Closed-Loop FES Control of a Hybrid Exoskeleton during Sit-to-Stand Exercises: Concept and First Evaluation

Abstract

Rehabilitation of paralysis caused by a stroke or a spinal cord injury remains a complex and time-consuming task. This work proposes a hybrid exoskeleton approach combining a traditional exoskeleton and functional electrical stimulation (FES) as a promising method in rehabilitation. However, hybrid exoskeletons with a closed-loop FES control strategy are functionally challenging to achieve and have not been reported often. Therefore, this study aimed to investigate a powered lower-limb exoskeleton with a closed-loop FES control for Sit-to-Stand (STS) movements. A body motion capture system was applied to record precise hip and knee trajectories of references for establishing the human model. A closed-loop control strategy with allocation factors is proposed featuring a two-layer cascaded proportional–integral–derivative (PID) controller for both FES and exoskeleton control. Experiments were performed on two participants to examine the feasibility of the hybrid exoskeleton and the closed-loop FES control. Both open- and closed-loop FES control showed the desired performance with a relatively low root-mean-squared error (max 1.3${}^{\circ }$ in open-loop and max 4.1${}^{\circ }$ in closed-loop) in hip and knee trajectories. Notably, the closed-loop FES control strategy can achieve the same performance with nearly 60% of the electrical power input compared to the open-loop control, which reduced muscle fatigue and improved robustness during the training. This study provides novel insights into body motion capture application and proposes a closed-loop FES control for hybrid exoskeletons.

1. Introduction

Stroke and spinal cord injury (SCI) are the leading causes of paralysis, impacting millions of people worldwide each year. These conditions not only affect the individuals directly, but also pose additional challenges for caregivers in terms of medical care, daily activities, and psychosocial well-being [1,2]. In the European Union, a significant proportion of adults, approximately 1.6%, have experienced a stroke or have been dealing with its chronic consequences within the past year [3]. Moreover, the COVID-19 pandemic has had a particularly noticeable impact on acute ischemic stroke (AIS) care within hospitals, leading to an increase in AIS case fatality rates [4,5]. One of the most-challenging daily activities for individuals with mobility impairments is standing up from a chair. The Sit-to-Stand (STS) movement is considered more demanding than other physical activities such as walking or climbing stairs [6,7]. The inability to perform this movement significantly affects an individual’s quality of life and autonomy.

Functional electrical stimulation (FES), also known as neuromuscular electrical stimulation (NMES), has long been recognized as a vital component of physical therapy [8,9]. Its applications date back to 1961, when FES was initially used to stimulate muscular nerves via electrodes to address foot drop issues [10]. Since then, FES has emerged as a promising approach in SCI and body movement rehabilitation [11]. In the past two decades, various FES applications, such as grasping, walking, cycling, and STS movements, have been widely utilized for paralyzed or paretic patients [12,13,14].

An exoskeleton is a wearable robotic device, offering mechanical support and assistance to the user. Exoskeletons have traditionally been designed to assist individuals with paraplegia, SCI, or muscular dystrophy in achieving basic mobility in their daily lives. Recently, innovative FES exoskeletons have emerged with advantages in providing additional torque support, reducing the exoskeleton weight, and offering extra degrees of freedom in the ankle joint [15,16,17,18,19,20,21]. The concept of hybrid FES exoskeletons originates from the demand to improve the effectiveness and efficiency of rehabilitation interventions. FES systems are primarily used to stimulate muscles and generate contractions, but they lacked the mechanical support necessary for functional movements. Besides, prolonged use of FES for exercise often leads to muscle fatigue, which complicates the rehabilitation process [22,23,24]. In a hybrid system, the exoskeleton primarily provides support and power, while FES provides complementary torque. By combining FES and exoskeleton technologies into a hybrid device, the integration overcomes the limitations of each approach and offers unique advantages in creating a synergistic effect and enhancing the overall rehabilitation outcomes.

