the value of which depends on the cadence of the subject [103].

The first processing step is to interpolate the joint angles, taking care to generate the same number of points as those of the EMG matrices. This makes it easier to align treatments later. It is then important to estimate the mean and the variance over several walking cycles to obtain an average curve that takes into account the variation margins for hip and knee angels, Figure 7.

Figure 7.

Mean and variance of angular curves of the knee and hip of a healthy subject.

The second step is to extract the TVC indices, to obtain continuous values. The goal is to have the data necessary to establish a relationship between joint angles and TVC indices during the gait cycle. It is thus possible to use a correlation method between each TVC index and the angular variation curve for each joint. This analysis is done for each part of the walking cycle in healthy subjects Figure 8.

Figure 8.

Example of processing of RF - quadriceps EMG signals and BF- hamstring EMG signals for a healthy subject during spontaneous walking at normal speed.

The femoral quadriceps is the largest muscle group in the human body. It is he who mainly supports the weight of the body and allows movement. It performs a role of knee extensor and hip flexor via two muscles included in our study: the rectus femoris (RF) and the vastus lateralis (VL), described below. The hamstring is a muscle group that allows hip extension and knee flexion by two muscles, the biceps femoris (BF) and the semi membranous (SM). From the perspective of a system using the minimum number of electrodes, we chose to measure the LF and RF. This is because the BF is a very large muscle, which extends from the lower pelvis above the knee. It is particularly involved in the flexion of the leg at the knee and in the extension of the thigh at the hip, playing a regulating role in the flexion of the knee during walking. The DF is also very extensive and also plays a role in stabilizing the knee during walking, as shown in Figure 9. The VL allows the extension of the leg at the knee and secondarily the external rotation of the leg but does not play a role in the flexion of the hip. SM is a thigh extensor muscle that flexes the leg on the thigh, but when the knee is flexed the SM acts as an internal rotator of the leg. The VL and the MS can therefore be left out for this study.

Figure 9.

Biceps Femoris (BF) and rectus Femoris (RF).

The CCM indices estimated in this work follow the methods described in the state of the art. In the reference articles, they are assessed in a discretized fashion for each part of the walking cycle. This type of approach does not allow continuous estimation, which is necessary to complete this work. We proposed a reformulation of Eqs. 1–3 which allows these TVC indices to be determined continuously, using a sliding window with overlap. The proposed reformulation of the TVC indices is shown below and is used for the estimates shown in Figure 10.

Figure 10.

Continuous assessment of co-contraction indices.

The method of Unnithan [76] and of Frost et al [53] (1) is thus rewritten:

$${\text{ICCM}}_1\left(t\right)=\frac{\left({\int }_{t_1}^{t_2}{\mathrm{ENV}}_{\mathrm{ago}}^{\mathrm{emg}}\left(t\right)\cap {\mathrm{ENV}}_{\text{anta}}^{\mathrm{emg}}\left(t\right)\mathit{dt}\right)}{{\int }_{t_1}^{t_2}{\mathrm{ENV}}_{\mathrm{ago}}^{\mathrm{emg}}\left(t\right)\cup {\mathrm{ENV}}_{\text{anta}}^{\mathrm{emg}}\left(t\right)\mathit{dt}}\times 100$$E4

The method of Hesse et al [81] (2) becomes:

$${\text{ICCM}}_2\left(t\right)=\frac{2\left({\int }_{t_1}^{t_2}{\mathrm{ENV}}_{\mathrm{ago}}^{\mathrm{emg}}\left(t\right)\cap {\mathrm{ENV}}_{\text{anta}}^{\mathrm{emg}}\left(t\right)\mathit{dt}\right)}{{\int }_{t_1}^{t_2}{\mathrm{ENV}}_{\mathrm{ago}}^{\mathrm{emg}}\left(t\right)+{\mathrm{ENV}}_{\text{anta}}^{\mathrm{emg}}\left(t\right)\mathit{dt}}\times 100$$E5

Finally, the method of Falconer and Winter [85] (3) can be reformulated as follows:

$${\text{ICCM}}_3\left(t\right)=\frac{2\left({\int }_{t_1}^{t_2}{\mathrm{ENV}}_{\mathrm{ago}}^{\mathrm{emg}}\left(t\right)\mathit{dt}+{\int }_{t_2}^{t_3}{\mathrm{ENV}}_{\text{anta}}^{\mathrm{emg}}\left(t\right)\mathit{dt}\right)}{{\int }_{t_1}^{t_3}{\mathrm{ENV}}_{\mathrm{ago}}^{\mathrm{emg}}\left(t\right)+{\mathrm{ENV}}_{\text{anta}}^{\mathrm{emg}}\left(t\right)\mathit{dt}}\times 100$$E6