Recent research in hybrid exoskeletons focused on improving the human model and enhancing hybrid exoskeleton mobility in individuals with paralysis or impairments [23,24]. These hybrid orthoses combined FES-induced muscle activation with the mechanical support and assistance, aiming to restore or regain the patients’ movement capabilities [25,26,27,28,29]. Furthermore, some research groups have focused on the optimization of muscle activation patterns and coordination. These studies addressed the challenge of providing effective mechanical assistance while ensuring synchronized and coordinated muscle activation through FES-induced stimulation [30], resulting in more-natural movement during rehabilitation. Various control strategies, such as cascade PID control, model predictive control (MPC), iterative learning control (ILC), and robust control, were also investigated to meet specific user requirements and adapt to environmental conditions [19,31,32,33]. Additionally, multiple sensors, including inertial measurement units (IMUs), electromyography (EMG), ultrasound, and others, have been widely applied for continuous monitoring and utilized as feedback signals [16,34]. By analyzing different bio-signals related to movement, muscle activation, and external forces, the control parameters of FES and exoskeletons could be dynamically adapted to enhance performance. These approaches allow hybrid orthoses to better accommodate changing conditions and meet individual user requirements. In addition to technical approaches, there are also researchers in this field conducting comprehensive experiments in clinical evaluation and user experience to assess the effectiveness, comfortableness, and safety of hybrid FES exoskeletons [35,36]. These evaluations involve individuals with diverse neurological conditions, including spinal cord injury, stroke, and multiple sclerosis. In particular, the impact of hybrid orthoses on muscle fatigue was considered to assess their effectiveness in rehabilitation.

One of the challenges of hybrid exoskeletons relates to energy consumption and muscle fatigue in hybrid FES exoskeletons [32,37,38]. We note that the mentioned research of the hybrid exoskeletons seldom tackled this problem. Either open-loop FES actuators cooperated with the exoskeleton or an event-based control strategy triggered the FES by measuring the motor torque, which both induced more current injection for the neurostimulation by FES and consumed more battery energy in the exoskeleton. Therefore, a closed-loop FES control strategy in hybrid exoskeleton could provide an effective approach to this situation. In terms of mechatronic objectives, closed-loop systems optimize power consumption and energy management by regulating FES and exoskeleton actuation based on the patient’s needs. It can intelligently distribute power resources to minimize energy consumption while ensuring adequate support and assistance during movement. Consequently, it could minimize unnecessary muscle activation, preventing fast muscle fatigue, prolonging the duration of wearable use, and enhancing the comfort during the rehabilitation activities. In terms of clinical objectives, closed-loop systems enable personalized and adaptive rehabilitation protocols. This adaptive control facilitated with more-effective and -targeted rehabilitation promotes neuroplasticity, which is the brain’s ability to reorganize and form new neural connections. By providing synchronized sensory feedback through the exoskeleton and electrical stimulation through FES, the closed-loop system can facilitate sensorimotor integration and promote neural reorganization.

This work involves proposing a closed-loop FES control strategy for a hybrid exoskeleton to aid in rehabilitation, which aims to investigate a powered lower-limb exoskeleton for STS movements. To achieve this, we used a body motion capture system to record precise hip and knee trajectories of references for establishing the human model, allowing us to design and test a closed-loop control strategy with allocation factors featuring a two-layer cascaded PI controller for both FES and exoskeleton control. The experimental setup involved performing experiments on two participants to examine the feasibility of the hybrid exoskeleton and closed-loop FES control. Then, we compared both open- and closed-loop FES control strategies and measured the RMS of the hip and knee trajectories to evaluate the performance. The Discussion Section elaborates on these findings and the effect of different allocation factors, highlighting the potential benefits of using hybrid exoskeletons with closed-loop FES control in rehabilitation. Overall, each section contributes to addressing the research problem by providing evidence for how closed-loop FES control can improve hybrid exoskeletons’ performance in terms of energy consumption savings, muscle fatigue prevention, and a robust control strategy.

2. Model Dynamics and Control Strategy

2.1. System Dynamics

In our work, the system to be controlled is the combination of the human body together with the FES and exoskeleton. We considered a lower-limb exoskeleton, the motors of which were attached at the hip and knee joints. Two pairs of electrodes of the FES were placed separately on the surface of the rectus femoris and vastus medialis leg muscles. The position of the electrodes was affixed above the nerves of the muscles to achieve the highest torque and cause minimum fatigue. Figure 1 shows a sketch diagram of the human body model with the exoskeleton and FES. Additionally, the exoskeleton was securely fastened to the body, and the electrodes are attached tightly to the skin, preventing any relative motion between the exoskeleton and the wearer.

The human body and the hybrid exoskeleton can be modeled as a three-link pendulum robot with three degrees of freedom at the hip, knee, and ankle joints (see Figure 1a). It is noted that we assumed the upper body to be rigid, the lower body and exoskeleton to be symmetrical, and synchronized movement during the STS motion, while the feet remained fixed on the ground without slipping. The model depicted in Figure 1b allows for the derivation of kinematic and potential equations (see Appendix A), which subsequently leads to the conventional Euler–Lagrange equations governing the dynamics of the hybrid exoskeleton during lower-limb body motion. These equations are expressed as:

$$\mathit{M}\left(\mathit{\theta }\right)\ddot{\mathit{\theta }}+\mathit{C}(\mathit{\theta },\dot{\mathit{\theta }})\dot{\mathit{\theta }}+\mathit{g}\left(\mathit{\theta }\right)+\mathit{\tau }(\mathit{\theta },\dot{\mathit{\theta }})=\mathit{T},$$

(1)

where $\mathit{M}\left(\mathit{\theta }\right)\in {\mathbb{R}}^{3\times 3}$ represents the mass inertia matrix and $\mathit{\theta }={\left[{\theta }_A{\theta }_K{\theta }_H\right]}^T$ denotes the joint angle vector. The vectors $\mathit{\theta }$, $\dot{\mathit{\theta }}$, and $\ddot{\mathit{\theta }}$ correspond to the angles, velocities, and accelerations of all joints, respectively. The matrix $\mathit{C}(\mathit{\theta },\dot{\mathit{\theta }})\in {\mathbb{R}}^{3\times 3}$ accounts for the centrifugal Coriolis forces, while $\mathit{g}\left(\mathit{\theta }\right)$ incorporates the gravitational force factor. Additionally, $\mathit{\tau }(\mathit{\theta },\dot{\mathit{\theta }})$ represents the equivalent joint elastic moments.

The inputs to the hybrid exoskeleton are derived from the actuators, which include the stimulated muscles and the two pairs of exoskeleton motors. Specifically, the muscle contraction induced by FES can be considered as an additional torque applied to the knee joint, when the exoskeleton provides torque to both the knee and hip joints. Therefore, the input vector $\mathit{T}$, encompassing contributions from the motors, FES, and the human body, can be expressed as follows:

$$\mathit{T}={\left[0T_{\mathrm{K},\mathrm{Motor}}{T}_{\mathrm{H},\mathrm{Motor}}\right]}^T+{\left[0T_{\mathrm{FES}}0\right]}^T+{\left[T_{\mathrm{A},\mathrm{Human}}{T}_{\mathrm{K},\mathrm{Human}}{T}_{\mathrm{H},\mathrm{Human}}\right]}^T,$$

(2)

where $T_{\mathrm{K},\mathrm{Motor}}$, $T_{\mathrm{H},\mathrm{Motor}}$, and $T_{\mathrm{FES}}$ represent the torque from the knee motor, hip motor, and FES actuator, respectively. In our work, we made the assumption that there are no torques generated by the human body, and the ankle joint remains fixed and unaffected by both FES and the exoskeleton. Consequently, the torque vector contributed by the human body, represented by $T_{\mathrm{A},\mathrm{Human}}$, $T_{\mathrm{K},\mathrm{Human}}$, and $T_{\mathrm{H},\mathrm{Human}}$, was assumed to be zero.