For the study of the knee, ENV^emg_anta is the envelope of the muscles that train the slack knee flexion (quadriceps) and ENV^emg_ago is the envelope of the muscles that are train the knee extension movement (hamstrings). For the study of the soft-hip, ENV^emg_anta is the envelope of the muscles that causes movement flexion of the hip (hamstring) and ENV^emg_ago is the envelope of the muscles that evokes the hip extension movement (quadriceps). For the ICCM₁(t) and ICCM₂(t) indices, the period of t₁ and t₂ represents a complete cycle of walking while for ICCM₃(t), the period of t₁ to t₂ denotes the period when the agonist EMG is lower than the antagonist EMG, whereas from t₂ to t₃, they denote the period when the antagonist EMG is lower than the agonist EMG.

Compared to Figure 10, ICCM₃ shows a discontinuity at the start of each step cycle sub-phase. The other two indices ICCM₁ and ICCM₂ are continuous variables relative to the walking cycle. So, in the rest of this work, we will leave ICCM₃ aside. We examined the relation between ICCM selected and joint angles, the result was, no relation, especially in extension (Figures 11 and 12).

This research paper was introduced one new movement indicator- Neuro Motor Indices (NMI). The NMI is by construction defined as an index of continuous muscle co-contraction and directly dependent on the trajectory of the joint angles. Our proposal is to combine the two chosen CCM indices, ICCM₁ and ICCM₂, to gain the advantages of each, using a relation that depends on the flexion/extension of each joint during a movement. This approach is inspired by the flexion/extension of each joint during the gait cycle. The NMI is thus built as a relation between envelopes of an agonist/antagonist muscle pair according to this flexion/extension. In practice, the combination of TVC indices is performed with a nonlinear regression derived from a polynomic Hermitian. Thus, INM is a non-linear combination of ICCM₁ and ICCM₂. TVC indices are defined as indices of joint stability and are calculated from EMG signals that detect the subject’s intention. The INM, therefore, takes advantage of these properties and defines a new relationship between EMGs and joint angles, which depends on the capacity of the muscles and therefore on the expertise of the patient.

The formal definition of INM is as follows:

$$\begin{array}{c}\mathrm{A}\left(t\right)=\frac{2\left({\int }_{t_1}^{t_2}{\mathrm{ENV}}_{\mathrm{ago}}^{\mathrm{emg}}\left(t\right)\cap {\mathrm{ENV}}_{\text{anta}}^{\mathrm{emg}}\left(t\right)\mathit{dt}\right)}{{\int }_{t_1}^{t_2}{\mathrm{ENV}}_{\mathrm{ago}}^{\mathrm{emg}}\left(t\right)+{\mathrm{ENV}}_{\text{anta}}^{\mathrm{emg}}\left(t\right)\mathit{dt}}\times 100\\ \mathrm{B}\left(t\right)=\frac{2\left({\int }_{t_1}^{t_2}{\mathrm{ENV}}_{\mathrm{ago}}^{\mathrm{emg}}\left(t\right)\cap {\mathrm{ENV}}_{\text{anta}}^{\mathrm{emg}}\left(t\right)\mathit{dt}\right)}{{\int }_{t_1}^{t_2}{\mathrm{ENV}}_{\mathrm{ago}}^{\mathrm{emg}}\left(t\right)\cup {\mathrm{ENV}}_{\text{anta}}^{\mathrm{emg}}\left(t\right)\mathit{dt}}\times 100\end{array}$$E7

$$\mathit{INM}=\frac{1}{2}\left\{\mathrm{A}\left(t\right)+R_x\left(t\right)\mathrm{B}\left(t\right)\right\}$$

$$R_x\left(t\right)=h_1\left(t\right)f_0+h_2\left(t\right)p_0+h_3\left(t\right)f_1+h_4\left(t\right){\mathit{fp}}_1,$$

$$f\left(t\right)={\mathrm{ENV}}_{\mathrm{ant}}^{\mathrm{emg}}\left(t\right)\cap {\mathrm{ENV}}_{\mathrm{ago}}^{\mathrm{emg}}\left(t\right)$$

for which h₁, h₂, h₃, h4 ∈ P are the roots of a Hermitian polynomial, and p₀, p₁ are the tangents to f₀ and f₁, and (t₂ - t₁) represents a complete operating cycle. To test the reliability of the INM, canonical correlation analysis is applied in an offline study. Then, to move on to the control stage, a second correlation analysis is applied in an online study.