2.2. Experiment Setup

2.2.1. Body Motion Capture and STS Trajectories

The desired and precise trajectories of lower-limb body motion were captured and analyzed in the motion lab of the Department of Geriatrics, University Hospital Aachen. To achieve this, a real-time motion-capture system consisting of ten cameras, Qualisys OPUS5+, and Qualisys Track Manager (QTM) software (Qualisys AB, Göteborg, Sweden) was employed. During the experiments, both participants were instructed to wear 56 markers distributed across their bodies, as depicted in Figure 2.

The marker set was constructed using CAST models [39,40]. In addition to the standard marker placements, two extra markers were attached to each knee, and three extra markers were attached to each hip, specifically positioned on the most-lateral prominence of the left and right greater trochanter. These additional markers, along with the CAST models, were simultaneously utilized to create a comprehensive whole-body movement model during the STS movement, enabling the acquisition of accurate trajectories. To capture as many data as possible, all of the cameras operated at a frequency of 120 Hz. The acquired data were then analyzed, and the trajectories were determined through a fitting process using an 8th-order Fourier series.

2.2.2. FES

The FES stimulator used in this study was the Rehamove 3 (Hasomed GmbH, Magdeburg, Germany), which was fit and mounted on the waist of the exoskeleton. In Figure 3a, two FES channels are utilized, and electrical pads are positioned on both legs, which are specifically attached above the nerves of the rectus femoris and vastus medialis muscles, to ensure optimal stimulation. Additionally, a test bench for torque–pulse width relationship experiments incorporated a rotating torque sensor DR-2477 (Lorenz Messtchnik GmbH, Alfdorf, Germany), a data acquisition system NI USB 6229 (National Instruments Corp, Austin, TX, USA), and a mechanical transmission system.

The participants were instructed to sit in an upright position with their knees flexed at approximately 90${}^{\circ }$, aligning the center of rotation of the knee joint with the sensor’s axis. When FES stimulation was initiated, the panel transmitted the torque to the sensor. Controlling the FES involved three parameters: frequency, pulse width, and current. It should be noted that the torque–pulse width relationships can vary from person to person. Therefore, prior to the experiment, an evaluation of FES-induced torque was conducted for each participant. This evaluation aimed to ensure the safety and optimal stimulation by identifying the appropriate parameters for each individual. Figure 3b presents the torque–pulse width characteristics of FES for the first participant, with a fixed frequency of 30 Hz. We conducted six groups of experiments with different currents ranging from 15 mA to 40 mA, while varying the pulse width from 50 $\mu$s to 900 $\mu$s in steps of 50 ms. Initially, the leg torque exhibited a linear relationship with the pulse width, but eventually reached saturation even with further increases in the pulse width. Consequently, for the subsequent experiments, we selected a constant current of 40 mA, a frequency of 30 Hz, and a pulse width control ranging from 50 $\mu$s to 300 $\mu$s. We fit the data with 95% confidence bounds using a linearized function between torque ($T_{\mathrm{FES}}$) and pulse width:

$$T_{\mathrm{FES}}\left(pw\right)=0.09343·pw-5.386.$$

(3)

2.2.3. Exoskeleton

Our lower-limb exoskeleton, ${\mathrm{L}}^2\mathrm{Exo}\_\mathrm{PE}$, incorporates parallel elastic actuators, which is the custom-built exoskeleton in our previous work [41,42]. It consists of two pairs of actuators: one pair attached to the hip and another pair attached to the knee. Each leg of the exoskeleton is equipped with the same module, which includes an EC flat frameless motor (Maxon Motor AG, Sachseln, Switzerland) and a 50/1 reduction gear transmission (Harmonic Drive AG, Limburg an der Lahn, Germany) (see Figure 4).

A total of four motors were utilized in the exoskeleton: two for the knee joints and two for the hip joints. All motors shared the same specifications, including a nominal torque of 49.4 Nm, a nominal speed of 3.7 rad s⁻¹, a no-load speed of 4.4 rad s⁻¹, a rated motor power of 260 W, and an axial length of 105 mm. These motors were connected to Escon 50/8 motor controllers from Maxon Motor AG, Sachseln, Switzerland. To ensure safe mechanical operation and prevent non-anatomical positions, a safety ring was employed as an end stopper to limit the range of motion for each joint. The hip joint had a restricted range of −15 ${}^{\circ }$ to 105 ${}^{\circ }$, and the knee joint was limited to 0 ${}^{\circ }$ to 120${}^{\circ }$.

2.2.4. Control Unit

Figure 4 provides an overview of the data structure of the hybrid exoskeleton prototype and its two separate systems enabling simultaneous cooperation. The first system involved the FES actuator, which interfaced with a data acquisition system (NI USB 6229). The control of the FES actuator was programmed in the C language and implemented as an S-function in Matlab/Simulink. Pulse width signals, generated in Simulink, were transmitted to the FES actuator via the UART interface.

The second system comprised the real-time MicroLabBox (dSPACE GmbH, Paderborn, Germany), which was responsible for controlling the exoskeleton. This device received data from the motor module and coordinated the operation of all subordinate control units. To ensure synchronization between the systems, the data acquisition instrument of the FES system was connected to the MicroLabBox to receive control signals. Furthermore, the NI data acquisition instrument played a role in synchronizing the operation of the two separate systems.

2.2.5. Experimental Procedure

Preliminary experiments were conducted with the participation of two young individuals who willingly supported the study (Subject A: height 1.78 m, weight 65 kg; Subject B: height 1.76 m, weight 74 kg). The experimental protocol underwent review by the Ethics Committee of RWTH Aachen University Hospital, and the corresponding statement can be referenced by the number EK23-029. To simplify the analysis, the weight of the body segments was measured and calculated [43], and the segment lengths of the exoskeleton were matched to those of the individuals. With the exception of the motor masses, we assumed a similar distribution of mass throughout each segment of the exoskeleton.

Table 1. Human and exoskeleton segment parameters for the mass, length, and distance of the center of mass (CoM).

Items	Human Body			Exoskeleton
	Shank	Thigh	Upper Body	Shank	Thigh
Sub.A M (kg)	7.8	13.0	44.2	8.2	9.0
Sub.B M (kg)	9.2	16.7	48.1	8.3	9.3
Sub.A L (m)	0.48	0.42	0.58	0.48	0.42
Sub.B L (m)	0.46	0.47	0.63	0.46	0.47
Sub.A $CoM$ (m)	0.24	0.23	0.18	0.24	0.21
Sub.B $CoM$ (m)	0.23	0.22	0.20	0.23	0.23

provides a summary of the physical properties of the subjects and the exoskeleton.

Figure 5 depicts the STS movement in three different conditions: Figure 5a shows STS with markers; Figure 5b illustrates STS in the QTM software; Figure 5c presents STS with the exoskeleton. Due to the markers attached to the human body, we were able to monitor the sitting position by the QTM software. This allowed us to guide the participants in adjusting the chair height and knee bending at a 90${}^{\circ }$ angle before the recording of the trajectories. Besides, we also used a timer to guide participants in performing the STS movement 20 times, with a fixed speed of 5 s for each repetition. To optimize the trajectories, the raw trajectory data were transferred from QTM to Matlab for selection. The individual-specific trajectories were then analyzed and applied as the trajectories for the control system. During the experimental phase of the hybrid exoskeleton, each participant wore the exoskeleton securely and ensured close attachment of the FES electrodes to the skin. The volunteers were instructed to relax, particularly in the lower limb muscle area, during the operation to minimize any effects on the motors and FES actuators.

2.3. Hybrid Control Strategy

The overall hybrid control model structure is depicted in Figure 6, where ${\theta }_{\mathrm{hip},\mathrm{ref}}$, ${\theta }_{\mathrm{knee},\mathrm{ref}}$, ${\theta }_{\mathrm{hip}}$, and ${\theta }_{\mathrm{knee}}$ represent the referenced and measured angle of hip and knee motor, ${\dot{\theta }}_{\mathrm{hip}}$ and ${\dot{\theta }}_{\mathrm{knee}}$ represent the angular velocity of each motor, $\mathit{K}$ represents the two-layer cascaded PI controller, and ${\mathit{I}}_{\mathrm{hip}}$ and ${\mathit{I}}_{\mathrm{knee}}$ represent the current input to the two actuators. The FES pulse width was limited to a range of 50 $\mu$s to 300 $\mu$s, ensuring that the muscles were adequately prepared to provide a linear torque response. Additionally, the motor current was limited to a range of −8 A to 8 A, corresponding to a linear torque response of −54.4 Nm to 54.4 Nm (with a gear ratio of 1:50).