Estimation	DS1	SS	DS2	SP
ϵICCM1 = f(θhip(t))	2.3	6.7	5.3	7.2
ϵICCM2 = f(θhip(t))	4.4	7.1	5.1	6.9
ϵICCM3 = f(θhip(t))	7.5	8.1	6.1	7.9
ϵINM = f(θhip(t))	1	0.9	0.7	0.5

Table 2.

Comparison of the error estimate for the indices studied for the case of the right hip of a healthy subject.

Figure 11.

Decomposition according to hip flexion and extension θ_hip as a function of ICCM₁.

Figure 12.

Decomposition according to hip flexion and extension θ_hip as a function of ICCM₂.

The results obtained from the analyzes of the estimate of error for the 20 subjects show that the healthy subjects (Example right hip of a healthy subject Table 2), that is to say, those who have a normal gait cycle, show a small change for all ICCMs with a slight advantage for the IMN. In the case of patients, the IMN provides valuable information and allows the evaluation of walks that deviate from prototypical cycles, as is most often the case in stroke or CP patients (Example right hip of a stroke subject Table 3). Analysis for neurological subjects shows that ICCMs are not appropriate. This calls into question their reliability in analyzing these cases, even though they are used in the state of the art. The cycles in stroke or CP subjects are poorly determined and they show very significant changes compared to normal walking cycles. However, they make it possible to achieve movement that allows movement and most often minimizes pain for patients. Thus, ICCMs worked well in subjects with normal walking, but the main source of error was in subjects with stroke or cerebral paralysis CP.

Estimation	DA1	SA	DA2	PO
$ϵ$ICCM1 = f(θhanche(t))	3.1	42.9	39.5	14.2
$ϵ$ICCM2 = f(θhanche(t))	3	54.9	49.8	20
$ϵ$ICCM3 = f(θhanche(t))	44.1	28.8	66.9	49.8
$ϵ$INM = f(θhanche(t))	2	1.9	1.8	0.9

Table 3.

Error estimates for the indices studied in the case of the right hip of a stroke subject.

5. Conclusion

The importance of the use of mechatronic systems, especially exoskeletons, reveals their role as a solution for the rehabilitation of walking in people with neurological problems or who are forced to stay in a wheelchair. This work is primarily concerned with the possibility of implementing effective rehabilitation using these systems. This requires close human-machine collaboration between the patient and the exoskeleton. The central place of the patient’s expertise in his movement and his ability to interact with the machine in the proposed work constitutes our main contribution to this research work. In this work, we, therefore, integrated the expertise of the patient into a controller that can be used on a rehabilitation exoskeleton. This model takes into account the effects of muscle weakness and spasticity. It is possible to use these results for a simple control strategy that is easy to evolve with the patient’s condition. This type of strategy adapts to all types of gaits and all speeds of movement. The gait of an evolving exoskeleton of the lower limbs is treated in a work by the team of Samer AlFayed [104]. This exoskeleton is designed with two active degrees of freedom on the hip and knee. We presented a state of the art on biomechanics of gait, the existing methods for the control of the exoskeletons of the lower limbs used in the world, the neurological problems causing gait disturbances such as stroke and the CP.

The basic premise of this work is that if a movement involves EMG activity, we can then estimate the movement from the EMGs to respect the muscle capacity and will of the patient. The research question addressed here, therefore, is to find a relationship between bio- signals (EMG) and kinematic parameters (here articular angles) to drive an assistance exoskeleton.

The solution that we have proposed is based on the detection of muscle co-contraction between agonist/antagonist muscles. Co-contraction plays a crucial role in detecting patient intentions and in characterizing joint angles when walking. This co-contraction as well as the patient’s intention are estimated from EMG measurements. We have defined the muscle function that allows us to estimate muscle co-contraction. We presented its evaluation in the state of the art from the indices of muscle co-contractions, and determined these indices in a continuous fashion between a pair of biarticular agonist/antagonist muscles for the hip and the knee. We then tested the correlation of these indices with joint angles and showed that the correlations are weak. This shortcoming is all the more limiting as the same co-contraction index is used in practice to assess two different joint angles. We, therefore, sought to introduce the specificity of each joint to these indices, by proposing to use a nonlinear regression related to the flexion/extension of each joint. We were thus able to propose a specific neuromotor index for each joint. We determined the error estimate for the INM and other state-of-the-art indices in the gait cycle sub-phases for healthy subjects, stroke, and CP. The INM showed better results than all other indices. To more fully validate this clue, we used a two-step canonical correlation analysis throughout the walk cycle. This evaluation showed the advantage of the INM over state-of-the-art indices. The INM shows a quasi-linear correlation with joint angles and the internal correlation was validated. In the future work will be comparatively studied all these indicators in different stages of the patient health and for all joints.