Considering that the human exoskeleton is a two-input four-output system with small coupling interaction, we employed a decentralized diagonal control strategy for the hip and knee joints. The reference trajectories for control were obtained from the body motion system, and the angle and velocity information were obtained from the built-in sensors of the motors.

For hip control, we utilized a two-layer PI controller exclusively for the motors. In the case of knee control, we introduced a torque distribution factor A for knee torque allocation. This factor distributes the controller signal from a two-layer PI cascade controller to the FES actuator and knee motors separately. The hybrid control logic is depicted in the dashed box in Figure 6. Initially, the knee torque control signal $I_{\mathrm{knee}}$ was divided based on the distribution factor weight term A$\in [0,1]$ and input to the FES and exoskeleton. When the allocation factor $A=0$, the knee motors produced the maximum torque, with the FES compensating for any deficiencies. Conversely, when the allocation factor $A=1$, the FES became the primary output, and the knee motors were engaged when the FES could not provide sufficient torque. Before the torque input signal was distributed to the FES actuator, an angle sensor detected knee motion, ensuring that the FES was activated only during the knee extension phase. During knee flexion, the FES was deactivated, and the control signal was redirected to the exoskeleton. Thus, the FES was integrated into the closed-loop feedback control of the hybrid exoskeleton and cooperated with the knee motors during STS movement.

3. Experimental Results

Our pilot trials aimed to demonstrate the feasibility and advantages of closed-loop FES in hybrid control. We designed two sets of experiments to compare open-loop and closed-loop strategies for STS body motion. Each experiment lasted approximately 3 min, involving three repetitions of STS motion. After each STS motion, there was a 1 min rest period to prevent muscle fatigue induced by FES.

In the first trial, we employed an open-loop strategy in the hybrid exoskeleton. The FES was activated during knee extension, starting at 2.3 s and continuing until the completion of the STS movement. In this experiment, the FES was activated using a constant pulse width of 300 $\mu$s, a current of 40 mA, and a frequency of 30 Hz. Figure 7a,c illustrate the comparison between the reference and actual angles during the STS movement. Figure 7b,d show the original and filtered values of hip and knee motor torque, respectively.

Because of the inherent measurement accuracy errors of the motor sensors, both hip and knee motors initially produced relatively high-frequency torque at the beginning and end of the STS movement. Other factors, such as sensor reading noise, drift, or non-linearity, may also influence the accuracy of torque measurements and contribute to high-frequency torque variations. Due to the deployment of a PID controller for current control, the derivative component exhibited a rapid response, leading to an overcompensation of these accuracy errors. Nevertheless, once the movement started, it remained stable without any oscillation. However, the specific high-frequency noise can be eliminated by applying a band-pass filter, which was also utilized in the experiments shown in Figure 8b,d, as well as Figure 9b,d.

The closed-loop strategy, which utilized the allocation factor A to determine the distribution of control signals between FES and knee motors during the STS movement, was further analyzed. Six experimental groups were conducted with A ranging from 0 to 1. By selecting different values of A, the trajectories, motor current, and FES pulse width varied among the groups. Figure 8 and Figure 9 illustrate two closed-loop strategies with $A=0.4$ and $A=0.8$, respectively. In comparison to the open-loop strategy, the hip and knee trajectories in the closed-loop strategy exhibited slight deviations (Figure 8a,c and Figure 9a,c). We observed that the maximum angular deviations of the knee were approximately 5${}^{\circ }$ for an allocation factor of $A=0.4$ and 8${}^{\circ }$ for an allocation factor of $A=0.8$. The largest deviation occurred around 2.3 s after knee extension, when the maximum torque was required for the STS movement. The motor torque with two different allocation factors in the FES closed-loop control method is depicted in Figure 8b,d and Figure 9b,d. The performance of the knee motor depended on the allocation factor, reaching its maximum torque at 3 s. The performance of the hip motors was comparable to that of the open-loop strategy, and it can be observed that they did not reach their limitation. The FES pulse width control signals for both legs are presented in Figure 8e and Figure 9e. A larger allocation factor, such as $A=0.8$, caused the FES actuator to quickly reach its maximum pulse width compared to a smaller allocation factor, such as $A=0.4$. The combined maximum torque of the two actuators was sufficient for the STS movement, allowing for successful STS execution at any allocation factor throughout the entire motion period.

Table 2. Root-mean-squared error (RMSE) (${}^{\circ }$) of the hip and knee trajectories in open-loop strategy and closed-loop strategy with different allocation factors.

Conditions	Trial	RMSE
		Left Hip	Left Knee	Right Hip	Right Knee
Open-loop	1	1.213	0.739	1.359	0.759
Open-loop	2	1.204	1.110	1.292	0.745
Open-loop	3	1.200	0.736	1.320	0.709
Open-loop	Ave.	1.206	0.862	1.324	0.738
A = 0	1	1.253	2.844	3.178	1.453
A = 0	2	1.266	4.537	5.115	1.054
A = 0	Ave.	1.260	3.691	4.147	1.254
A = 0.2	1	1.378	2.366	1.616	0.860
A = 0.2	2	1.269	5.573	1.580	1.341
A = 0.2	Ave.	1.324	3.970	1.598	1.341
A = 0.4	1	1.242	5.173	1.402	0.912
A = 0.4	2	1.248	2.480	1.418	0.796
A = 0.4	Ave.	1.245	3.827	1.410	0.854
A = 0.6	1	1.256	2.665	1.466	0.825
A = 0.6	2	1.245	1.101	1.454	0.832
A = 0.6	Ave.	1.251	1.883	1.460	0.829
A = 0.8	1	1.272	3.192	1.821	0.881
A = 0.8	2	1.254	1.987	2.190	0.849
A = 0.8	Ave.	1.263	2.590	2.006	0.865
A = 1.0	1	1.266	3.998	2.181	0.825
A = 1.0	2	1.260	4.530	2.324	0.867
A = 1.0	Ave.	1.255	3.685	2.253	0.846

presents the root-mean-squared error (RMSE) of the hip and knee angles for both legs using both open-loop and closed-loop control techniques. In the case of the open-loop control strategy, both legs were well regulated with an average RMSE of 1.4${}^{\circ }$. With regard to the closed-loop control strategy, the hybrid exoskeleton with different allocation factors also demonstrated good performance, with an average RMSE of less than 4${}^{\circ }$ overall. For $A=0$, the FES actuator only activated the muscle when the exoskeleton reached its maximum torque, functioning as an assisting actuator. With $A=0.6$, the RMSE was relatively low, indicating that the FES actuator could achieve similar results as the full power mode.

At the end of each experiment, the motor torque of the exoskeleton and the pulse width of the FES actuator were integrated over time and compared in Figure 10. The integration of the torque and pulse width with time can be viewed as the energy consumption of the hybrid exoskeleton. The calculated torque–time integration (TTI) of the exoskeleton without FES and the hybrid exoskeleton is shown in Figure 10a. Obviously, the exoskeleton alone, without FES assistance, consumed the most power. The hybrid combination of both open-loop and closed-loop strategies allowed for energy savings during the STS movement. The open-loop strategy saved nearly half of the energy on both motors, while the optimal closed-loop strategy ($A=0.4$) saved 40% of power on the knee motor.

Figure 10b depicts the pulse width–time integration (PTI) of the FES in the hybrid exoskeleton, which directly indicates the amount of energy applied to the muscle. The open-loop strategy, without feedback, delivered full electricity throughout the entire movement. On the other hand, the closed-loop FES strategy, proven to be advantageous for stimulation, delivered less current to both legs and dynamically stimulated the muscle for a shorter duration. When the allocation factor A was set to zero, the knee motors output the maximum torque, with the FES compensating for it. In this scenario, the FES only provided current when necessary, resulting in more than a 50% savings in current injection to the leg. Even with an allocation factor of $A=0.6$, where FES accounted for 60% of the output power, it still utilized 75% of the total electricity compared to the open-loop strategy.

4. Discussion

The findings of this study demonstrated that hybrid closed- and open-loop FES exoskeletons are more effective in performing the STS motion compared to the exoskeleton alone. Both control strategies exhibited excellent performance in terms of motor measurement trajectories.

An advantage of the open-loop strategy is its model independence and noise-free FES reference signal. However, this also implies that the open-loop FES exoskeleton has less tolerance for external disturbances and may be less reliable. Random FES pulse widths can result in unsuccessful attempts. On the other hand, hybrid control requires an accurate model, but can withstand certain perturbations to achieve the desired goals. In this study, determining the optimal allocation factor for proportionally allocating control is a compromise. It depends on which actuator serves as the main output and which acts as assistance. Different control strategy parameters can affect the level of actuation. The closed-loop FES control with different allocation factors is more robust, enabling smoother actuator load distribution and disturbance compensation. This robustness is achieved through the saturation surplus allocation, which is the second component of the strategy.

Additionally, closed-loop systems offer several benefits in both mechatronic and clinical objectives. From a mechatronic standpoint, this system optimizes power consumption and energy management by regulating the actuation of both the FES and exoskeleton according to the patient’s needs. This intelligent power distribution minimizes energy consumption while ensuring sufficient support and assistance during movement. As a result, it reduces unnecessary muscle activation, prevents rapid muscle fatigue, extends the wearable’s duration of use, and enhances comfort during rehabilitation activities. In terms of clinical objectives, closed-loop systems enable personalized and adaptive rehabilitation protocols. This adaptive control approach facilitates more-effective and -targeted rehabilitation, promoting neuroplasticity, which is the brain’s ability to reorganize and form new neural connections. By providing synchronized sensory feedback through the exoskeleton and electrical stimulation through FES, closed-loop systems facilitate sensorimotor integration and promote neural reorganization. This integration of sensory feedback and stimulation enhances the effectiveness of rehabilitation and facilitates the development of new neural pathways.

Nevertheless, this study has limitations, and further improvements are necessary in the future. Firstly, the controller for the hybrid exoskeleton is a two-layer cascade PID, which is unable to control the ankle joint. In the future, the controller will be improved, possibly using techniques such as ILC or Hinf robust control. During the experiments, the feet and ankle joints were fixed on the ground to maintain balance during the STS transition. This limitation will be addressed in future research by adding more FES channels to the soleus, tibialis anterior, and gastrocnemius muscles, which control the ankle joint for balance. Additionally, the experiments conducted in this study involved only two healthy individuals. In a subsequent clinical study following the standards of good clinical practice (GCP), the hybrid exoskeleton should be tested on individuals with spinal cord injury (SCI) or elderly individuals with impaired body motion to validate its feasibility. It is important to note that muscle fatigue is a crucial factor in all body motions. Our analysis of FES energy is limited to current time integration. The human model will also be supplemented with a muscle fatigue model, and muscle fatigue indicators will be quantified using electromyography or electrical impedance myography sensors.

5. Conclusions

This study introduced a closed-loop FES control strategy for STS movement using a hybrid exoskeleton. Precise trajectories obtained from a body motion capture system were utilized as references and in the development of human models. An allocation factor was proposed to distribute torque between the FES and exoskeleton. Through our experiments, we examined the viability of closed-loop FES control in exoskeleton applications. The performance of the closed-loop FES strategy was found to be comparable to that of the open-loop FES control strategy. This outcome suggests that a hybrid exoskeleton has the potential to reduce muscle fatigue and power consumption, and improve robustness during STS movements. To validate the feasibility of clinical trials, it is important to consider a muscle fatigue model and to conduct further experiments involving patients with SCI or stroke. Future research can also explore the feasibility of alternative controllers and multiple channels for achieving robust STS movement.